This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Keep PL_numeric_radix_sv always set
[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
3ca88433
KW
391 const char * radix = (use_locale)
392 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
86799d2d 393 /* FALSE => already in dest locale */
3ca88433 394 : ".";
2213a3be 395
86799d2d 396 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 397
7a393424
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),
401 SvCUR(PL_numeric_radix_sv))
c1284011 402 && _is_cur_LC_category_utf8(LC_NUMERIC))
28acfe03 403 {
86799d2d 404 SvUTF8_on(PL_numeric_radix_sv);
28acfe03 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
501 if (DEBUG_L_TEST || debug_initialization) {
502 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
503 }
504
4c28b29c
KW
505 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
506 * have to worry about the radix being a non-dot. (Core operations that
507 * need the underlying locale change to it temporarily). */
508 set_numeric_standard();
509
98994639 510#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 511
98994639
HS
512}
513
514void
515Perl_set_numeric_standard(pTHX)
516{
7d4bcc4a 517
98994639 518#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 519
28c1bf33
KW
520 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
521 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
522 * macro avoids calling this routine if toggling isn't necessary according
523 * to our records (which could be wrong if some XS code has changed the
524 * locale behind our back) */
0d071d52 525
837ce802 526 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 527 PL_numeric_standard = TRUE;
4c68b815 528 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 529 set_numeric_radix(0);
7d4bcc4a
KW
530
531# ifdef DEBUGGING
532
2fcc0ca9
KW
533 if (DEBUG_L_TEST || debug_initialization) {
534 PerlIO_printf(Perl_debug_log,
58e4a467 535 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 536 }
98994639 537
7d4bcc4a 538# endif
98994639 539#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 540
98994639
HS
541}
542
543void
5792c642 544Perl_set_numeric_underlying(pTHX)
98994639 545{
7d4bcc4a 546
98994639 547#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 548
28c1bf33 549 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
550 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
551 * instead of calling this directly. The macro avoids calling this routine
552 * if toggling isn't necessary according to our records (which could be
553 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 554
837ce802 555 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 556 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 557 PL_numeric_underlying = TRUE;
3c62fd46 558 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
559
560# ifdef DEBUGGING
561
2fcc0ca9
KW
562 if (DEBUG_L_TEST || debug_initialization) {
563 PerlIO_printf(Perl_debug_log,
58e4a467 564 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
565 PL_numeric_name);
566 }
98994639 567
7d4bcc4a 568# endif
98994639 569#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 570
98994639
HS
571}
572
573/*
574 * Set up for a new ctype locale.
575 */
a4f00dcc
KW
576STATIC void
577S_new_ctype(pTHX_ const char *newctype)
98994639 578{
7d4bcc4a
KW
579
580#ifndef USE_LOCALE_CTYPE
581
582 PERL_ARGS_ASSERT_NEW_CTYPE;
583 PERL_UNUSED_ARG(newctype);
584 PERL_UNUSED_CONTEXT;
585
586#else
0d071d52 587
291a84fb 588 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
589 * core Perl this and that 'newctype' is the name of the new locale.
590 *
591 * This function sets up the folding arrays for all 256 bytes, assuming
592 * that tofold() is tolc() since fold case is not a concept in POSIX,
593 *
594 * Any code changing the locale (outside this file) should use
595 * POSIX::setlocale, which calls this function. Therefore this function
596 * should be called directly only from this file and from
597 * POSIX::setlocale() */
598
27da23d5 599 dVAR;
68067e4e 600 UV i;
98994639 601
7918f24d
NC
602 PERL_ARGS_ASSERT_NEW_CTYPE;
603
215c5139
KW
604 /* We will replace any bad locale warning with 1) nothing if the new one is
605 * ok; or 2) a new warning for the bad new locale */
606 if (PL_warn_locale) {
607 SvREFCNT_dec_NN(PL_warn_locale);
608 PL_warn_locale = NULL;
609 }
610
c1284011 611 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
612
613 /* A UTF-8 locale gets standard rules. But note that code still has to
614 * handle this specially because of the three problematic code points */
615 if (PL_in_utf8_CTYPE_locale) {
616 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
617 }
618 else {
8c6180a9
KW
619 /* Assume enough space for every character being bad. 4 spaces each
620 * for the 94 printable characters that are output like "'x' "; and 5
621 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
622 * NUL */
623 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
624
cc9eaeb0
KW
625 /* Don't check for problems if we are suppressing the warnings */
626 bool check_for_problems = ckWARN_d(WARN_LOCALE)
627 || UNLIKELY(DEBUG_L_TEST);
8c6180a9
KW
628 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
629 to start */
630 unsigned int bad_count = 0; /* Count of bad characters */
631
baa60164 632 for (i = 0; i < 256; i++) {
5a8a8f21
KW
633 if (isupper(i))
634 PL_fold_locale[i] = (U8) tolower(i);
635 else if (islower(i))
636 PL_fold_locale[i] = (U8) toupper(i);
baa60164
KW
637 else
638 PL_fold_locale[i] = (U8) i;
8c6180a9
KW
639
640 /* If checking for locale problems, see if the native ASCII-range
641 * printables plus \n and \t are in their expected categories in
642 * the new locale. If not, this could mean big trouble, upending
643 * Perl's and most programs' assumptions, like having a
644 * metacharacter with special meaning become a \w. Fortunately,
645 * it's very rare to find locales that aren't supersets of ASCII
646 * nowadays. It isn't a problem for most controls to be changed
647 * into something else; we check only \n and \t, though perhaps \r
648 * could be an issue as well. */
7d4bcc4a 649 if ( check_for_problems
8c6180a9
KW
650 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
651 {
adf947ba
KW
652 if ( cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC(i))
653 || cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i))
654 || cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i))
655 || cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i))
656 || cBOOL(islower(i)) != cBOOL(isLOWER_A(i))
657 || cBOOL(isprint(i)) != cBOOL(isPRINT_A(i))
658 || cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i))
659 || cBOOL(isspace(i)) != cBOOL(isSPACE_A(i))
660 || cBOOL(isupper(i)) != cBOOL(isUPPER_A(i))
661 || cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i))
662 || tolower(i) != (int) toLOWER_A(i)
663 || toupper(i) != (int) toUPPER_A(i)
664 || (i == '\n' && ! isCNTRL_LC(i)))
8c6180a9
KW
665 {
666 if (bad_count) { /* Separate multiple entries with a
667 blank */
668 bad_chars_list[bad_count++] = ' ';
669 }
670 bad_chars_list[bad_count++] = '\'';
671 if (isPRINT_A(i)) {
672 bad_chars_list[bad_count++] = (char) i;
673 }
674 else {
675 bad_chars_list[bad_count++] = '\\';
676 if (i == '\n') {
677 bad_chars_list[bad_count++] = 'n';
678 }
679 else {
680 assert(i == '\t');
681 bad_chars_list[bad_count++] = 't';
682 }
683 }
684 bad_chars_list[bad_count++] = '\'';
685 bad_chars_list[bad_count] = '\0';
686 }
687 }
688 }
689
7d4bcc4a
KW
690# ifdef MB_CUR_MAX
691
8c6180a9 692 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 693 * this locale requires more than one byte, there are going to be
8c6180a9 694 * problems. */
9c8a6dc2
KW
695 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
696 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
697 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
698
ba1a4362
KW
699 if (check_for_problems && MB_CUR_MAX > 1
700
701 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
702 * locale. Just assume that the implementation for them (plus
703 * for POSIX) is correct and the > 1 value is spurious. (Since
704 * these are specially handled to never be considered UTF-8
705 * locales, as long as this is the only problem, everything
706 * should work fine */
707 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
708 {
8c6180a9
KW
709 multi_byte_locale = TRUE;
710 }
7d4bcc4a
KW
711
712# endif
8c6180a9
KW
713
714 if (bad_count || multi_byte_locale) {
780fcc9f 715 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 716 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 717 newctype,
8c6180a9
KW
718 (multi_byte_locale)
719 ? " Some characters in it are not recognized by"
720 " Perl."
721 : "",
722 (bad_count)
723 ? "\nThe following characters (and maybe others)"
724 " may not have the same meaning as the Perl"
725 " program expects:\n"
726 : "",
727 (bad_count)
728 ? bad_chars_list
729 : ""
730 );
cc9eaeb0 731 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
732 * output the message now. If not in that scope, we save the
733 * message to be output at the first operation using this locale,
734 * if that actually happens. Most programs don't use locales, so
735 * they are immune to bad ones. */
cc9eaeb0 736 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f
KW
737
738 /* We have to save 'newctype' because the setlocale() just
739 * below may destroy it. The next setlocale() further down
740 * should restore it properly so that the intermediate change
741 * here is transparent to this function's caller */
742 const char * const badlocale = savepv(newctype);
743
837ce802 744 do_setlocale_c(LC_CTYPE, "C");
780fcc9f
KW
745
746 /* The '0' below suppresses a bogus gcc compiler warning */
747 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 748
837ce802 749 do_setlocale_c(LC_CTYPE, badlocale);
c0f3a893 750 Safefree(badlocale);
bddebb56
KW
751
752 if (IN_LC(LC_CTYPE)) {
753 SvREFCNT_dec_NN(PL_warn_locale);
754 PL_warn_locale = NULL;
755 }
780fcc9f 756 }
baa60164 757 }
31f05a37 758 }
98994639
HS
759
760#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 761
98994639
HS
762}
763
98994639 764void
2726666d
KW
765Perl__warn_problematic_locale()
766{
2726666d
KW
767
768#ifdef USE_LOCALE_CTYPE
769
5f04a188
KW
770 dTHX;
771
772 /* Internal-to-core function that outputs the message in PL_warn_locale,
773 * and then NULLS it. Should be called only through the macro
774 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
775
2726666d 776 if (PL_warn_locale) {
2726666d
KW
777 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
778 SvPVX(PL_warn_locale),
779 0 /* dummy to avoid compiler warning */ );
2726666d
KW
780 SvREFCNT_dec_NN(PL_warn_locale);
781 PL_warn_locale = NULL;
782 }
783
784#endif
785
786}
787
a4f00dcc
KW
788STATIC void
789S_new_collate(pTHX_ const char *newcoll)
98994639 790{
7d4bcc4a
KW
791
792#ifndef USE_LOCALE_COLLATE
793
794 PERL_UNUSED_ARG(newcoll);
795 PERL_UNUSED_CONTEXT;
796
797#else
0d071d52 798
291a84fb 799 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
800 * core Perl this and that 'newcoll' is the name of the new locale.
801 *
d35fca5f
KW
802 * The design of locale collation is that every locale change is given an
803 * index 'PL_collation_ix'. The first time a string particpates in an
804 * operation that requires collation while locale collation is active, it
805 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
806 * magic includes the collation index, and the transformation of the string
807 * by strxfrm(), q.v. That transformation is used when doing comparisons,
808 * instead of the string itself. If a string changes, the magic is
809 * cleared. The next time the locale changes, the index is incremented,
810 * and so we know during a comparison that the transformation is not
811 * necessarily still valid, and so is recomputed. Note that if the locale
812 * changes enough times, the index could wrap (a U32), and it is possible
813 * that a transformation would improperly be considered valid, leading to
814 * an unlikely bug */
0d071d52 815
98994639
HS
816 if (! newcoll) {
817 if (PL_collation_name) {
818 ++PL_collation_ix;
819 Safefree(PL_collation_name);
820 PL_collation_name = NULL;
821 }
822 PL_collation_standard = TRUE;
00bf60ca 823 is_standard_collation:
98994639
HS
824 PL_collxfrm_base = 0;
825 PL_collxfrm_mult = 2;
165a1c52 826 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 827 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 828 PL_strxfrm_max_cp = 0;
98994639
HS
829 return;
830 }
831
d35fca5f 832 /* If this is not the same locale as currently, set the new one up */
98994639
HS
833 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
834 ++PL_collation_ix;
835 Safefree(PL_collation_name);
836 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 837 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
838 if (PL_collation_standard) {
839 goto is_standard_collation;
840 }
98994639 841
165a1c52 842 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 843 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 844 PL_strxfrm_max_cp = 0;
165a1c52 845
59c018b9
KW
846 /* A locale collation definition includes primary, secondary, tertiary,
847 * etc. weights for each character. To sort, the primary weights are
848 * used, and only if they compare equal, then the secondary weights are
849 * used, and only if they compare equal, then the tertiary, etc.
850 *
851 * strxfrm() works by taking the input string, say ABC, and creating an
852 * output transformed string consisting of first the primary weights,
853 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
854 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
855 * may not have weights at every level. In our example, let's say B
856 * doesn't have a tertiary weight, and A doesn't have a secondary
857 * weight. The constructed string is then going to be
858 * A¹B¹C¹ B²C² A³C³ ....
859 * This has the desired effect that strcmp() will look at the secondary
860 * or tertiary weights only if the strings compare equal at all higher
861 * priority weights. The spaces shown here, like in
c342d20e 862 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
863 * are not just for readability. In the general case, these must
864 * actually be bytes, which we will call here 'separator weights'; and
865 * they must be smaller than any other weight value, but since these
866 * are C strings, only the terminating one can be a NUL (some
867 * implementations may include a non-NUL separator weight just before
868 * the NUL). Implementations tend to reserve 01 for the separator
869 * weights. They are needed so that a shorter string's secondary
870 * weights won't be misconstrued as primary weights of a longer string,
871 * etc. By making them smaller than any other weight, the shorter
872 * string will sort first. (Actually, if all secondary weights are
873 * smaller than all primary ones, there is no need for a separator
874 * weight between those two levels, etc.)
875 *
876 * The length of the transformed string is roughly a linear function of
877 * the input string. It's not exactly linear because some characters
878 * don't have weights at all levels. When we call strxfrm() we have to
879 * allocate some memory to hold the transformed string. The
880 * calculations below try to find coefficients 'm' and 'b' for this
881 * locale so that m*x + b equals how much space we need, given the size
882 * of the input string in 'x'. If we calculate too small, we increase
883 * the size as needed, and call strxfrm() again, but it is better to
884 * get it right the first time to avoid wasted expensive string
885 * transformations. */
886
98994639 887 {
79f120c8
KW
888 /* We use the string below to find how long the tranformation of it
889 * is. Almost all locales are supersets of ASCII, or at least the
890 * ASCII letters. We use all of them, half upper half lower,
891 * because if we used fewer, we might hit just the ones that are
892 * outliers in a particular locale. Most of the strings being
893 * collated will contain a preponderance of letters, and even if
894 * they are above-ASCII, they are likely to have the same number of
895 * weight levels as the ASCII ones. It turns out that digits tend
896 * to have fewer levels, and some punctuation has more, but those
897 * are relatively sparse in text, and khw believes this gives a
898 * reasonable result, but it could be changed if experience so
899 * dictates. */
900 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
901 char * x_longer; /* Transformed 'longer' */
902 Size_t x_len_longer; /* Length of 'x_longer' */
903
904 char * x_shorter; /* We also transform a substring of 'longer' */
905 Size_t x_len_shorter;
906
a4a439fb 907 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
908 * are interested only in its length). It is used because it has
909 * the intelligence to handle all cases, but to work, it needs some
910 * values of 'm' and 'b' to get it started. For the purposes of
911 * this calculation we use a very conservative estimate of 'm' and
912 * 'b'. This assumes a weight can be multiple bytes, enough to
913 * hold any UV on the platform, and there are 5 levels, 4 weight
914 * bytes, and a trailing NUL. */
915 PL_collxfrm_base = 5;
916 PL_collxfrm_mult = 5 * sizeof(UV);
917
918 /* Find out how long the transformation really is */
a4a439fb
KW
919 x_longer = _mem_collxfrm(longer,
920 sizeof(longer) - 1,
921 &x_len_longer,
922
923 /* We avoid converting to UTF-8 in the
924 * called function by telling it the
925 * string is in UTF-8 if the locale is a
926 * UTF-8 one. Since the string passed
927 * here is invariant under UTF-8, we can
928 * claim it's UTF-8 even though it isn't.
929 * */
930 PL_in_utf8_COLLATE_locale);
79f120c8
KW
931 Safefree(x_longer);
932
933 /* Find out how long the transformation of a substring of 'longer'
934 * is. Together the lengths of these transformations are
935 * sufficient to calculate 'm' and 'b'. The substring is all of
936 * 'longer' except the first character. This minimizes the chances
937 * of being swayed by outliers */
a4a439fb 938 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 939 sizeof(longer) - 2,
a4a439fb
KW
940 &x_len_shorter,
941 PL_in_utf8_COLLATE_locale);
79f120c8
KW
942 Safefree(x_shorter);
943
944 /* If the results are nonsensical for this simple test, the whole
945 * locale definition is suspect. Mark it so that locale collation
946 * is not active at all for it. XXX Should we warn? */
947 if ( x_len_shorter == 0
948 || x_len_longer == 0
949 || x_len_shorter >= x_len_longer)
950 {
951 PL_collxfrm_mult = 0;
952 PL_collxfrm_base = 0;
953 }
954 else {
955 SSize_t base; /* Temporary */
956
957 /* We have both: m * strlen(longer) + b = x_len_longer
958 * m * strlen(shorter) + b = x_len_shorter;
959 * subtracting yields:
960 * m * (strlen(longer) - strlen(shorter))
961 * = x_len_longer - x_len_shorter
962 * But we have set things up so that 'shorter' is 1 byte smaller
963 * than 'longer'. Hence:
964 * m = x_len_longer - x_len_shorter
965 *
966 * But if something went wrong, make sure the multiplier is at
967 * least 1.
968 */
969 if (x_len_longer > x_len_shorter) {
970 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
971 }
972 else {
973 PL_collxfrm_mult = 1;
974 }
975
976 /* mx + b = len
977 * so: b = len - mx
978 * but in case something has gone wrong, make sure it is
979 * non-negative */
980 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
981 if (base < 0) {
982 base = 0;
983 }
984
985 /* Add 1 for the trailing NUL */
986 PL_collxfrm_base = base + 1;
987 }
58eebef2 988
7d4bcc4a
KW
989# ifdef DEBUGGING
990
58eebef2
KW
991 if (DEBUG_L_TEST || debug_initialization) {
992 PerlIO_printf(Perl_debug_log,
b07929e4
KW
993 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
994 "x_len_longer=%zu,"
995 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
996 __FILE__, __LINE__,
997 PL_in_utf8_COLLATE_locale,
998 x_len_shorter, x_len_longer,
999 PL_collxfrm_mult, PL_collxfrm_base);
1000 }
7d4bcc4a
KW
1001# endif
1002
98994639
HS
1003 }
1004 }
1005
1006#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1007
98994639
HS
1008}
1009
d2b24094 1010#ifdef WIN32
b8cc575c 1011
a4f00dcc 1012STATIC char *
b8cc575c 1013S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1014{
1015 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1016 * difference between the two unless the input locale is "", which normally
1017 * means on Windows to get the machine default, which is set via the
1018 * computer's "Regional and Language Options" (or its current equivalent).
1019 * In POSIX, it instead means to find the locale from the user's
1020 * environment. This routine changes the Windows behavior to first look in
1021 * the environment, and, if anything is found, use that instead of going to
1022 * the machine default. If there is no environment override, the machine
1023 * default is used, by calling the real setlocale() with "".
1024 *
1025 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1026 * use the particular category's variable if set; otherwise to use the LANG
1027 * variable. */
b385bb4d 1028
175c4cf9 1029 bool override_LC_ALL = FALSE;
89f7b9aa 1030 char * result;
e5f10d49 1031 unsigned int i;
89f7b9aa 1032
b385bb4d 1033 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1034
1035# ifdef LC_ALL
1036
b385bb4d
KW
1037 locale = PerlEnv_getenv("LC_ALL");
1038 if (! locale) {
e5f10d49
KW
1039 if (category == LC_ALL) {
1040 override_LC_ALL = TRUE;
1041 }
1042 else {
7d4bcc4a
KW
1043
1044# endif
7d4bcc4a 1045
e5f10d49
KW
1046 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1047 if (category == categories[i]) {
1048 locale = PerlEnv_getenv(category_names[i]);
1049 goto found_locale;
1050 }
1051 }
7d4bcc4a 1052
b385bb4d 1053 locale = PerlEnv_getenv("LANG");
481465ea 1054 if (! locale) {
b385bb4d
KW
1055 locale = "";
1056 }
e5f10d49
KW
1057
1058 found_locale: ;
7d4bcc4a
KW
1059
1060# ifdef LC_ALL
1061
e5f10d49 1062 }
b385bb4d 1063 }
7d4bcc4a
KW
1064
1065# endif
1066
b385bb4d
KW
1067 }
1068
89f7b9aa 1069 result = setlocale(category, locale);
bbc98134 1070 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
a4f00dcc 1071 setlocale_debug_string(category, locale, result)));
89f7b9aa 1072
481465ea 1073 if (! override_LC_ALL) {
89f7b9aa
KW
1074 return result;
1075 }
1076
dfd77d7a 1077 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
1078 * LANG variable or the system default if there is no LANG. But these have
1079 * lower priority than the other LC_foo variables, so override it for each
1080 * one that is set. (If they are set to "", it means to use the same thing
1081 * we just set LC_ALL to, so can skip) */
7d4bcc4a 1082
948523db 1083 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
1084 result = PerlEnv_getenv(category_names[i]);
1085 if (result && strNE(result, "")) {
1086 setlocale(categories[i], result);
1087 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1088 __FILE__, __LINE__,
1089 setlocale_debug_string(categories[i], result, "not captured")));
1090 }
89f7b9aa 1091 }
7d4bcc4a 1092
bbc98134
KW
1093 result = setlocale(LC_ALL, NULL);
1094 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1095 __FILE__, __LINE__,
a4f00dcc 1096 setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 1097
bbc98134 1098 return result;
b385bb4d
KW
1099}
1100
1101#endif
1102
a4f00dcc
KW
1103char *
1104Perl_setlocale(int category, const char * locale)
1105{
1106 /* This wraps POSIX::setlocale() */
1107
1108 char * retval;
1159483a 1109 char * newlocale;
a4f00dcc
KW
1110 dTHX;
1111
a4f00dcc
KW
1112#ifdef USE_LOCALE_NUMERIC
1113
291a84fb
KW
1114 /* A NULL locale means only query what the current one is. We have the
1115 * LC_NUMERIC name saved, because we are normally switched into the C
1116 * locale for it. For an LC_ALL query, switch back to get the correct
1117 * results. All other categories don't require special handling */
a4f00dcc
KW
1118 if (locale == NULL) {
1119 if (category == LC_NUMERIC) {
1120 return savepv(PL_numeric_name);
1121 }
1122
7d4bcc4a 1123# ifdef LC_ALL
a4f00dcc 1124
c0d737a8
KW
1125 else if (category == LC_ALL && ! PL_numeric_underlying) {
1126
a4f00dcc
KW
1127 SET_NUMERIC_UNDERLYING();
1128 }
1129
7d4bcc4a 1130# endif
a4f00dcc
KW
1131
1132 }
1133
1134#endif
1135
e5f10d49
KW
1136 /* Save retval since subsequent setlocale() calls may overwrite it. */
1137 retval = savepv(do_setlocale_r(category, locale));
a4f00dcc
KW
1138
1139 DEBUG_L(PerlIO_printf(Perl_debug_log,
1140 "%s:%d: %s\n", __FILE__, __LINE__,
1141 setlocale_debug_string(category, locale, retval)));
1142 if (! retval) {
1143 /* Should never happen that a query would return an error, but be
1144 * sure and reset to C locale */
1145 if (locale == 0) {
1146 SET_NUMERIC_STANDARD();
1147 }
7d4bcc4a 1148
a4f00dcc
KW
1149 return NULL;
1150 }
1151
a4f00dcc
KW
1152 /* If locale == NULL, we are just querying the state, but may have switched
1153 * to NUMERIC_UNDERLYING. Switch back before returning. */
1154 if (locale == NULL) {
1155 SET_NUMERIC_STANDARD();
1156 return retval;
1157 }
a4f00dcc 1158
1159483a
KW
1159 /* Now that have switched locales, we have to update our records to
1160 * correspond. */
a4f00dcc 1161
1159483a 1162 switch (category) {
a4f00dcc 1163
1159483a 1164#ifdef USE_LOCALE_CTYPE
a4f00dcc 1165
1159483a
KW
1166 case LC_CTYPE:
1167 new_ctype(retval);
1168 break;
a4f00dcc 1169
1159483a 1170#endif
a4f00dcc
KW
1171#ifdef USE_LOCALE_COLLATE
1172
1159483a
KW
1173 case LC_COLLATE:
1174 new_collate(retval);
1175 break;
a4f00dcc 1176
1159483a 1177#endif
a4f00dcc
KW
1178#ifdef USE_LOCALE_NUMERIC
1179
1159483a
KW
1180 case LC_NUMERIC:
1181 new_numeric(retval);
1182 break;
a4f00dcc 1183
1159483a
KW
1184#endif
1185#ifdef LC_ALL
a4f00dcc 1186
1159483a 1187 case LC_ALL:
a4f00dcc 1188
1159483a
KW
1189 /* LC_ALL updates all the things we care about. The values may not
1190 * be the same as 'retval', as the locale "" may have set things
1191 * individually */
a4f00dcc 1192
1159483a 1193# ifdef USE_LOCALE_CTYPE
a4f00dcc 1194
1159483a
KW
1195 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1196 new_ctype(newlocale);
a4f00dcc 1197
1159483a
KW
1198# endif /* USE_LOCALE_CTYPE */
1199# ifdef USE_LOCALE_COLLATE
1200
1201 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1202 new_collate(newlocale);
a4f00dcc 1203
7d4bcc4a 1204# endif
1159483a 1205# ifdef USE_LOCALE_NUMERIC
a4f00dcc 1206
1159483a
KW
1207 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1208 new_numeric(newlocale);
a4f00dcc 1209
1159483a
KW
1210# endif /* USE_LOCALE_NUMERIC */
1211#endif /* LC_ALL */
a4f00dcc 1212
1159483a
KW
1213 default:
1214 break;
a4f00dcc
KW
1215 }
1216
1217 return retval;
1218
f7416781
KW
1219
1220}
1221
1222PERL_STATIC_INLINE const char *
1223S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1224{
1225 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1226 * growing it if necessary */
1227
1228 const Size_t string_size = strlen(string) + offset + 1;
1229
1230 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1231
1232 if (*buf_size == 0) {
1233 Newx(*buf, string_size, char);
1234 *buf_size = string_size;
1235 }
1236 else if (string_size > *buf_size) {
1237 Renew(*buf, string_size, char);
1238 *buf_size = string_size;
1239 }
1240
1241 Copy(string, *buf + offset, string_size - offset, char);
1242 return *buf;
1243}
1244
1245/*
1246
1247=head1 Locale-related functions and macros
1248
1249=for apidoc Perl_langinfo
1250
7d4bcc4a 1251This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
1252taking the same C<item> parameter values, and returning the same information.
1253But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1254of Perl's locale handling from your code, and can be used on systems that lack
1255a native C<nl_langinfo>.
1256
1257Expanding on these:
1258
1259=over
1260
1261=item *
1262
1263It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1264without you having to write extra code. The reason for the extra code would be
1265because these are from the C<LC_NUMERIC> locale category, which is normally
1266kept set to the C locale by Perl, no matter what the underlying locale is
1267supposed to be, and so to get the expected results, you have to temporarily
1268toggle into the underlying locale, and later toggle back. (You could use
1269plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1270but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1271keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1272expecting the radix (decimal point) character to be a dot.)
1273
1274=item *
1275
1276Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1277makes your code more portable. Of the fifty-some possible items specified by
1278the POSIX 2008 standard,
1279L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1280only two are completely unimplemented. It uses various techniques to recover
1281the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1282both of which are specified in C89, so should be always be available. Later
1283C<strftime()> versions have additional capabilities; C<""> is returned for
1284those not available on your system.
1285
1286The details for those items which may differ from what this emulation returns
1287and what a native C<nl_langinfo()> would return are:
1288
1289=over
1290
1291=item C<CODESET>
1292
1293=item C<ERA>
1294
1295Unimplemented, so returns C<"">.
1296
1297=item C<YESEXPR>
1298
c1566110
KW
1299=item C<YESSTR>
1300
f7416781
KW
1301=item C<NOEXPR>
1302
c1566110
KW
1303=item C<NOSTR>
1304
1305Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
1306removed from POSIX 2008, and are retained for backwards compatibility. Your
1307platform's C<nl_langinfo> may not support them.
f7416781
KW
1308
1309=item C<D_FMT>
1310
1311Always evaluates to C<%x>, the locale's appropriate date representation.
1312
1313=item C<T_FMT>
1314
1315Always evaluates to C<%X>, the locale's appropriate time representation.
1316
1317=item C<D_T_FMT>
1318
1319Always evaluates to C<%c>, the locale's appropriate date and time
1320representation.
1321
1322=item C<CRNCYSTR>
1323
1324The return may be incorrect for those rare locales where the currency symbol
1325replaces the radix character.
1326Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1327to work differently.
1328
1329=item C<ALT_DIGITS>
1330
1331Currently this gives the same results as Linux does.
1332Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1333to work differently.
1334
1335=item C<ERA_D_FMT>
1336
1337=item C<ERA_T_FMT>
1338
1339=item C<ERA_D_T_FMT>
1340
1341=item C<T_FMT_AMPM>
1342
1343These are derived by using C<strftime()>, and not all versions of that function
1344know about them. C<""> is returned for these on such systems.
1345
1346=back
1347
1348When using C<Perl_langinfo> on systems that don't have a native
1349C<nl_langinfo()>, you must
1350
1351 #include "perl_langinfo.h"
1352
1353before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1354C<#include> with this one. (Doing it this way keeps out the symbols that plain
1355C<langinfo.h> imports into the namespace for code that doesn't need it.)
1356
1357You also should not use the bare C<langinfo.h> item names, but should preface
1358them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1359The C<PERL_I<foo>> versions will also work for this function on systems that do
1360have a native C<nl_langinfo>.
1361
1362=item *
1363
1364It is thread-friendly, returning its result in a buffer that won't be
1365overwritten by another thread, so you don't have to code for that possibility.
1366The buffer can be overwritten by the next call to C<nl_langinfo> or
1367C<Perl_langinfo> in the same thread.
1368
1369=item *
1370
7d4bcc4a 1371ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
1372*>>, but you are (only by documentation) forbidden to write into the buffer.
1373By declaring this C<const>, the compiler enforces this restriction. The extra
1374C<const> is why this isn't an unequivocal drop-in replacement for
1375C<nl_langinfo>.
1376
1377=back
1378
1379The original impetus for C<Perl_langinfo()> was so that code that needs to
1380find out the current currency symbol, floating point radix character, or digit
1381grouping separator can use, on all systems, the simpler and more
1382thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1383pain to make thread-friendly. For other fields returned by C<localeconv>, it
1384is better to use the methods given in L<perlcall> to call
1385L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1386
1387=cut
1388
1389*/
1390
1391const char *
1392#ifdef HAS_NL_LANGINFO
1393Perl_langinfo(const nl_item item)
1394#else
1395Perl_langinfo(const int item)
1396#endif
1397{
f61748ac
KW
1398 return my_nl_langinfo(item, TRUE);
1399}
1400
1401const char *
1402#ifdef HAS_NL_LANGINFO
1403S_my_nl_langinfo(const nl_item item, bool toggle)
1404#else
1405S_my_nl_langinfo(const int item, bool toggle)
1406#endif
1407{
ae74815b 1408 dTHX;
f7416781 1409
5a854ab3
KW
1410 /* We only need to toggle into the underlying LC_NUMERIC locale for these
1411 * two items, and only if not already there */
1412 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
1413 || PL_numeric_underlying))
1414 {
1415 toggle = FALSE;
1416 }
1417
ab340fff
KW
1418#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1419#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1420
ab340fff 1421 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1422 * for those items dependent on it. This must be copied to a buffer before
1423 * switching back, as some systems destroy the buffer when setlocale() is
1424 * called */
f7416781 1425
038d3702
KW
1426 {
1427 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
1428
b2fee59c 1429 if (toggle) {
038d3702 1430 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 1431 }
f7416781 1432
5acc4454
KW
1433 LOCALE_LOCK; /* Prevent interference from another thread executing
1434 this code section (the only call to nl_langinfo in
1435 the core) */
1436
b2fee59c
KW
1437 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf,
1438 &PL_langinfo_bufsize, 0);
f7416781 1439
5acc4454
KW
1440 LOCALE_UNLOCK;
1441
b2fee59c 1442 if (toggle) {
038d3702 1443 RESTORE_LC_NUMERIC();
b2fee59c 1444 }
f7416781
KW
1445 }
1446
ab340fff
KW
1447# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1448
5a854ab3 1449 {
b2fee59c
KW
1450 bool do_free = FALSE;
1451 locale_t cur = uselocale((locale_t) 0);
ab340fff 1452
b2fee59c
KW
1453 if (cur == LC_GLOBAL_LOCALE) {
1454 cur = duplocale(LC_GLOBAL_LOCALE);
1455 do_free = TRUE;
1456 }
ab340fff 1457
b2fee59c
KW
1458 if (toggle) {
1459 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1460 do_free = TRUE;
1461 }
ab340fff 1462
b2fee59c
KW
1463 save_to_buffer(nl_langinfo_l(item, cur),
1464 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1465 if (do_free) {
1466 freelocale(cur);
1467 }
5a854ab3 1468 }
ab340fff 1469
c1566110
KW
1470# endif
1471
1472 if (strEQ(PL_langinfo_buf, "")) {
1473 if (item == PERL_YESSTR) {
1474 return "yes";
1475 }
1476 if (item == PERL_NOSTR) {
1477 return "no";
1478 }
1479 }
1480
ab340fff
KW
1481 return PL_langinfo_buf;
1482
f7416781 1483#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
1484
1485 {
1486
f7416781
KW
1487# ifdef HAS_LOCALECONV
1488
43dd6b15 1489 const struct lconv* lc;
038d3702 1490 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
1491
1492# endif
1493# ifdef HAS_STRFTIME
1494
43dd6b15
KW
1495 struct tm tm;
1496 bool return_format = FALSE; /* Return the %format, not the value */
1497 const char * format;
f7416781
KW
1498
1499# endif
1500
43dd6b15
KW
1501 /* We copy the results to a per-thread buffer, even if not
1502 * multi-threaded. This is in part to simplify this code, and partly
1503 * because we need a buffer anyway for strftime(), and partly because a
1504 * call of localeconv() could otherwise wipe out the buffer, and the
1505 * programmer would not be expecting this, as this is a nl_langinfo()
1506 * substitute after all, so s/he might be thinking their localeconv()
1507 * is safe until another localeconv() call. */
f7416781 1508
43dd6b15
KW
1509 switch (item) {
1510 Size_t len;
1511 const char * retval;
f7416781 1512
43dd6b15
KW
1513 /* These 2 are unimplemented */
1514 case PERL_CODESET:
1515 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 1516
43dd6b15
KW
1517 default:
1518 return "";
f7416781 1519
43dd6b15
KW
1520 /* We use only an English set, since we don't know any more */
1521 case PERL_YESEXPR: return "^[+1yY]";
1522 case PERL_YESSTR: return "yes";
1523 case PERL_NOEXPR: return "^[-0nN]";
1524 case PERL_NOSTR: return "no";
f7416781
KW
1525
1526# ifdef HAS_LOCALECONV
1527
43dd6b15 1528 case PERL_CRNCYSTR:
f7416781 1529
43dd6b15
KW
1530 /* We don't bother with localeconv_l() because any system that
1531 * has it is likely to also have nl_langinfo() */
291a84fb 1532
5acc4454
KW
1533 LOCALE_LOCK; /* Prevent interference with other threads
1534 using localeconv() */
1535
43dd6b15
KW
1536 lc = localeconv();
1537 if ( ! lc
1538 || ! lc->currency_symbol
1539 || strEQ("", lc->currency_symbol))
1540 {
1541 LOCALE_UNLOCK;
1542 return "";
1543 }
f7416781 1544
43dd6b15
KW
1545 /* Leave the first spot empty to be filled in below */
1546 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1547 &PL_langinfo_bufsize, 1);
1548 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1549 { /* khw couldn't figure out how the localedef specifications
1550 would show that the $ should replace the radix; this is
1551 just a guess as to how it might work.*/
1552 *PL_langinfo_buf = '.';
1553 }
1554 else if (lc->p_cs_precedes) {
1555 *PL_langinfo_buf = '-';
1556 }
1557 else {
1558 *PL_langinfo_buf = '+';
1559 }
f7416781 1560
43dd6b15
KW
1561 LOCALE_UNLOCK;
1562 break;
f7416781 1563
43dd6b15
KW
1564 case PERL_RADIXCHAR:
1565 case PERL_THOUSEP:
f7416781 1566
43dd6b15 1567 if (toggle) {
038d3702 1568 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 1569 }
f7416781 1570
5acc4454
KW
1571 LOCALE_LOCK; /* Prevent interference with other threads
1572 using localeconv() */
1573
43dd6b15
KW
1574 lc = localeconv();
1575 if (! lc) {
33394adc
KW
1576 retval = "";
1577 }
43dd6b15
KW
1578 else {
1579 retval = (item == PERL_RADIXCHAR)
1580 ? lc->decimal_point
1581 : lc->thousands_sep;
1582 if (! retval) {
1583 retval = "";
1584 }
1585 }
f7416781 1586
43dd6b15
KW
1587 save_to_buffer(retval, &PL_langinfo_buf,
1588 &PL_langinfo_bufsize, 0);
f7416781 1589
5acc4454
KW
1590 LOCALE_UNLOCK;
1591
43dd6b15 1592 if (toggle) {
038d3702 1593 RESTORE_LC_NUMERIC();
43dd6b15 1594 }
f7416781 1595
43dd6b15 1596 break;
f7416781
KW
1597
1598# endif
1599# ifdef HAS_STRFTIME
1600
43dd6b15
KW
1601 /* These are defined by C89, so we assume that strftime supports
1602 * them, and so are returned unconditionally; they may not be what
1603 * the locale actually says, but should give good enough results
1604 * for someone using them as formats (as opposed to trying to parse
1605 * them to figure out what the locale says). The other format
1606 * items are actually tested to verify they work on the platform */
1607 case PERL_D_FMT: return "%x";
1608 case PERL_T_FMT: return "%X";
1609 case PERL_D_T_FMT: return "%c";
1610
1611 /* These formats are only available in later strfmtime's */
1612 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1613 case PERL_T_FMT_AMPM:
1614
1615 /* The rest can be gotten from most versions of strftime(). */
1616 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1617 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1618 case PERL_ABDAY_7:
1619 case PERL_ALT_DIGITS:
1620 case PERL_AM_STR: case PERL_PM_STR:
1621 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1622 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1623 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1624 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1625 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1626 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1627 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1628 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1629 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
1630 case PERL_MON_12:
1631
1632 LOCALE_LOCK;
1633
1634 init_tm(&tm); /* Precaution against core dumps */
1635 tm.tm_sec = 30;
1636 tm.tm_min = 30;
1637 tm.tm_hour = 6;
1638 tm.tm_year = 2017 - 1900;
1639 tm.tm_wday = 0;
1640 tm.tm_mon = 0;
1641 switch (item) {
1642 default:
1643 LOCALE_UNLOCK;
1644 Perl_croak(aTHX_
1645 "panic: %s: %d: switch case: %d problem",
1646 __FILE__, __LINE__, item);
1647 NOT_REACHED; /* NOTREACHED */
1648
1649 case PERL_PM_STR: tm.tm_hour = 18;
1650 case PERL_AM_STR:
1651 format = "%p";
1652 break;
1653
1654 case PERL_ABDAY_7: tm.tm_wday++;
1655 case PERL_ABDAY_6: tm.tm_wday++;
1656 case PERL_ABDAY_5: tm.tm_wday++;
1657 case PERL_ABDAY_4: tm.tm_wday++;
1658 case PERL_ABDAY_3: tm.tm_wday++;
1659 case PERL_ABDAY_2: tm.tm_wday++;
1660 case PERL_ABDAY_1:
1661 format = "%a";
1662 break;
1663
1664 case PERL_DAY_7: tm.tm_wday++;
1665 case PERL_DAY_6: tm.tm_wday++;
1666 case PERL_DAY_5: tm.tm_wday++;
1667 case PERL_DAY_4: tm.tm_wday++;
1668 case PERL_DAY_3: tm.tm_wday++;
1669 case PERL_DAY_2: tm.tm_wday++;
1670 case PERL_DAY_1:
1671 format = "%A";
1672 break;
1673
1674 case PERL_ABMON_12: tm.tm_mon++;
1675 case PERL_ABMON_11: tm.tm_mon++;
1676 case PERL_ABMON_10: tm.tm_mon++;
1677 case PERL_ABMON_9: tm.tm_mon++;
1678 case PERL_ABMON_8: tm.tm_mon++;
1679 case PERL_ABMON_7: tm.tm_mon++;
1680 case PERL_ABMON_6: tm.tm_mon++;
1681 case PERL_ABMON_5: tm.tm_mon++;
1682 case PERL_ABMON_4: tm.tm_mon++;
1683 case PERL_ABMON_3: tm.tm_mon++;
1684 case PERL_ABMON_2: tm.tm_mon++;
1685 case PERL_ABMON_1:
1686 format = "%b";
1687 break;
1688
1689 case PERL_MON_12: tm.tm_mon++;
1690 case PERL_MON_11: tm.tm_mon++;
1691 case PERL_MON_10: tm.tm_mon++;
1692 case PERL_MON_9: tm.tm_mon++;
1693 case PERL_MON_8: tm.tm_mon++;
1694 case PERL_MON_7: tm.tm_mon++;
1695 case PERL_MON_6: tm.tm_mon++;
1696 case PERL_MON_5: tm.tm_mon++;
1697 case PERL_MON_4: tm.tm_mon++;
1698 case PERL_MON_3: tm.tm_mon++;
1699 case PERL_MON_2: tm.tm_mon++;
1700 case PERL_MON_1:
1701 format = "%B";
1702 break;
1703
1704 case PERL_T_FMT_AMPM:
1705 format = "%r";
1706 return_format = TRUE;
1707 break;
1708
1709 case PERL_ERA_D_FMT:
1710 format = "%Ex";
1711 return_format = TRUE;
1712 break;
1713
1714 case PERL_ERA_T_FMT:
1715 format = "%EX";
1716 return_format = TRUE;
1717 break;
1718
1719 case PERL_ERA_D_T_FMT:
1720 format = "%Ec";
1721 return_format = TRUE;
1722 break;
1723
1724 case PERL_ALT_DIGITS:
1725 tm.tm_wday = 0;
1726 format = "%Ow"; /* Find the alternate digit for 0 */
1727 break;
1728 }
f7416781 1729
43dd6b15
KW
1730 /* We can't use my_strftime() because it doesn't look at
1731 * tm_wday */
1732 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1733 format, &tm))
1734 {
1735 /* A zero return means one of:
1736 * a) there wasn't enough space in PL_langinfo_buf
1737 * b) the format, like a plain %p, returns empty
1738 * c) it was an illegal format, though some
1739 * implementations of strftime will just return the
1740 * illegal format as a plain character sequence.
1741 *
1742 * To quickly test for case 'b)', try again but precede
1743 * the format with a plain character. If that result is
1744 * still empty, the problem is either 'a)' or 'c)' */
1745
1746 Size_t format_size = strlen(format) + 1;
1747 Size_t mod_size = format_size + 1;
1748 char * mod_format;
1749 char * temp_result;
1750
1751 Newx(mod_format, mod_size, char);
1752 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 1753 *mod_format = ' ';
43dd6b15
KW
1754 my_strlcpy(mod_format + 1, format, mod_size);
1755 len = strftime(temp_result,
1756 PL_langinfo_bufsize,
1757 mod_format, &tm);
1758 Safefree(mod_format);
1759 Safefree(temp_result);
1760
1761 /* If 'len' is non-zero, it means that we had a case like
1762 * %p which means the current locale doesn't use a.m. or
1763 * p.m., and that is valid */
1764 if (len == 0) {
1765
1766 /* Here, still didn't work. If we get well beyond a
1767 * reasonable size, bail out to prevent an infinite
1768 * loop. */
1769
1770 if (PL_langinfo_bufsize > 100 * format_size) {
1771 *PL_langinfo_buf = '\0';
1772 }
1773 else {
1774 /* Double the buffer size to retry; Add 1 in case
1775 * original was 0, so we aren't stuck at 0. */
1776 PL_langinfo_bufsize *= 2;
1777 PL_langinfo_bufsize++;
1778 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1779 continue;
1780 }
1781 }
f7416781 1782
f7416781 1783 break;
43dd6b15 1784 }
f7416781 1785
43dd6b15
KW
1786 /* Here, we got a result.
1787 *
1788 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1789 * alternate format for wday 0. If the value is the same as
1790 * the normal 0, there isn't an alternate, so clear the buffer.
1791 * */
1792 if ( item == PERL_ALT_DIGITS
1793 && strEQ(PL_langinfo_buf, "0"))
1794 {
1795 *PL_langinfo_buf = '\0';
1796 }
f7416781 1797
43dd6b15
KW
1798 /* ALT_DIGITS is problematic. Experiments on it showed that
1799 * strftime() did not always work properly when going from
1800 * alt-9 to alt-10. Only a few locales have this item defined,
1801 * and in all of them on Linux that khw was able to find,
1802 * nl_langinfo() merely returned the alt-0 character, possibly
1803 * doubled. Most Unicode digits are in blocks of 10
1804 * consecutive code points, so that is sufficient information
1805 * for those scripts, as we can infer alt-1, alt-2, .... But
1806 * for a Japanese locale, a CJK ideographic 0 is returned, and
1807 * the CJK digits are not in code point order, so you can't
1808 * really infer anything. The localedef for this locale did
1809 * specify the succeeding digits, so that strftime() works
1810 * properly on them, without needing to infer anything. But
1811 * the nl_langinfo() return did not give sufficient information
1812 * for the caller to understand what's going on. So until
1813 * there is evidence that it should work differently, this
1814 * returns the alt-0 string for ALT_DIGITS.
1815 *
1816 * wday was chosen because its range is all a single digit.
1817 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 1818
43dd6b15 1819 LOCALE_UNLOCK;
f7416781 1820
43dd6b15
KW
1821 /* If to return the format, not the value, overwrite the buffer
1822 * with it. But some strftime()s will keep the original format
1823 * if illegal, so change those to "" */
1824 if (return_format) {
1825 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
1826 *PL_langinfo_buf = '\0';
1827 }
43dd6b15
KW
1828 else {
1829 save_to_buffer(format, &PL_langinfo_buf,
1830 &PL_langinfo_bufsize, 0);
f7416781
KW
1831 }
1832 }
1833
1834 break;
f7416781
KW
1835
1836# endif
1837
43dd6b15 1838 }
f7416781
KW
1839 }
1840
1841 return PL_langinfo_buf;
1842
1843#endif
1844
a4f00dcc 1845}
b385bb4d 1846
98994639
HS
1847/*
1848 * Initialize locale awareness.
1849 */
1850int
1851Perl_init_i18nl10n(pTHX_ int printwarn)
1852{
0e92a118
KW
1853 /* printwarn is
1854 *
1855 * 0 if not to output warning when setup locale is bad
1856 * 1 if to output warning based on value of PERL_BADLANG
1857 * >1 if to output regardless of PERL_BADLANG
1858 *
1859 * returns
98994639 1860 * 1 = set ok or not applicable,
0e92a118
KW
1861 * 0 = fallback to a locale of lower priority
1862 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1863 *
1864 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1865 * set, debugging information is output.
1866 *
1867 * This looks more complicated than it is, mainly due to the #ifdefs.
1868 *
1869 * We try to set LC_ALL to the value determined by the environment. If
1870 * there is no LC_ALL on this platform, we try the individual categories we
1871 * know about. If this works, we are done.
1872 *
1873 * But if it doesn't work, we have to do something else. We search the
1874 * environment variables ourselves instead of relying on the system to do
1875 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1876 * think there is one), and the ultimate fallback "C". This is all done in
1877 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
1878 * more complex. The 'trial_locales' array is initialized with just one
1879 * element; it causes the behavior described in the paragraph above this to
1880 * happen. If that fails, we add elements to 'trial_locales', and do extra
1881 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
1882 *
1883 * On Ultrix, the locale MUST come from the environment, so there is
1884 * preliminary code to set it. I (khw) am not sure that it is necessary,
1885 * and that this couldn't be folded into the loop, but barring any real
1886 * platforms to test on, it's staying as-is
1887 *
1888 * A slight complication is that in embedded Perls, the locale may already
1889 * be set-up, and we don't want to get it from the normal environment
1890 * variables. This is handled by having a special environment variable
1891 * indicate we're in this situation. We simply set setlocale's 2nd
1892 * parameter to be a NULL instead of "". That indicates to setlocale that
1893 * it is not to change anything, but to return the current value,
1894 * effectively initializing perl's db to what the locale already is.
1895 *
1896 * We play the same trick with NULL if a LC_ALL succeeds. We call
1897 * setlocale() on the individual categores with NULL to get their existing
1898 * values for our db, instead of trying to change them.
1899 * */
98994639 1900
0e92a118
KW
1901 int ok = 1;
1902
7d4bcc4a
KW
1903#ifndef USE_LOCALE
1904
1905 PERL_UNUSED_ARG(printwarn);
1906
1907#else /* USE_LOCALE */
7d4bcc4a
KW
1908# ifdef __GLIBC__
1909
175c4cf9 1910 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
1911
1912# endif
65ebb059 1913
ccd65d51
KW
1914 /* NULL uses the existing already set up locale */
1915 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
1916 ? NULL
1917 : "";
c3fcd832
KW
1918 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
1919 unsigned int trial_locales_count;
175c4cf9
KW
1920 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
1921 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 1922 bool setlocale_failure = FALSE;
65ebb059 1923 unsigned int i;
175c4cf9
KW
1924
1925 /* A later getenv() could zap this, so only use here */
1926 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
1927
1928 const bool locwarn = (printwarn > 1
e5f10d49
KW
1929 || ( printwarn
1930 && ( ! bad_lang_use_once
22ff3130 1931 || (
e5f10d49
KW
1932 /* disallow with "" or "0" */
1933 *bad_lang_use_once
1934 && strNE("0", bad_lang_use_once)))));
ea92aad8 1935
291a84fb 1936 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 1937 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
1938
1939 /* current locale for given category; should have been copied so aren't
1940 * volatile */
8de4332b 1941 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 1942
7d4bcc4a
KW
1943# ifdef WIN32
1944
6bce99ee
JH
1945 /* In some systems you can find out the system default locale
1946 * and use that as the fallback locale. */
7d4bcc4a
KW
1947# define SYSTEM_DEFAULT_LOCALE
1948# endif
1949# ifdef SYSTEM_DEFAULT_LOCALE
1950
65ebb059 1951 const char *system_default_locale = NULL;
98994639 1952
7d4bcc4a 1953# endif
948523db
KW
1954
1955# ifndef DEBUGGING
1956# define DEBUG_LOCALE_INIT(a,b,c)
1957# else
7d4bcc4a 1958
8298454c 1959 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
1960
1961# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
1962 STMT_START { \
1963 if (debug_initialization) { \
1964 PerlIO_printf(Perl_debug_log, \
1965 "%s:%d: %s\n", \
1966 __FILE__, __LINE__, \
a4f00dcc 1967 setlocale_debug_string(category, \
2fcc0ca9
KW
1968 locale, \
1969 result)); \
1970 } \
1971 } STMT_END
2fcc0ca9 1972
948523db
KW
1973/* Make sure the parallel arrays are properly set up */
1974# ifdef USE_LOCALE_NUMERIC
1975 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
1976 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
1977# endif
1978# ifdef USE_LOCALE_CTYPE
1979 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
1980 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
1981# endif
1982# ifdef USE_LOCALE_COLLATE
1983 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
1984 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
1985# endif
1986# ifdef USE_LOCALE_TIME
1987 assert(categories[LC_TIME_INDEX] == LC_TIME);
1988 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
1989# endif
1990# ifdef USE_LOCALE_MESSAGES
1991 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
1992 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
1993# endif
1994# ifdef USE_LOCALE_MONETARY
1995 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
1996 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
1997# endif
9821811f
KW
1998# ifdef USE_LOCALE_ADDRESS
1999 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
2000 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
2001# endif
2002# ifdef USE_LOCALE_IDENTIFICATION
2003 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
2004 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
2005# endif
2006# ifdef USE_LOCALE_MEASUREMENT
2007 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
2008 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
2009# endif
2010# ifdef USE_LOCALE_PAPER
2011 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
2012 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
2013# endif
2014# ifdef USE_LOCALE_TELEPHONE
2015 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
2016 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
2017# endif
948523db
KW
2018# ifdef LC_ALL
2019 assert(categories[LC_ALL_INDEX] == LC_ALL);
2020 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
2021 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
2022# endif
2023# endif /* DEBUGGING */
47280b20
KW
2024
2025 /* Initialize the cache of the program's UTF-8ness for the always known
2026 * locales C and POSIX */
2027 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
2028 sizeof(PL_locale_utf8ness));
2029
3ca88433
KW
2030 PL_numeric_radix_sv = newSVpvs(".");
2031
b56c4436 2032# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
2033
2034 /*
2035 * Ultrix setlocale(..., "") fails if there are no environment
2036 * variables from which to get a locale name.
2037 */
2038
b56c4436
KW
2039# ifndef LC_ALL
2040# error Ultrix without LC_ALL not implemented
2041# else
7d4bcc4a 2042
b56c4436
KW
2043 {
2044 bool done = FALSE;
ea92aad8
KW
2045 if (lang) {
2046 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
2047 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
2048 if (sl_result[LC_ALL_INDEX])
2049 done = TRUE;
2050 else
e5f10d49 2051 setlocale_failure = TRUE;
ea92aad8
KW
2052 }
2053 if (! setlocale_failure) {
2054 const char * locale_param;
2055 for (i = 0; i < LC_ALL_INDEX; i++) {
2056 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
2057 ? setlocale_init
2058 : NULL;
2059 sl_result[i] = do_setlocale_r(categories[i], locale_param);
2060 if (! sl_result[i]) {
2061 setlocale_failure = TRUE;
2062 }
2063 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 2064 }
c835d6be 2065 }
7d4bcc4a
KW
2066 }
2067
2068# endif /* LC_ALL */
e5f10d49 2069# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 2070
65ebb059 2071 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
2072 * the list. Normally the loop is executed just once. But if setting the
2073 * locale fails, inside the loop we add fallback trials to the array and so
2074 * will execute the loop multiple times */
c3fcd832
KW
2075 trial_locales[0] = setlocale_init;
2076 trial_locales_count = 1;
7d4bcc4a 2077
65ebb059
KW
2078 for (i= 0; i < trial_locales_count; i++) {
2079 const char * trial_locale = trial_locales[i];
2080
2081 if (i > 0) {
2082
2083 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
2084 * when i==0, but I (khw) don't think that behavior makes much
2085 * sense */
2086 setlocale_failure = FALSE;
2087
7d4bcc4a 2088# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 2089# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 2090
65ebb059
KW
2091 /* On Windows machines, an entry of "" after the 0th means to use
2092 * the system default locale, which we now proceed to get. */
2093 if (strEQ(trial_locale, "")) {
2094 unsigned int j;
2095
2096 /* Note that this may change the locale, but we are going to do
2097 * that anyway just below */
837ce802 2098 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 2099 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 2100
7d4bcc4a 2101 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
2102 * try */
2103 if (! system_default_locale) {
2104 goto next_iteration;
2105 }
2106 for (j = 0; j < trial_locales_count; j++) {
2107 if (strEQ(system_default_locale, trial_locales[j])) {
2108 goto next_iteration;
2109 }
2110 }
2111
2112 trial_locale = system_default_locale;
2113 }
ec0202b5
KW
2114# else
2115# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
2116# endif
7d4bcc4a 2117# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
2118
2119 } /* For i > 0 */
65ebb059 2120
7d4bcc4a
KW
2121# ifdef LC_ALL
2122
948523db
KW
2123 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
2124 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
2125 if (! sl_result[LC_ALL_INDEX]) {
49c85077 2126 setlocale_failure = TRUE;
7cd8b568
KW
2127 }
2128 else {
2129 /* Since LC_ALL succeeded, it should have changed all the other
2130 * categories it can to its value; so we massage things so that the
2131 * setlocales below just return their category's current values.
2132 * This adequately handles the case in NetBSD where LC_COLLATE may
2133 * not be defined for a locale, and setting it individually will
7d4bcc4a 2134 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
2135 * the POSIX locale. */
2136 trial_locale = NULL;
2137 }
7d4bcc4a
KW
2138
2139# endif /* LC_ALL */
98994639 2140
e5f10d49
KW
2141 if (! setlocale_failure) {
2142 unsigned int j;
2143 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2144 curlocales[j]
2145 = savepv(do_setlocale_r(categories[j], trial_locale));
2146 if (! curlocales[j]) {
2147 setlocale_failure = TRUE;
2148 }
2149 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2150 }
c835d6be 2151
e5f10d49
KW
2152 if (! setlocale_failure) { /* All succeeded */
2153 break; /* Exit trial_locales loop */
49c85077 2154 }
65ebb059 2155 }
98994639 2156
49c85077
KW
2157 /* Here, something failed; will need to try a fallback. */
2158 ok = 0;
65ebb059 2159
49c85077
KW
2160 if (i == 0) {
2161 unsigned int j;
98994639 2162
65ebb059 2163 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2164
2165# ifdef LC_ALL
98994639 2166
49c85077
KW
2167 PerlIO_printf(Perl_error_log,
2168 "perl: warning: Setting locale failed.\n");
98994639 2169
7d4bcc4a 2170# else /* !LC_ALL */
98994639 2171
49c85077
KW
2172 PerlIO_printf(Perl_error_log,
2173 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2174
e5f10d49
KW
2175 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2176 if (! curlocales[j]) {
2177 PerlIO_printf(Perl_error_log, category_names[j]);
2178 }
2179 else {
2180 Safefree(curlocales[j]);
2181 }
2182 }
7d4bcc4a 2183
7d4bcc4a 2184# endif /* LC_ALL */
98994639 2185
49c85077
KW
2186 PerlIO_printf(Perl_error_log,
2187 "perl: warning: Please check that your locale settings:\n");
98994639 2188
7d4bcc4a
KW
2189# ifdef __GLIBC__
2190
49c85077
KW
2191 PerlIO_printf(Perl_error_log,
2192 "\tLANGUAGE = %c%s%c,\n",
2193 language ? '"' : '(',
2194 language ? language : "unset",
2195 language ? '"' : ')');
7d4bcc4a 2196# endif
98994639 2197
49c85077
KW
2198 PerlIO_printf(Perl_error_log,
2199 "\tLC_ALL = %c%s%c,\n",
2200 lc_all ? '"' : '(',
2201 lc_all ? lc_all : "unset",
2202 lc_all ? '"' : ')');
98994639 2203
7d4bcc4a
KW
2204# if defined(USE_ENVIRON_ARRAY)
2205
49c85077 2206 {
cd999af9 2207 char **e;
d5e32b93
KW
2208
2209 /* Look through the environment for any variables of the
2210 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2211 * already handled above. These are assumed to be locale
2212 * settings. Output them and their values. */
cd999af9 2213 for (e = environ; *e; e++) {
d5e32b93
KW
2214 const STRLEN prefix_len = sizeof("LC_") - 1;
2215 STRLEN uppers_len;
2216
cd999af9 2217 if ( strBEGINs(*e, "LC_")
c8b388b0 2218 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2219 && (uppers_len = strspn(*e + prefix_len,
2220 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2221 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2222 {
2223 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2224 (int) (prefix_len + uppers_len), *e,
2225 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2226 }
2227 }
49c85077 2228 }
7d4bcc4a
KW
2229
2230# else
2231
49c85077
KW
2232 PerlIO_printf(Perl_error_log,
2233 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2234
2235# endif
98994639 2236
49c85077
KW
2237 PerlIO_printf(Perl_error_log,
2238 "\tLANG = %c%s%c\n",
2239 lang ? '"' : '(',
2240 lang ? lang : "unset",
2241 lang ? '"' : ')');
98994639 2242
49c85077
KW
2243 PerlIO_printf(Perl_error_log,
2244 " are supported and installed on your system.\n");
2245 }
98994639 2246
65ebb059 2247 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2248 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2249 * usually change the upper bound of the loop we are in.
2250 *
2251 * Since the system's default way of setting the locale has not
2252 * found one that works, We use Perl's defined ordering: LC_ALL,
2253 * LANG, and the C locale. We don't try the same locale twice, so
2254 * don't add to the list if already there. (On POSIX systems, the
2255 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2256 * but there's no harm done by doing it explicitly.
2257 *
2258 * Note that this tries the LC_ALL environment variable even on
2259 * systems which have no LC_ALL locale setting. This may or may
2260 * not have been originally intentional, but there's no real need
2261 * to change the behavior. */
65ebb059
KW
2262 if (lc_all) {
2263 for (j = 0; j < trial_locales_count; j++) {
2264 if (strEQ(lc_all, trial_locales[j])) {
2265 goto done_lc_all;
2266 }
2267 }
2268 trial_locales[trial_locales_count++] = lc_all;
2269 }
2270 done_lc_all:
98994639 2271
65ebb059
KW
2272 if (lang) {
2273 for (j = 0; j < trial_locales_count; j++) {
2274 if (strEQ(lang, trial_locales[j])) {
2275 goto done_lang;
2276 }
2277 }
2278 trial_locales[trial_locales_count++] = lang;
2279 }
2280 done_lang:
2281
7d4bcc4a
KW
2282# if defined(WIN32) && defined(LC_ALL)
2283
65ebb059
KW
2284 /* For Windows, we also try the system default locale before "C".
2285 * (If there exists a Windows without LC_ALL we skip this because
2286 * it gets too complicated. For those, the "C" is the next
2287 * fallback possibility). The "" is the same as the 0th element of
2288 * the array, but the code at the loop above knows to treat it
2289 * differently when not the 0th */
2290 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2291
2292# endif
65ebb059
KW
2293
2294 for (j = 0; j < trial_locales_count; j++) {
2295 if (strEQ("C", trial_locales[j])) {
2296 goto done_C;
2297 }
2298 }
2299 trial_locales[trial_locales_count++] = "C";
98994639 2300
65ebb059
KW
2301 done_C: ;
2302 } /* end of first time through the loop */
98994639 2303
7d4bcc4a
KW
2304# ifdef WIN32
2305
65ebb059 2306 next_iteration: ;
7d4bcc4a
KW
2307
2308# endif
65ebb059
KW
2309
2310 } /* end of looping through the trial locales */
2311
2312 if (ok < 1) { /* If we tried to fallback */
2313 const char* msg;
2314 if (! setlocale_failure) { /* fallback succeeded */
2315 msg = "Falling back to";
2316 }
2317 else { /* fallback failed */
e5f10d49 2318 unsigned int j;
98994639 2319
65ebb059
KW
2320 /* We dropped off the end of the loop, so have to decrement i to
2321 * get back to the value the last time through */
2322 i--;
98994639 2323
65ebb059
KW
2324 ok = -1;
2325 msg = "Failed to fall back to";
2326
2327 /* To continue, we should use whatever values we've got */
7d4bcc4a 2328
e5f10d49
KW
2329 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2330 Safefree(curlocales[j]);
2331 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2332 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2333 }
65ebb059
KW
2334 }
2335
2336 if (locwarn) {
2337 const char * description;
2338 const char * name = "";
2339 if (strEQ(trial_locales[i], "C")) {
2340 description = "the standard locale";
2341 name = "C";
2342 }
7d4bcc4a
KW
2343
2344# ifdef SYSTEM_DEFAULT_LOCALE
2345
65ebb059
KW
2346 else if (strEQ(trial_locales[i], "")) {
2347 description = "the system default locale";
2348 if (system_default_locale) {
2349 name = system_default_locale;
2350 }
2351 }
7d4bcc4a
KW
2352
2353# endif /* SYSTEM_DEFAULT_LOCALE */
2354
65ebb059
KW
2355 else {
2356 description = "a fallback locale";
2357 name = trial_locales[i];
2358 }
2359 if (name && strNE(name, "")) {
2360 PerlIO_printf(Perl_error_log,
2361 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2362 }
2363 else {
2364 PerlIO_printf(Perl_error_log,
2365 "perl: warning: %s %s.\n", msg, description);
2366 }
2367 }
2368 } /* End of tried to fallback */
98994639 2369
e5f10d49
KW
2370 /* Done with finding the locales; update our records */
2371
7d4bcc4a
KW
2372# ifdef USE_LOCALE_CTYPE
2373
948523db 2374 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2375
e5f10d49 2376# endif
7d4bcc4a
KW
2377# ifdef USE_LOCALE_COLLATE
2378
948523db 2379 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2380
e5f10d49 2381# endif
7d4bcc4a
KW
2382# ifdef USE_LOCALE_NUMERIC
2383
948523db 2384 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2385
2386# endif
2387
948523db 2388 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20
KW
2389
2390# if defined(USE_ITHREADS)
2391
2392 /* This caches whether each category's locale is UTF-8 or not. This
2393 * may involve changing the locale. It is ok to do this at
2394 * initialization time before any threads have started, but not later.
2395 * Caching means that if the program heeds our dictate not to change
2396 * locales in threaded applications, this data will remain valid, and
2397 * it may get queried without changing locales. If the environment is
2398 * such that all categories have the same locale, this isn't needed, as
2399 * the code will not change the locale; but this handles the uncommon
2400 * case where the environment has disparate locales for the categories
2401 * */
2402 (void) _is_cur_LC_category_utf8(categories[i]);
2403
2404# endif
2405
e5f10d49
KW
2406 Safefree(curlocales[i]);
2407 }
b310b053 2408
7d4bcc4a
KW
2409# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2410
49c85077 2411 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
2412 * locale is UTF-8. The call to new_ctype() just above has already
2413 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
2414 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
2415 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
2416 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
2417 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 2418
a05d7ebb 2419 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2420 This is an alternative to using the -C command line switch
2421 (the -C if present will override this). */
2422 {
dd374669 2423 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2424 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2425 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2426 PL_utf8cache = -1;
b310b053
JH
2427 }
2428
7d4bcc4a 2429# endif
7d4bcc4a
KW
2430# ifdef __GLIBC__
2431
175c4cf9 2432 Safefree(language);
7d4bcc4a
KW
2433
2434# endif
175c4cf9
KW
2435
2436 Safefree(lc_all);
2437 Safefree(lang);
2438
e3305790 2439#endif /* USE_LOCALE */
2fcc0ca9 2440#ifdef DEBUGGING
7d4bcc4a 2441
2fcc0ca9 2442 /* So won't continue to output stuff */
27cdc72e 2443 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2444
2fcc0ca9
KW
2445#endif
2446
98994639
HS
2447 return ok;
2448}
2449
98994639
HS
2450#ifdef USE_LOCALE_COLLATE
2451
a4a439fb 2452char *
a4a439fb
KW
2453Perl__mem_collxfrm(pTHX_ const char *input_string,
2454 STRLEN len, /* Length of 'input_string' */
2455 STRLEN *xlen, /* Set to length of returned string
2456 (not including the collation index
2457 prefix) */
2458 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2459 )
98994639 2460{
a4a439fb
KW
2461
2462 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2463 * differences. First, it handles embedded NULs. Second, it allocates a bit
2464 * more memory than needed for the transformed data itself. The real
55e5378d 2465 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2466 * the length of that, and doesn't include the collation index size.
2467 * Please see sv_collxfrm() to see how this is used. */
2468
55e5378d
KW
2469#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2470
6696cfa7
KW
2471 char * s = (char *) input_string;
2472 STRLEN s_strlen = strlen(input_string);
79f120c8 2473 char *xbuf = NULL;
55e5378d 2474 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2475 STRLEN length_in_chars;
c664130f 2476 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2477
a4a439fb
KW
2478 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2479
2480 /* Must be NUL-terminated */
2481 assert(*(input_string + len) == '\0');
7918f24d 2482
79f120c8
KW
2483 /* If this locale has defective collation, skip */
2484 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2485 DEBUG_L(PerlIO_printf(Perl_debug_log,
2486 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2487 goto bad;
2488 }
2489
6696cfa7
KW
2490 /* Replace any embedded NULs with the control that sorts before any others.
2491 * This will give as good as possible results on strings that don't
2492 * otherwise contain that character, but otherwise there may be
2493 * less-than-perfect results with that character and NUL. This is
fdc080f3 2494 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2495 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2496 NUL */
6696cfa7
KW
2497 char * e = s + len;
2498 char * sans_nuls;
fdc080f3 2499 STRLEN sans_nuls_len;
94762aa0 2500 int try_non_controls;
afc4976f
KW
2501 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2502 making sure 2nd byte is NUL.
2503 */
2504 STRLEN this_replacement_len;
2505
1e4c9676
KW
2506 /* If we don't know what non-NUL control character sorts lowest for
2507 * this locale, find it */
f28f4d2a 2508 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2509 int j;
afc4976f 2510 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2511 includes the collation index
2512 prefixed. */
2513
91c0e2e0 2514 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2515
2516 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2517 * NUL, in which case we instead look for any character. Controls
2518 * are preferred because collation order is, in general, context
2519 * sensitive, with adjoining characters affecting the order, and
2520 * controls are less likely to have such interactions, allowing the
2521 * NUL-replacement to stand on its own. (Another way to look at it
2522 * is to imagine what would happen if the NUL were replaced by a
2523 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2524 for (try_non_controls = 0;
2525 try_non_controls < 2;
2526 try_non_controls++)
2527 {
d4ff9586
KW
2528 /* Look through all legal code points (NUL isn't) */
2529 for (j = 1; j < 256; j++) {
2530 char * x; /* j's xfrm plus collation index */
2531 STRLEN x_len; /* length of 'x' */
2532 STRLEN trial_len = 1;
736a4fed 2533 char cur_source[] = { '\0', '\0' };
d4ff9586 2534
736a4fed
KW
2535 /* Skip non-controls the first time through the loop. The
2536 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2537 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2538 ? ! isCNTRL_L1(j)
2539 : ! isCNTRL_LC(j))
2540 {
d4ff9586 2541 continue;
6696cfa7 2542 }
6696cfa7 2543
736a4fed
KW
2544 /* Create a 1-char string of the current code point */
2545 cur_source[0] = (char) j;
2546
d4ff9586
KW
2547 /* Then transform it */
2548 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2549 0 /* The string is not in UTF-8 */);
6696cfa7 2550
1e4c9676 2551 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2552 * */
2553 if (! x) {
2554 continue;
2555 }
6696cfa7 2556
d4ff9586
KW
2557 /* If this character's transformation is lower than
2558 * the current lowest, this one becomes the lowest */
2559 if ( cur_min_x == NULL
2560 || strLT(x + COLLXFRM_HDR_LEN,
2561 cur_min_x + COLLXFRM_HDR_LEN))
2562 {
f28f4d2a 2563 PL_strxfrm_NUL_replacement = j;
d4ff9586 2564 cur_min_x = x;
d4ff9586
KW
2565 }
2566 else {
2567 Safefree(x);
2568 }
1e4c9676 2569 } /* end of loop through all 255 characters */
6696cfa7 2570
1e4c9676 2571 /* Stop looking if found */
94762aa0
KW
2572 if (cur_min_x) {
2573 break;
2574 }
2575
2576 /* Unlikely, but possible, if there aren't any controls that
2577 * work in the locale, repeat the loop, looking for any
2578 * character that works */
2579 DEBUG_L(PerlIO_printf(Perl_debug_log,
2580 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2581 } /* End of loop to try first the controls, then any char */
6696cfa7 2582
94762aa0
KW
2583 if (! cur_min_x) {
2584 DEBUG_L(PerlIO_printf(Perl_debug_log,
2585 "_mem_collxfrm: Couldn't find any character to replace"
2586 " embedded NULs in locale %s with", PL_collation_name));
2587 goto bad;
58eebef2
KW
2588 }
2589
94762aa0
KW
2590 DEBUG_L(PerlIO_printf(Perl_debug_log,
2591 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2592 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2593
6696cfa7 2594 Safefree(cur_min_x);
1e4c9676 2595 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2596
2597 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2598 * UTF8-ness of the original */
f28f4d2a
KW
2599 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2600 this_replacement_char[0] =
2601 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2602 this_replacement_char[1] =
2603 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2604 this_replacement_len = 2;
2605 }
2606 else {
f28f4d2a 2607 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2608 /* this_replacement_char[1] = '\0' was done at initialization */
2609 this_replacement_len = 1;
6696cfa7
KW
2610 }
2611
2612 /* The worst case length for the replaced string would be if every
2613 * character in it is NUL. Multiply that by the length of each
2614 * replacement, and allow for a trailing NUL */
afc4976f 2615 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2616 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2617 *sans_nuls = '\0';
2618
6696cfa7
KW
2619 /* Replace each NUL with the lowest collating control. Loop until have
2620 * exhausted all the NULs */
2621 while (s + s_strlen < e) {
6069d6c5 2622 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2623
2624 /* Do the actual replacement */
6069d6c5 2625 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2626
2627 /* Move past the input NUL */
2628 s += s_strlen + 1;
2629 s_strlen = strlen(s);
2630 }
2631
2632 /* And add anything that trails the final NUL */
6069d6c5 2633 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2634
2635 /* Switch so below we transform this modified string */
2636 s = sans_nuls;
2637 len = strlen(s);
1e4c9676 2638 } /* End of replacing NULs */
6696cfa7 2639
a4a439fb
KW
2640 /* Make sure the UTF8ness of the string and locale match */
2641 if (utf8 != PL_in_utf8_COLLATE_locale) {
2642 const char * const t = s; /* Temporary so we can later find where the
2643 input was */
2644
2645 /* Here they don't match. Change the string's to be what the locale is
2646 * expecting */
2647
2648 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2649 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2650 utf8 = TRUE;
2651 }
2652 else { /* locale is not UTF-8; but input is; downgrade the input */
2653
2654 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2655
2656 /* If the downgrade was successful we are done, but if the input
2657 * contains things that require UTF-8 to represent, have to do
2658 * damage control ... */
2659 if (UNLIKELY(utf8)) {
2660
2661 /* What we do is construct a non-UTF-8 string with
2662 * 1) the characters representable by a single byte converted
2663 * to be so (if necessary);
2664 * 2) and the rest converted to collate the same as the
2665 * highest collating representable character. That makes
2666 * them collate at the end. This is similar to how we
2667 * handle embedded NULs, but we use the highest collating
2668 * code point instead of the smallest. Like the NUL case,
2669 * this isn't perfect, but is the best we can reasonably
2670 * do. Every above-255 code point will sort the same as
2671 * the highest-sorting 0-255 code point. If that code
2672 * point can combine in a sequence with some other code
2673 * points for weight calculations, us changing something to
2674 * be it can adversely affect the results. But in most
2675 * cases, it should work reasonably. And note that this is
2676 * really an illegal situation: using code points above 255
2677 * on a locale where only 0-255 are valid. If two strings
2678 * sort entirely equal, then the sort order for the
2679 * above-255 code points will be in code point order. */
2680
2681 utf8 = FALSE;
2682
2683 /* If we haven't calculated the code point with the maximum
2684 * collating order for this locale, do so now */
2685 if (! PL_strxfrm_max_cp) {
2686 int j;
2687
2688 /* The current transformed string that collates the
2689 * highest (except it also includes the prefixed collation
2690 * index. */
2691 char * cur_max_x = NULL;
2692
2693 /* Look through all legal code points (NUL isn't) */
2694 for (j = 1; j < 256; j++) {
2695 char * x;
2696 STRLEN x_len;
736a4fed 2697 char cur_source[] = { '\0', '\0' };
a4a439fb 2698
736a4fed
KW
2699 /* Create a 1-char string of the current code point */
2700 cur_source[0] = (char) j;
a4a439fb
KW
2701
2702 /* Then transform it */
2703 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2704
2705 /* If something went wrong (which it shouldn't), just
2706 * ignore this code point */
94762aa0 2707 if (! x) {
a4a439fb
KW
2708 continue;
2709 }
2710
2711 /* If this character's transformation is higher than
2712 * the current highest, this one becomes the highest */
2713 if ( cur_max_x == NULL
55e5378d
KW
2714 || strGT(x + COLLXFRM_HDR_LEN,
2715 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2716 {
2717 PL_strxfrm_max_cp = j;
2718 cur_max_x = x;
2719 }
2720 else {
2721 Safefree(x);
2722 }
2723 }
2724
94762aa0
KW
2725 if (! cur_max_x) {
2726 DEBUG_L(PerlIO_printf(Perl_debug_log,
2727 "_mem_collxfrm: Couldn't find any character to"
2728 " replace above-Latin1 chars in locale %s with",
2729 PL_collation_name));
2730 goto bad;
2731 }
2732
58eebef2
KW
2733 DEBUG_L(PerlIO_printf(Perl_debug_log,
2734 "_mem_collxfrm: highest 1-byte collating character"
2735 " in locale %s is 0x%02X\n",
2736 PL_collation_name,
2737 PL_strxfrm_max_cp));
58eebef2 2738
a4a439fb
KW
2739 Safefree(cur_max_x);
2740 }
2741
2742 /* Here we know which legal code point collates the highest.
2743 * We are ready to construct the non-UTF-8 string. The length
2744 * will be at least 1 byte smaller than the input string
2745 * (because we changed at least one 2-byte character into a
2746 * single byte), but that is eaten up by the trailing NUL */
2747 Newx(s, len, char);
2748
2749 {
2750 STRLEN i;
2751 STRLEN d= 0;
042d9e50 2752 char * e = (char *) t + len;
a4a439fb
KW
2753
2754 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2755 U8 cur_char = t[i];
2756 if (UTF8_IS_INVARIANT(cur_char)) {
2757 s[d++] = cur_char;
2758 }
042d9e50 2759 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2760 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2761 }
2762 else { /* Replace illegal cp with highest collating
2763 one */
2764 s[d++] = PL_strxfrm_max_cp;
2765 }
2766 }
2767 s[d++] = '\0';
2768 Renew(s, d, char); /* Free up unused space */
2769 }
2770 }
2771 }
2772
2773 /* Here, we have constructed a modified version of the input. It could
2774 * be that we already had a modified copy before we did this version.
2775 * If so, that copy is no longer needed */
2776 if (t != input_string) {
2777 Safefree(t);
2778 }
2779 }
2780
17f41037
KW
2781 length_in_chars = (utf8)
2782 ? utf8_length((U8 *) s, (U8 *) s + len)
2783 : len;
2784
59c018b9
KW
2785 /* The first element in the output is the collation id, used by
2786 * sv_collxfrm(); then comes the space for the transformed string. The
2787 * equation should give us a good estimate as to how much is needed */
55e5378d 2788 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2789 + PL_collxfrm_base
17f41037 2790 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2791 Newx(xbuf, xAlloc, char);
c7202dee
KW
2792 if (UNLIKELY(! xbuf)) {
2793 DEBUG_L(PerlIO_printf(Perl_debug_log,
2794 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2795 goto bad;
c7202dee 2796 }
98994639 2797
d35fca5f 2798 /* Store the collation id */
98994639 2799 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2800
2801 /* Then the transformation of the input. We loop until successful, or we
2802 * give up */
4ebeff16 2803 for (;;) {
1adab0a7 2804
55e5378d 2805 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2806
2807 /* If the transformed string occupies less space than we told strxfrm()
2808 * was available, it means it successfully transformed the whole
2809 * string. */
55e5378d 2810 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2811
1adab0a7
KW
2812 /* Some systems include a trailing NUL in the returned length.
2813 * Ignore it, using a loop in case multiple trailing NULs are
2814 * returned. */
2815 while ( (*xlen) > 0
2816 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2817 {
2818 (*xlen)--;
2819 }
2820
17f41037
KW
2821 /* If the first try didn't get it, it means our prediction was low.
2822 * Modify the coefficients so that we predict a larger value in any
2823 * future transformations */
2824 if (! first_time) {
2825 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2826 STRLEN computed_guess = PL_collxfrm_base
2827 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2828
2829 /* On zero-length input, just keep current slope instead of
2830 * dividing by 0 */
2831 const STRLEN new_m = (length_in_chars != 0)
2832 ? needed / length_in_chars
2833 : PL_collxfrm_mult;
17f41037
KW
2834
2835 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2836 "%s: %d: initial size of %zu bytes for a length "
2837 "%zu string was insufficient, %zu needed\n",
17f41037 2838 __FILE__, __LINE__,
b07929e4 2839 computed_guess, length_in_chars, needed));
17f41037
KW
2840
2841 /* If slope increased, use it, but discard this result for
2842 * length 1 strings, as we can't be sure that it's a real slope
2843 * change */
2844 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
2845
2846# ifdef DEBUGGING
2847
17f41037
KW
2848 STRLEN old_m = PL_collxfrm_mult;
2849 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
2850
2851# endif
2852
17f41037
KW
2853 PL_collxfrm_mult = new_m;
2854 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2855 computed_guess = PL_collxfrm_base
2856 + (PL_collxfrm_mult * length_in_chars);
2857 if (computed_guess < needed) {
2858 PL_collxfrm_base += needed - computed_guess;
2859 }
2860
2861 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2862 "%s: %d: slope is now %zu; was %zu, base "
2863 "is now %zu; was %zu\n",
17f41037 2864 __FILE__, __LINE__,
b07929e4
KW
2865 PL_collxfrm_mult, old_m,
2866 PL_collxfrm_base, old_b));
17f41037
KW
2867 }
2868 else { /* Slope didn't change, but 'b' did */
2869 const STRLEN new_b = needed
2870 - computed_guess
2871 + PL_collxfrm_base;
2872 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2873 "%s: %d: base is now %zu; was %zu\n",
17f41037 2874 __FILE__, __LINE__,
b07929e4 2875 new_b, PL_collxfrm_base));
17f41037
KW
2876 PL_collxfrm_base = new_b;
2877 }
2878 }
2879
4ebeff16
KW
2880 break;
2881 }
bb0f664e 2882
c7202dee
KW
2883 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2884 DEBUG_L(PerlIO_printf(Perl_debug_log,
2885 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2886 *xlen, PERL_INT_MAX));
4ebeff16 2887 goto bad;
c7202dee 2888 }
d35fca5f 2889
c664130f 2890 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
2891 * (usually not including the trailing NUL) when it fails due to not
2892 * enough space being provided. Assume that this is the case unless
2893 * it's been proven otherwise */
c664130f 2894 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 2895 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
2896 }
2897 else { /* Here, either:
2898 * 1) The strxfrm() has previously shown bad behavior; or
2899 * 2) It isn't the first time through the loop, which means
2900 * that the strxfrm() is now showing bad behavior, because
2901 * we gave it what it said was needed in the previous
2902 * iteration, and it came back saying it needed still more.
2903 * (Many versions of cygwin fit this. When the buffer size
2904 * isn't sufficient, they return the input size instead of
2905 * how much is needed.)
d4ff9586
KW
2906 * Increase the buffer size by a fixed percentage and try again.
2907 * */
6ddd902c 2908 xAlloc += (xAlloc / 4) + 1;
c664130f 2909 PL_strxfrm_is_behaved = FALSE;
c664130f 2910
7d4bcc4a
KW
2911# ifdef DEBUGGING
2912
58eebef2
KW
2913 if (DEBUG_Lv_TEST || debug_initialization) {
2914 PerlIO_printf(Perl_debug_log,
2915 "_mem_collxfrm required more space than previously calculated"
b07929e4 2916 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 2917 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 2918 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 2919 }
7d4bcc4a
KW
2920
2921# endif
2922
58eebef2 2923 }
c664130f 2924
4ebeff16 2925 Renew(xbuf, xAlloc, char);
c7202dee
KW
2926 if (UNLIKELY(! xbuf)) {
2927 DEBUG_L(PerlIO_printf(Perl_debug_log,
2928 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 2929 goto bad;
c7202dee 2930 }
c664130f
KW
2931
2932 first_time = FALSE;
4ebeff16 2933 }
98994639 2934
6696cfa7 2935
7d4bcc4a
KW
2936# ifdef DEBUGGING
2937
58eebef2 2938 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
2939
2940 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
2941 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
2942 PerlIO_printf(Perl_debug_log, "%s\n",
2943 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
2944 *xlen, 1));
58eebef2 2945 }
7d4bcc4a
KW
2946
2947# endif
58eebef2 2948
3c5f993e 2949 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 2950 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 2951
6696cfa7
KW
2952 if (s != input_string) {
2953 Safefree(s);
98994639
HS
2954 }
2955
98994639
HS
2956 return xbuf;
2957
2958 bad:
2959 Safefree(xbuf);
6696cfa7
KW
2960 if (s != input_string) {
2961 Safefree(s);
2962 }
98994639 2963 *xlen = 0;
7d4bcc4a
KW
2964
2965# ifdef DEBUGGING
2966
58eebef2 2967 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 2968 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 2969 }
7d4bcc4a
KW
2970
2971# endif
2972
98994639
HS
2973 return NULL;
2974}
2975
7d4bcc4a 2976# ifdef DEBUGGING
c7202dee 2977
4cbaac56 2978STATIC void
c7202dee
KW
2979S_print_collxfrm_input_and_return(pTHX_
2980 const char * const s,
2981 const char * const e,
2982 const STRLEN * const xlen,
2983 const bool is_utf8)
2984{
c7202dee
KW
2985
2986 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
2987
511e4ff7
DM
2988 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
2989 (UV)PL_collation_ix);
c7202dee 2990 if (xlen) {
08b6dc1d 2991 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
2992 }
2993 else {
2994 PerlIO_printf(Perl_debug_log, "NULL");
2995 }
2996 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
2997 PL_collation_name);
9c8a6dc2
KW
2998 print_bytes_for_locale(s, e, is_utf8);
2999
3000 PerlIO_printf(Perl_debug_log, "'\n");
3001}
3002
3003STATIC void
3004S_print_bytes_for_locale(pTHX_
3005 const char * const s,
3006 const char * const e,
3007 const bool is_utf8)
3008{
3009 const char * t = s;
3010 bool prev_was_printable = TRUE;
3011 bool first_time = TRUE;
3012
3013 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
3014
3015 while (t < e) {
3016 UV cp = (is_utf8)
3017 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
3018 : * (U8 *) t;
3019 if (isPRINT(cp)) {
3020 if (! prev_was_printable) {
3021 PerlIO_printf(Perl_debug_log, " ");
3022 }
3023 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
3024 prev_was_printable = TRUE;
3025 }
3026 else {
3027 if (! first_time) {
3028 PerlIO_printf(Perl_debug_log, " ");
3029 }
147e3846 3030 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
3031 prev_was_printable = FALSE;
3032 }
3033 t += (is_utf8) ? UTF8SKIP(t) : 1;
3034 first_time = FALSE;
3035 }
c7202dee
KW
3036}
3037
7d4bcc4a 3038# endif /* #ifdef DEBUGGING */
98994639 3039#endif /* USE_LOCALE_COLLATE */
58eebef2 3040
8ef6e574
KW
3041#ifdef USE_LOCALE
3042
c1284011
KW
3043bool
3044Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
3045{
3046 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
3047 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 3048 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
3049 * could give the wrong result. The result will very likely be correct for
3050 * languages that have commonly used non-ASCII characters, but for notably
3051 * English, it comes down to if the locale's name ends in something like
3052 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61 3053
47280b20 3054 /* Name of current locale corresponding to the input category */
8de4332b 3055 const char *save_input_locale = NULL;
47280b20
KW
3056
3057 bool is_utf8 = FALSE; /* The return value */
7d74bb61
KW
3058 STRLEN final_pos;
3059
47280b20
KW
3060 /* The variables below are for the cache of previous lookups using this
3061 * function. The cache is a C string, described at the definition for
3062 * 'C_and_POSIX_utf8ness'.
3063 *
3064 * The first part of the cache is fixed, for the C and POSIX locales. The
3065 * varying part starts just after them. */
3066 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
3067
3068 Size_t utf8ness_cache_size; /* Size of the varying portion */
3069 Size_t input_name_len; /* Length in bytes of save_input_locale */
3070 Size_t input_name_len_with_overhead; /* plus extra chars used to store
3071 the name in the cache */
3072 char * delimited; /* The name plus the delimiters used to store
3073 it in the cache */
3074 char * name_pos; /* position of 'delimited' in the cache, or 0
3075 if not there */
3076
3077
7d4bcc4a
KW
3078# ifdef LC_ALL
3079
7d74bb61 3080 assert(category != LC_ALL);
7d4bcc4a
KW
3081
3082# endif
7d74bb61 3083
47280b20 3084 /* Get the desired category's locale */
837ce802 3085 save_input_locale = do_setlocale_r(category, NULL);
7d74bb61 3086 if (! save_input_locale) {
47280b20
KW
3087 Perl_croak(aTHX_
3088 "panic: %s: %d: Could not find current locale for %s\n",
3089 __FILE__, __LINE__, category_name(category));
7d74bb61 3090 }
47280b20 3091
b07fffd1 3092 save_input_locale = stdize_locale(savepv(save_input_locale));
47280b20
KW
3093 DEBUG_L(PerlIO_printf(Perl_debug_log,
3094 "Current locale for %s is %s\n",
3095 category_name(category), save_input_locale));
3096
3097 input_name_len = strlen(save_input_locale);
3098
3099 /* In our cache, each name is accompanied by two delimiters and a single
3100 * utf8ness digit */
3101 input_name_len_with_overhead = input_name_len + 3;
3102
3103 /* Allocate and populate space for a copy of the name surrounded by the
3104 * delimiters */
3105 Newx(delimited, input_name_len_with_overhead, char);
3106 delimited[0] = UTF8NESS_SEP[0];
3107 Copy(save_input_locale, delimited + 1, input_name_len, char);
3108 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
3109 delimited[input_name_len+2] = '\0';
3110
3111 /* And see if that is in the cache */
3112 name_pos = instr(PL_locale_utf8ness, delimited);
3113 if (name_pos) {
3114 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
3115
3116# ifdef DEBUGGING
3117
3118 if (DEBUG_Lv_TEST || debug_initialization) {
3119 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
3120 save_input_locale, is_utf8);
3121 }
3122
3123# endif
3124
3125 /* And, if not already in that position, move it to the beginning of
3126 * the non-constant portion of the list, since it is the most recently
3127 * used. (We don't have to worry about overflow, since just moving
3128 * existing names around) */
3129 if (name_pos > utf8ness_cache) {
3130 Move(utf8ness_cache,
3131 utf8ness_cache + input_name_len_with_overhead,
3132 name_pos - utf8ness_cache, char);
3133 Copy(delimited,
3134 utf8ness_cache,
3135 input_name_len_with_overhead - 1, char);
3136 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3137 }
3138
3139 Safefree(delimited);
b07fffd1 3140 Safefree(save_input_locale);
47280b20 3141 return is_utf8;
7d74bb61
KW
3142 }
3143
47280b20
KW
3144 /* Here we don't have stored the utf8ness for the input locale. We have to
3145 * calculate it */
3146
7d4bcc4a 3147# if defined(USE_LOCALE_CTYPE) \
1d958db2 3148 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 3149
1d958db2 3150 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
3151
3152 char *save_ctype_locale = NULL;
7d74bb61 3153
119ee68b 3154 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
3155
3156 /* Get the current LC_CTYPE locale */
837ce802 3157 save_ctype_locale = do_setlocale_c(LC_CTYPE, NULL);
7d74bb61 3158 if (! save_ctype_locale) {
69014004
KW
3159 DEBUG_L(PerlIO_printf(Perl_debug_log,
3160 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
3161 goto cant_use_nllanginfo;
3162 }
4f72bb37 3163 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
3164
3165 /* If LC_CTYPE and the desired category use the same locale, this
3166 * means that finding the value for LC_CTYPE is the same as finding
3167 * the value for the desired category. Otherwise, switch LC_CTYPE
3168 * to the desired category's locale */
3169 if (strEQ(save_ctype_locale, save_input_locale)) {
3170 Safefree(save_ctype_locale);
3171 save_ctype_locale = NULL;
3172 }
837ce802 3173 else if (! do_setlocale_c(LC_CTYPE, save_input_locale)) {
69014004
KW
3174 DEBUG_L(PerlIO_printf(Perl_debug_log,
3175 "Could not change LC_CTYPE locale to %s\n",
3176 save_input_locale));
7d74bb61
KW
3177 Safefree(save_ctype_locale);
3178 goto cant_use_nllanginfo;
3179 }
3180 }
3181
69014004
KW
3182 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
3183 save_input_locale));
3184
7d74bb61 3185 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
3186 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
3187 * should give the correct results */
119ee68b 3188
7d4bcc4a
KW
3189# if defined(HAS_NL_LANGINFO) && defined(CODESET)
3190
291a84fb 3191 { /* The task is easiest if the platform has this POSIX 2001 function */
c70a3e68
KW
3192 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
3193 /* FALSE => already in dest locale */
119ee68b 3194
c70a3e68
KW
3195 DEBUG_L(PerlIO_printf(Perl_debug_log,
3196 "\tnllanginfo returned CODESET '%s'\n", codeset));
3197
3198 if (codeset && strNE(codeset, "")) {
1d958db2
KW
3199 /* If we switched LC_CTYPE, switch back */
3200 if (save_ctype_locale) {
837ce802 3201 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2
KW
3202 Safefree(save_ctype_locale);
3203 }
3204
1d075173
KW
3205 /* If the implementation of foldEQ() somehow were
3206 * to change to not go byte-by-byte, this could
3207 * read past end of string, as only one length is
3208 * checked. But currently, a premature NUL will
3209 * compare false, and it will stop there */
3210 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
3211 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
1d958db2 3212
69014004
KW
3213 DEBUG_L(PerlIO_printf(Perl_debug_log,
3214 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
3215 codeset, is_utf8));
47280b20 3216 goto finish_and_return;
1d958db2 3217 }
119ee68b
KW
3218 }
3219
7d4bcc4a
KW
3220# endif
3221# ifdef MB_CUR_MAX
1d958db2
KW
3222
3223 /* Here, either we don't have nl_langinfo, or it didn't return a
3224 * codeset. Try MB_CUR_MAX */
3225
119ee68b
KW
3226 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3227 * Unicode code point. Since UTF-8 is the only non-single byte
3228 * encoding we handle, we just say any such encoding is UTF-8, and if
3229 * turns out to be wrong, other things will fail */
5f7616bd 3230 is_utf8 = (unsigned) MB_CUR_MAX >= STRLENs(MAX_UNICODE_UTF8);
119ee68b 3231
69014004
KW
3232 DEBUG_L(PerlIO_printf(Perl_debug_log,
3233 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
3234 (int) MB_CUR_MAX, is_utf8));
3235
119ee68b
KW
3236 Safefree(save_input_locale);
3237
7d4bcc4a 3238# ifdef HAS_MBTOWC
119ee68b
KW
3239
3240 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
3241 * since they are both in the C99 standard. We can feed a known byte
3242 * string to the latter function, and check that it gives the expected
3243 * result */
3244 if (is_utf8) {
3245 wchar_t wc;
51fc4b19
KW
3246 int len;
3247
856b881c 3248 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 3249 errno = 0;
b1d4925c 3250 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
51fc4b19 3251
b1d4925c
KW
3252
3253 if ( len != STRLENs(REPLACEMENT_CHARACTER_UTF8)
5579633c 3254 || wc != (wchar_t) UNICODE_REPLACEMENT)
119ee68b
KW
3255 {
3256 is_utf8 = FALSE;
b1d4925c
KW
3257 DEBUG_L(PerlIO_printf(Perl_debug_log, "\replacement=U+%x\n",
3258 (unsigned int)wc));
69014004
KW
3259 DEBUG_L(PerlIO_printf(Perl_debug_log,
3260 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
51fc4b19 3261 len, errno));
119ee68b
KW
3262 }
3263 }
7d4bcc4a
KW
3264
3265# endif
119ee68b 3266
1d958db2
KW
3267 /* If we switched LC_CTYPE, switch back */
3268 if (save_ctype_locale) {
837ce802 3269 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2 3270 Safefree(save_ctype_locale);
119ee68b 3271 }
7d74bb61 3272
47280b20 3273 goto finish_and_return;
7d4bcc4a
KW
3274
3275# endif
3276
7d74bb61 3277 }
119ee68b 3278
7d74bb61
KW
3279 cant_use_nllanginfo:
3280
7d4bcc4a 3281# else /* nl_langinfo should work if available, so don't bother compiling this
0080c90a
KW
3282 fallback code. The final fallback of looking at the name is
3283 compiled, and will be executed if nl_langinfo fails */
7d74bb61 3284
97f4de96
KW
3285 /* nl_langinfo not available or failed somehow. Next try looking at the
3286 * currency symbol to see if it disambiguates things. Often that will be
3287 * in the native script, and if the symbol isn't in UTF-8, we know that the
3288 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
3289 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
3290 * */
fa9b773e 3291
7d4bcc4a
KW
3292# ifdef HAS_LOCALECONV
3293# ifdef USE_LOCALE_MONETARY
3294
fa9b773e
KW
3295 {
3296 char *save_monetary_locale = NULL;
fa9b773e 3297 bool only_ascii = FALSE;
13542a67 3298 struct lconv* lc;
fa9b773e 3299
97f4de96
KW
3300 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
3301 * the desired category, if it isn't that locale already */
3302
fa9b773e
KW
3303 if (category != LC_MONETARY) {
3304
837ce802 3305 save_monetary_locale = do_setlocale_c(LC_MONETARY, NULL);
fa9b773e 3306 if (! save_monetary_locale) {
69014004
KW
3307 DEBUG_L(PerlIO_printf(Perl_debug_log,
3308 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
3309 goto cant_use_monetary;
3310 }
4f72bb37 3311 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 3312
13542a67
KW
3313 if (strEQ(save_monetary_locale, save_input_locale)) {
3314 Safefree(save_monetary_locale);
3315 save_monetary_locale = NULL;
3316 }
837ce802 3317 else if (! do_setlocale_c(LC_MONETARY, save_input_locale)) {
59c234b4
KW
3318 DEBUG_L(PerlIO_printf(Perl_debug_log,
3319 "Could not change LC_MONETARY locale to %s\n",
3320 save_input_locale));
3321 Safefree(save_monetary_locale);
3322 goto cant_use_monetary;
fa9b773e
KW
3323 }
3324 }
3325
3326 /* Here the current LC_MONETARY is set to the locale of the category
3327 * whose information is desired. */
3328
13542a67
KW
3329 lc = localeconv();
3330 if (! lc
3331 || ! lc->currency_symbol
c5f058df 3332 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
3333 {
3334 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));
3335 only_ascii = TRUE;
3336 }
3337 else {
3338 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
3339 }
3340
3341 /* If we changed it, restore LC_MONETARY to its original locale */
3342 if (save_monetary_locale) {
837ce802 3343 do_setlocale_c(LC_MONETARY, save_monetary_locale);
fa9b773e
KW
3344 Safefree(save_monetary_locale);
3345 }
3346
13542a67 3347 if (! only_ascii) {
fa9b773e 3348
59c234b4
KW
3349 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3350 * otherwise assume the locale is UTF-8 if and only if the symbol
3351 * is non-ascii UTF-8. */
3352 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
3353 save_input_locale, is_utf8));
47280b20 3354 goto finish_and_return;
13542a67 3355 }
fa9b773e
KW
3356 }
3357 cant_use_monetary:
3358
7d4bcc4a
KW
3359# endif /* USE_LOCALE_MONETARY */
3360# endif /* HAS_LOCALECONV */
fa9b773e 3361
7d4bcc4a 3362# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e
KW
3363
3364/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3365 * the names of the months and weekdays, timezone, and am/pm indicator */
3366 {
3367 char *save_time_locale = NULL;
3368 int hour = 10;
3369 bool is_dst = FALSE;
3370 int dom = 1;
3371 int month = 0;
3372 int i;
3373 char * formatted_time;
3374
3375
3376 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
3377 * desired category, if it isn't that locale already */
3378
3379 if (category != LC_TIME) {
3380
837ce802 3381 save_time_locale = do_setlocale_c(LC_TIME, NULL);
15f7e74e
KW
3382 if (! save_time_locale) {
3383 DEBUG_L(PerlIO_printf(Perl_debug_log,
3384 "Could not find current locale for LC_TIME\n"));
3385 goto cant_use_time;
3386 }
3387 save_time_locale = stdize_locale(savepv(save_time_locale));
3388
3389 if (strEQ(save_time_locale, save_input_locale)) {
3390 Safefree(save_time_locale);
3391 save_time_locale = NULL;
3392 }
837ce802 3393 else if (! do_setlocale_c(LC_TIME, save_input_locale)) {
15f7e74e
KW
3394 DEBUG_L(PerlIO_printf(Perl_debug_log,
3395 "Could not change LC_TIME locale to %s\n",
3396 save_input_locale));
3397 Safefree(save_time_locale);
3398 goto cant_use_time;
3399 }
3400 }
3401
3402 /* Here the current LC_TIME is set to the locale of the category
3403 * whose information is desired. Look at all the days of the week and
9f10db87 3404 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
3405 * characters. The first such a one found will tell us if the locale
3406 * is UTF-8 or not */
3407
3408 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3409 formatted_time = my_strftime("%A %B %Z %p",
3ae5cd07 3410 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
c5f058df
KW
3411 if ( ! formatted_time
3412 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3413 {
15f7e74e
KW
3414
3415 /* Here, we didn't find a non-ASCII. Try the next time through
3416 * with the complemented dst and am/pm, and try with the next
3417 * weekday. After we have gotten all weekdays, try the next
3418 * month */
3419 is_dst = ! is_dst;
3420 hour = (hour + 12) % 24;
3421 dom++;
3422 if (i > 6) {
3423 month++;
3424 }
3425 continue;
3426 }
3427
3428 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3429 * false otherwise. But first, restore LC_TIME to its original
3430 * locale if we changed it */
3431 if (save_time_locale) {
837ce802 3432 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3433 Safefree(save_time_locale);
3434 }
3435
3436 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3437 save_input_locale,
3438 is_utf8_string((U8 *) formatted_time, 0)));
47280b20
KW
3439 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
3440 goto finish_and_return;
15f7e74e
KW
3441 }
3442
3443 /* Falling off the end of the loop indicates all the names were just
3444 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3445 * to its original locale */
3446 if (save_time_locale) {
837ce802 3447 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3448 Safefree(save_time_locale);
3449 }
3450 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));
3451 }
3452 cant_use_time:
3453
7d4bcc4a 3454# endif
15f7e74e 3455
7d4bcc4a 3456# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
3457
3458/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
3459 * on our dromedary test machine, which has over 700 locales. There, this
3460 * added no value to looking at the currency symbol and the time strings. I
3461 * left it in so as to avoid rewriting it if real-world experience indicates
3462 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
3463 * haven't found illegal utf8, we continue on and examine all the strerror()
3464 * messages on the platform for utf8ness. If all are ASCII, we still don't
3465 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
3466 * utf8ness. The reason this doesn't help much is that the messages may not
3467 * have been translated into the locale. The currency symbol and time strings
3468 * are much more likely to have been translated. */
3469 {
855aeb93 3470 int e;
5857e934 3471 bool non_ascii = FALSE;
855aeb93 3472 char *save_messages_locale = NULL;
5857e934 3473 const char * errmsg = NULL;
855aeb93 3474
5857e934
KW
3475 /* Like above, we set LC_MESSAGES to the locale of the desired
3476 * category, if it isn't that locale already */
855aeb93
JH
3477
3478 if (category != LC_MESSAGES) {
3479
837ce802 3480 save_messages_locale = do_setlocale_c(LC_MESSAGES, NULL);
855aeb93 3481 if (! save_messages_locale) {
5857e934
KW
3482 DEBUG_L(PerlIO_printf(Perl_debug_log,
3483 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
3484 goto cant_use_messages;
3485 }
5857e934 3486 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
3487
3488 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
3489 Safefree(save_messages_locale);
3490 save_messages_locale = NULL;
855aeb93 3491 }
837ce802 3492 else if (! do_setlocale_c(LC_MESSAGES, save_input_locale)) {
5857e934
KW
3493 DEBUG_L(PerlIO_printf(Perl_debug_log,
3494 "Could not change LC_MESSAGES locale to %s\n",
3495 save_input_locale));
855aeb93
JH
3496 Safefree(save_messages_locale);
3497 goto cant_use_messages;
3498 }
3499 }
3500
3501 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
3502 * whose information is desired. Look through all the messages. We
3503 * can't use Strerror() here because it may expand to code that
3504 * segfaults in miniperl */
855aeb93 3505
5857e934
KW
3506 for (e = 0; e <= sys_nerr; e++) {
3507 errno = 0;
3508 errmsg = sys_errlist[e];
3509 if (errno || !errmsg) {
855aeb93
JH
3510 break;
3511 }
5857e934 3512 errmsg = savepv(errmsg);
c5f058df 3513 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
3514 non_ascii = TRUE;
3515 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3516 break;
855aeb93
JH
3517 }
3518 }
5857e934 3519 Safefree(errmsg);
855aeb93
JH
3520
3521 /* And, if we changed it, restore LC_MESSAGES to its original locale */
3522 if (save_messages_locale) {
837ce802 3523 do_setlocale_c(LC_MESSAGES, save_messages_locale);
855aeb93
JH
3524 Safefree(save_messages_locale);
3525 }
3526
5857e934
KW
3527 if (non_ascii) {
3528
3529 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3530 * any non-ascii means it is one; otherwise we assume it isn't */
3531 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3532 save_input_locale,
3533 is_utf8));
47280b20 3534 goto finish_and_return;
5857e934 3535 }
855aeb93 3536
5857e934 3537 DEBUG_L(PerlIO_printf(Perl_debug_log, "All error messages for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
855aeb93
JH
3538 }
3539 cant_use_messages:
3540
7d4bcc4a
KW
3541# endif
3542# endif /* the code that is compiled when no nl_langinfo */
0080c90a 3543
7d4bcc4a 3544# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3545 UTF-8 locale */
7d4bcc4a 3546
97f4de96
KW
3547 /* As a last resort, look at the locale name to see if it matches
3548 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3549 * return of setlocale(), is actually defined to be opaque, so we can't
3550 * really rely on the absence of various substrings in the name to indicate
3551 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3552 * be a UTF-8 locale. Similarly for the other common names */
3553
3554 final_pos = strlen(save_input_locale) - 1;
3555 if (final_pos >= 3) {
8de4332b 3556 const char *name = save_input_locale;
97f4de96
KW
3557
3558 /* Find next 'U' or 'u' and look from there */
3559 while ((name += strcspn(name, "Uu") + 1)
3560 <= save_input_locale + final_pos - 2)
3561 {
c72493e0 3562 if ( isALPHA_FOLD_NE(*name, 't')
305b8651 3563 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
3564 {
3565 continue;
3566 }
3567 name += 2;
3568 if (*(name) == '-') {
3569 if ((name > save_input_locale + final_pos - 1)) {
3570 break;
3571 }
3572 name++;
3573 }
3574 if (*(name) == '8') {
97f4de96
KW
3575 DEBUG_L(PerlIO_printf(Perl_debug_log,
3576 "Locale %s ends with UTF-8 in name\n",
3577 save_input_locale));
47280b20
KW
3578 is_utf8 = TRUE;
3579 goto finish_and_return;
97f4de96
KW
3580 }
3581 }
3582 DEBUG_L(PerlIO_printf(Perl_debug_log,
3583 "Locale %s doesn't end with UTF-8 in name\n",
3584 save_input_locale));
3585 }
3586
4d8d465a 3587# ifdef WIN32
7d4bcc4a 3588
97f4de96 3589 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
8a0832a1 3590 if (memENDs(save_input_locale, final_pos, "65001")) {
97f4de96 3591 DEBUG_L(PerlIO_printf(Perl_debug_log,
8a0832a1 3592 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
97f4de96 3593 save_input_locale));
47280b20
KW
3594 is_utf8 = TRUE;
3595 goto finish_and_return;
97f4de96 3596 }
7d4bcc4a 3597
4d8d465a
KW
3598# endif
3599# endif
97f4de96
KW
3600
3601 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3602 * since we are about to return FALSE anyway, there is no point in doing
3603 * this extra work */
7d4bcc4a
KW
3604
3605# if 0
97f4de96
KW
3606 if (instr(save_input_locale, "8859")) {
3607 DEBUG_L(PerlIO_printf(Perl_debug_log,
3608 "Locale %s has 8859 in name, not UTF-8 locale\n",
3609 save_input_locale));
47280b20
KW
3610 is_utf8 = FALSE;
3611 goto finish_and_return;
97f4de96 3612 }
7d4bcc4a 3613# endif
97f4de96 3614
69014004
KW
3615 DEBUG_L(PerlIO_printf(Perl_debug_log,
3616 "Assuming locale %s is not a UTF-8 locale\n",
3617 save_input_locale));
47280b20
KW
3618 is_utf8 = FALSE;
3619
3620 finish_and_return:
3621
3622 /* Cache this result so we don't have to go through all this next time. */
3623 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
3624 - (utf8ness_cache - PL_locale_utf8ness);
3625
3626 /* But we can't save it if it is too large for the total space available */
3627 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
3628 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
3629
3630 /* Here it can fit, but we may need to clear out the oldest cached
3631 * result(s) to do so. Check */
3632 if (utf8ness_cache_len + input_name_len_with_overhead
3633 >= utf8ness_cache_size)
3634 {
3635 /* Here we have to clear something out to make room for this.
3636 * Start looking at the rightmost place where it could fit and find
3637 * the beginning of the entry that extends past that. */
3638 char * cutoff = (char *) my_memrchr(utf8ness_cache,
3639 UTF8NESS_SEP[0],
3640 utf8ness_cache_size
3641 - input_name_len_with_overhead);
3642
3643 assert(cutoff);
3644 assert(cutoff >= utf8ness_cache);
3645
3646 /* This and all subsequent entries must be removed */
3647 *cutoff = '\0';
3648 utf8ness_cache_len = strlen(utf8ness_cache);
3649 }
3650
3651 /* Make space for the new entry */
3652 Move(utf8ness_cache,
3653 utf8ness_cache + input_name_len_with_overhead,
3654 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
3655
3656 /* And insert it */
3657 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
3658 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3659
3660 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
3661 & (PERL_UINTMAX_T) ~1) != '0')
3662 {
3663 Perl_croak(aTHX_
3664 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%u,"
3665 " inserted_name=%s, its_len=%u\n",
3666 __FILE__, __LINE__,
3667 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
3668 delimited, input_name_len_with_overhead);
3669 }
3670 }
3671
3672# ifdef DEBUGGING
3673
3674 if (DEBUG_Lv_TEST || debug_initialization) {
3675 PerlIO_printf(Perl_debug_log,
3676 "PL_locale_utf8ness is now %s; returning %d\n",
3677 PL_locale_utf8ness, is_utf8);
3678 }
3679
3680# endif
3681
3682 Safefree(delimited);
fa9b773e 3683 Safefree(save_input_locale);
47280b20 3684 return is_utf8;
7d74bb61
KW
3685}
3686
8ef6e574 3687#endif
7d74bb61 3688
d6ded950
KW
3689
3690bool
3691Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3692{
1a4f13e1 3693 dVAR;
d6ded950
KW
3694 /* Internal function which returns if we are in the scope of a pragma that
3695 * enables the locale category 'category'. 'compiling' should indicate if
3696 * this is during the compilation phase (TRUE) or not (FALSE). */
3697
3698 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
3699
3700 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
3701 if (! categories || categories == &PL_sv_placeholder) {
3702 return FALSE;
3703 }
3704
3705 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
3706 * a valid unsigned */
3707 assert(category >= -1);
3708 return cBOOL(SvUV(categories) & (1U << (category + 1)));
3709}
3710
2c6ee1a7 3711char *
6ebbc862
KW
3712Perl_my_strerror(pTHX_ const int errnum)
3713{
3714 /* Returns a mortalized copy of the text of the error message associated
3715 * with 'errnum'. It uses the current locale's text unless the platform
3716 * doesn't have the LC_MESSAGES category or we are not being called from
3717 * within the scope of 'use locale'. In the former case, it uses whatever
3718 * strerror returns; in the latter case it uses the text from the C locale.
3719 *
3720 * The function just calls strerror(), but temporarily switches, if needed,
3721 * to the C locale */
3722
3723 char *errstr;
52770946 3724 dVAR;
6ebbc862 3725
52770946 3726#ifndef USE_LOCALE_MESSAGES
6ebbc862 3727
52770946
KW