This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Check for anomalies in UTF-8 locales
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
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
0dec74cd
KW
44#ifdef I_WCHAR
45# include <wchar.h>
46#endif
47
2fcc0ca9
KW
48/* If the environment says to, we can output debugging information during
49 * initialization. This is done before option parsing, and before any thread
50 * creation, so can be a file-level static */
5a4b0634
KW
51#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
52# define debug_initialization 0
53# define DEBUG_INITIALIZATION_set(v)
54#else
2fcc0ca9 55static bool debug_initialization = FALSE;
5a4b0634 56# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
57#endif
58
291a84fb
KW
59/* strlen() of a literal string constant. We might want this more general,
60 * but using it in just this file for now. A problem with more generality is
61 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
62#define STRLENs(s) (sizeof("" s "") - 1)
63
63e5b0d7
KW
64/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
65 * return of setlocale(), then this is extremely likely to be the C or POSIX
66 * locale. However, the output of setlocale() is documented to be opaque, but
67 * the odds are extremely small that it would return these two strings for some
68 * other locale. Note that VMS in these two locales includes many non-ASCII
69 * characters as controls and punctuation (below are hex bytes):
70 * cntrl: 84-97 9B-9F
71 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
72 * Oddly, none there are listed as alphas, though some represent alphabetics
73 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
74#define isNAME_C_OR_POSIX(name) \
75 ( (name) != NULL \
76 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
77 || strEQ((name), "POSIX")))
78
8ef6e574
KW
79#ifdef USE_LOCALE
80
47280b20
KW
81/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
82 * looked up. This is in the form of a C string: */
83
84#define UTF8NESS_SEP "\v"
85#define UTF8NESS_PREFIX "\f"
86
87/* So, the string looks like:
98994639 88 *
47280b20 89 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 90 *
47280b20
KW
91 * where the digit 0 after the \a indicates that the locale starting just
92 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
93
94STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
95STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
96
97#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
98 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
99
100/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
101 * kept there always. The remining portion of the cache is LRU, with the
102 * oldest looked-up locale at the tail end */
103
98994639
HS
104STATIC char *
105S_stdize_locale(pTHX_ char *locs)
106{
47280b20
KW
107 /* Standardize the locale name from a string returned by 'setlocale',
108 * possibly modifying that string.
109 *
110 * The typical return value of setlocale() is either
111 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
112 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
113 * (the space-separated values represent the various sublocales,
114 * in some unspecified order). This is not handled by this function.
115 *
116 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
117 * which is harmful for further use of the string in setlocale(). This
118 * function removes the trailing new line and everything up through the '='
119 * */
120
7452cf6a 121 const char * const s = strchr(locs, '=');
98994639
HS
122 bool okay = TRUE;
123
7918f24d
NC
124 PERL_ARGS_ASSERT_STDIZE_LOCALE;
125
8772537c
AL
126 if (s) {
127 const char * const t = strchr(s, '.');
98994639 128 okay = FALSE;
8772537c
AL
129 if (t) {
130 const char * const u = strchr(t, '\n');
131 if (u && (u[1] == 0)) {
132 const STRLEN len = u - s;
133 Move(s + 1, locs, len, char);
134 locs[len] = 0;
135 okay = TRUE;
98994639
HS
136 }
137 }
138 }
139
140 if (!okay)
141 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
142
143 return locs;
144}
145
e5f10d49
KW
146/* Two parallel arrays; first the locale categories Perl uses on this system;
147 * the second array is their names. These arrays are in mostly arbitrary
148 * order. */
149
150const int categories[] = {
151
152# ifdef USE_LOCALE_NUMERIC
153 LC_NUMERIC,
154# endif
155# ifdef USE_LOCALE_CTYPE
156 LC_CTYPE,
157# endif
158# ifdef USE_LOCALE_COLLATE
159 LC_COLLATE,
160# endif
161# ifdef USE_LOCALE_TIME
162 LC_TIME,
163# endif
164# ifdef USE_LOCALE_MESSAGES
165 LC_MESSAGES,
166# endif
167# ifdef USE_LOCALE_MONETARY
168 LC_MONETARY,
169# endif
9821811f
KW
170# ifdef USE_LOCALE_ADDRESS
171 LC_ADDRESS,
172# endif
173# ifdef USE_LOCALE_IDENTIFICATION
174 LC_IDENTIFICATION,
175# endif
176# ifdef USE_LOCALE_MEASUREMENT
177 LC_MEASUREMENT,
178# endif
179# ifdef USE_LOCALE_PAPER
180 LC_PAPER,
181# endif
182# ifdef USE_LOCALE_TELEPHONE
183 LC_TELEPHONE,
184# endif
e5f10d49
KW
185# ifdef LC_ALL
186 LC_ALL,
187# endif
188 -1 /* Placeholder because C doesn't allow a
189 trailing comma, and it would get complicated
190 with all the #ifdef's */
191};
192
193/* The top-most real element is LC_ALL */
194
195const char * category_names[] = {
196
197# ifdef USE_LOCALE_NUMERIC
198 "LC_NUMERIC",
199# endif
200# ifdef USE_LOCALE_CTYPE
201 "LC_CTYPE",
202# endif
203# ifdef USE_LOCALE_COLLATE
204 "LC_COLLATE",
205# endif
206# ifdef USE_LOCALE_TIME
207 "LC_TIME",
208# endif
209# ifdef USE_LOCALE_MESSAGES
210 "LC_MESSAGES",
211# endif
212# ifdef USE_LOCALE_MONETARY
213 "LC_MONETARY",
214# endif
9821811f
KW
215# ifdef USE_LOCALE_ADDRESS
216 "LC_ADDRESS",
217# endif
218# ifdef USE_LOCALE_IDENTIFICATION
219 "LC_IDENTIFICATION",
220# endif
221# ifdef USE_LOCALE_MEASUREMENT
222 "LC_MEASUREMENT",
223# endif
224# ifdef USE_LOCALE_PAPER
225 "LC_PAPER",
226# endif
227# ifdef USE_LOCALE_TELEPHONE
228 "LC_TELEPHONE",
229# endif
e5f10d49
KW
230# ifdef LC_ALL
231 "LC_ALL",
232# endif
233 NULL /* Placeholder */
234 };
235
236# ifdef LC_ALL
237
238 /* On systems with LC_ALL, it is kept in the highest index position. (-2
239 * to account for the final unused placeholder element.) */
240# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
241
242# else
243
244 /* On systems without LC_ALL, we pretend it is there, one beyond the real
245 * top element, hence in the unused placeholder element. */
246# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
247
248# endif
249
250/* Pretending there is an LC_ALL element just above allows us to avoid most
251 * special cases. Most loops through these arrays in the code below are
252 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
253 * on either type of system. But the code must be written to not access the
948523db
KW
254 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
255 * checked for at compile time by using the #define LC_ALL_INDEX which is only
256 * defined if we do have LC_ALL. */
e5f10d49 257
b09aaf40
KW
258STATIC const char *
259S_category_name(const int category)
260{
261 unsigned int i;
262
263#ifdef LC_ALL
264
265 if (category == LC_ALL) {
266 return "LC_ALL";
267 }
268
269#endif
270
271 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
272 if (category == categories[i]) {
273 return category_names[i];
274 }
275 }
276
277 {
278 const char suffix[] = " (unknown)";
279 int temp = category;
280 Size_t length = sizeof(suffix) + 1;
281 char * unknown;
282 dTHX;
283
284 if (temp < 0) {
285 length++;
286 temp = - temp;
287 }
288
289 /* Calculate the number of digits */
290 while (temp >= 10) {
291 temp /= 10;
292 length++;
293 }
294
295 Newx(unknown, length, char);
296 my_snprintf(unknown, length, "%d%s", category, suffix);
297 SAVEFREEPV(unknown);
298 return unknown;
299 }
300}
301
948523db
KW
302/* Now create LC_foo_INDEX #defines for just those categories on this system */
303# ifdef USE_LOCALE_NUMERIC
304# define LC_NUMERIC_INDEX 0
305# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
306# else
307# define _DUMMY_NUMERIC -1
308# endif
309# ifdef USE_LOCALE_CTYPE
310# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
311# define _DUMMY_CTYPE LC_CTYPE_INDEX
312# else
313# define _DUMMY_CTYPE _DUMMY_NUMERIC
314# endif
315# ifdef USE_LOCALE_COLLATE
316# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
317# define _DUMMY_COLLATE LC_COLLATE_INDEX
318# else
319# define _DUMMY_COLLATE _DUMMY_COLLATE
320# endif
321# ifdef USE_LOCALE_TIME
322# define LC_TIME_INDEX _DUMMY_COLLATE + 1
323# define _DUMMY_TIME LC_TIME_INDEX
324# else
325# define _DUMMY_TIME _DUMMY_COLLATE
326# endif
327# ifdef USE_LOCALE_MESSAGES
328# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
329# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
330# else
331# define _DUMMY_MESSAGES _DUMMY_TIME
332# endif
333# ifdef USE_LOCALE_MONETARY
334# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
335# define _DUMMY_MONETARY LC_MONETARY_INDEX
336# else
337# define _DUMMY_MONETARY _DUMMY_MESSAGES
338# endif
9821811f
KW
339# ifdef USE_LOCALE_ADDRESS
340# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
341# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
342# else
343# define _DUMMY_ADDRESS _DUMMY_MONETARY
344# endif
345# ifdef USE_LOCALE_IDENTIFICATION
346# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
347# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
348# else
349# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
350# endif
351# ifdef USE_LOCALE_MEASUREMENT
352# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
353# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
354# else
355# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
356# endif
357# ifdef USE_LOCALE_PAPER
358# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
359# define _DUMMY_PAPER LC_PAPER_INDEX
360# else
361# define _DUMMY_PAPER _DUMMY_MEASUREMENT
362# endif
363# ifdef USE_LOCALE_TELEPHONE
364# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
365# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
366# else
367# define _DUMMY_TELEPHONE _DUMMY_PAPER
368# endif
948523db 369# ifdef LC_ALL
9821811f 370# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
371# endif
372#endif /* ifdef USE_LOCALE */
8ef6e574 373
d2b24094
KW
374/* Windows requres a customized base-level setlocale() */
375# ifdef WIN32
376# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
377# else
378# define my_setlocale(cat, locale) setlocale(cat, locale)
379# endif
380
837ce802
KW
381/* Just placeholders for now. "_c" is intended to be called when the category
382 * is a constant known at compile time; "_r", not known until run time */
d2b24094
KW
383# define do_setlocale_c(category, locale) my_setlocale(category, locale)
384# define do_setlocale_r(category, locale) my_setlocale(category, locale)
837ce802 385
a4f00dcc 386STATIC void
86799d2d 387S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 388{
86799d2d
KW
389 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
390 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 391
86799d2d
KW
392#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
393 || defined(HAS_NL_LANGINFO))
98994639 394
87f8e8e7
KW
395 const char * radix = (use_locale)
396 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
397 /* FALSE => already in dest locale */
398 : ".";
2213a3be 399
87f8e8e7 400 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 401
87f8e8e7
KW
402 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
403 * a UTF-8 locale, then mark the radix as being in UTF-8 */
404 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 405 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
406 && _is_cur_LC_category_utf8(LC_NUMERIC))
407 {
408 SvUTF8_on(PL_numeric_radix_sv);
409 }
86799d2d
KW
410
411# ifdef DEBUGGING
7d4bcc4a 412
2fcc0ca9
KW
413 if (DEBUG_L_TEST || debug_initialization) {
414 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
415 SvPVX(PL_numeric_radix_sv),
416 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 417 }
69014004 418
86799d2d
KW
419# endif
420#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 421
98994639
HS
422}
423
a39edc4c 424
98994639 425void
8772537c 426Perl_new_numeric(pTHX_ const char *newnum)
98994639 427{
7d4bcc4a
KW
428
429#ifndef USE_LOCALE_NUMERIC
430
431 PERL_UNUSED_ARG(newnum);
432
433#else
0d071d52 434
291a84fb 435 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
436 * core Perl this and that 'newnum' is the name of the new locale.
437 * It installs this locale as the current underlying default.
438 *
439 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
440 * set_numeric_underlying() and set_numeric_standard() functions, which
441 * should probably not be called directly, but only via macros like
0d071d52
KW
442 * SET_NUMERIC_STANDARD() in perl.h.
443 *
444 * The toggling is necessary mainly so that a non-dot radix decimal point
445 * character can be output, while allowing internal calculations to use a
446 * dot.
447 *
448 * This sets several interpreter-level variables:
bb304765 449 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 450 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
451 * that the current locale is the program's underlying
452 * locale
453 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
454 * that the current locale is the C locale or
455 * indistinguishable from the C locale. If non-zero, it
456 * is in C; if > 1, it means it may not be toggled away
7738054c 457 * from C.
4c68b815
KW
458 * PL_numeric_underlying_is_standard A bool kept by this function
459 * indicating that the underlying locale and the standard
460 * C locale are indistinguishable for the purposes of
461 * LC_NUMERIC. This happens when both of the above two
462 * variables are true at the same time. (Toggling is a
463 * no-op under these circumstances.) This variable is
464 * used to avoid having to recalculate.
0d071d52
KW
465 * Any code changing the locale (outside this file) should use
466 * POSIX::setlocale, which calls this function. Therefore this function
467 * should be called directly only from this file and from
468 * POSIX::setlocale() */
469
b03f34cf 470 char *save_newnum;
98994639
HS
471
472 if (! newnum) {
43c5f42d
NC
473 Safefree(PL_numeric_name);
474 PL_numeric_name = NULL;
98994639 475 PL_numeric_standard = TRUE;
892e6465 476 PL_numeric_underlying = TRUE;
4c68b815 477 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
478 return;
479 }
480
b03f34cf 481 save_newnum = stdize_locale(savepv(newnum));
892e6465 482 PL_numeric_underlying = TRUE;
4c68b815
KW
483 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
484
485 /* If its name isn't C nor POSIX, it could still be indistinguishable from
486 * them */
487 if (! PL_numeric_standard) {
488 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
489 FALSE /* Don't toggle locale */ ))
490 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
491 FALSE)));
492 }
abe1abcf 493
4c68b815 494 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 495 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 496 Safefree(PL_numeric_name);
b03f34cf 497 PL_numeric_name = save_newnum;
b03f34cf 498 }
abe1abcf
KW
499 else {
500 Safefree(save_newnum);
501 }
4c28b29c 502
4c68b815
KW
503 PL_numeric_underlying_is_standard = PL_numeric_standard;
504
7e5377f7 505# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
506
507 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
508 PL_numeric_name,
509 PL_underlying_numeric_obj);
510
511#endif
512
4c68b815
KW
513 if (DEBUG_L_TEST || debug_initialization) {
514 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
515 }
516
4c28b29c
KW
517 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
518 * have to worry about the radix being a non-dot. (Core operations that
519 * need the underlying locale change to it temporarily). */
520 set_numeric_standard();
521
98994639 522#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 523
98994639
HS
524}
525
526void
527Perl_set_numeric_standard(pTHX)
528{
7d4bcc4a 529
98994639 530#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 531
28c1bf33
KW
532 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
533 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
534 * macro avoids calling this routine if toggling isn't necessary according
535 * to our records (which could be wrong if some XS code has changed the
536 * locale behind our back) */
0d071d52 537
837ce802 538 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 539 PL_numeric_standard = TRUE;
4c68b815 540 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 541 set_numeric_radix(0);
7d4bcc4a
KW
542
543# ifdef DEBUGGING
544
2fcc0ca9
KW
545 if (DEBUG_L_TEST || debug_initialization) {
546 PerlIO_printf(Perl_debug_log,
58e4a467 547 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 548 }
98994639 549
7d4bcc4a 550# endif
98994639 551#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 552
98994639
HS
553}
554
555void
5792c642 556Perl_set_numeric_underlying(pTHX)
98994639 557{
7d4bcc4a 558
98994639 559#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 560
28c1bf33 561 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
562 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
563 * instead of calling this directly. The macro avoids calling this routine
564 * if toggling isn't necessary according to our records (which could be
565 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 566
837ce802 567 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 568 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 569 PL_numeric_underlying = TRUE;
3c62fd46 570 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
571
572# ifdef DEBUGGING
573
2fcc0ca9
KW
574 if (DEBUG_L_TEST || debug_initialization) {
575 PerlIO_printf(Perl_debug_log,
58e4a467 576 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
577 PL_numeric_name);
578 }
98994639 579
7d4bcc4a 580# endif
98994639 581#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 582
98994639
HS
583}
584
585/*
586 * Set up for a new ctype locale.
587 */
a4f00dcc
KW
588STATIC void
589S_new_ctype(pTHX_ const char *newctype)
98994639 590{
7d4bcc4a
KW
591
592#ifndef USE_LOCALE_CTYPE
593
594 PERL_ARGS_ASSERT_NEW_CTYPE;
595 PERL_UNUSED_ARG(newctype);
596 PERL_UNUSED_CONTEXT;
597
598#else
0d071d52 599
291a84fb 600 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
601 * core Perl this and that 'newctype' is the name of the new locale.
602 *
603 * This function sets up the folding arrays for all 256 bytes, assuming
604 * that tofold() is tolc() since fold case is not a concept in POSIX,
605 *
606 * Any code changing the locale (outside this file) should use
607 * POSIX::setlocale, which calls this function. Therefore this function
608 * should be called directly only from this file and from
609 * POSIX::setlocale() */
610
27da23d5 611 dVAR;
68067e4e 612 UV i;
98994639 613
8b7358b9
KW
614 /* Don't check for problems if we are suppressing the warnings */
615 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
616
7918f24d
NC
617 PERL_ARGS_ASSERT_NEW_CTYPE;
618
215c5139
KW
619 /* We will replace any bad locale warning with 1) nothing if the new one is
620 * ok; or 2) a new warning for the bad new locale */
621 if (PL_warn_locale) {
622 SvREFCNT_dec_NN(PL_warn_locale);
623 PL_warn_locale = NULL;
624 }
625
c1284011 626 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
627
628 /* A UTF-8 locale gets standard rules. But note that code still has to
629 * handle this specially because of the three problematic code points */
630 if (PL_in_utf8_CTYPE_locale) {
631 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
632 }
8b7358b9
KW
633
634 /* We don't populate the other lists if a UTF-8 locale, but do check that
635 * everything works as expected, unless checking turned off */
636 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
637 /* Assume enough space for every character being bad. 4 spaces each
638 * for the 94 printable characters that are output like "'x' "; and 5
639 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
640 * NUL */
8b7358b9 641 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
642 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
643 to start */
644 unsigned int bad_count = 0; /* Count of bad characters */
645
baa60164 646 for (i = 0; i < 256; i++) {
8b7358b9 647 if (! PL_in_utf8_CTYPE_locale) {
5a8a8f21
KW
648 if (isupper(i))
649 PL_fold_locale[i] = (U8) tolower(i);
650 else if (islower(i))
651 PL_fold_locale[i] = (U8) toupper(i);
baa60164
KW
652 else
653 PL_fold_locale[i] = (U8) i;
8b7358b9 654 }
8c6180a9
KW
655
656 /* If checking for locale problems, see if the native ASCII-range
657 * printables plus \n and \t are in their expected categories in
658 * the new locale. If not, this could mean big trouble, upending
659 * Perl's and most programs' assumptions, like having a
660 * metacharacter with special meaning become a \w. Fortunately,
661 * it's very rare to find locales that aren't supersets of ASCII
662 * nowadays. It isn't a problem for most controls to be changed
663 * into something else; we check only \n and \t, though perhaps \r
664 * could be an issue as well. */
7d4bcc4a 665 if ( check_for_problems
8c6180a9
KW
666 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
667 {
8b7358b9
KW
668 bool is_bad = FALSE;
669 char name[3] = { '\0' };
670
671 /* Convert the name into a string */
672 if (isPRINT_A(i)) {
673 name[0] = i;
674 name[1] = '\0';
675 }
676 else if (i == '\n') {
677 my_strlcpy(name, "\n", sizeof(name));
678 }
679 else {
680 my_strlcpy(name, "\t", sizeof(name));
681 }
682
683 /* Check each possibe class */
684 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
685 is_bad = TRUE;
686 DEBUG_L(PerlIO_printf(Perl_debug_log,
687 "isalnum('%s') unexpectedly is %d\n",
688 name, cBOOL(isalnum(i))));
689 }
690 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
691 is_bad = TRUE;
692 DEBUG_L(PerlIO_printf(Perl_debug_log,
693 "isalpha('%s') unexpectedly is %d\n",
694 name, cBOOL(isalpha(i))));
695 }
696 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
697 is_bad = TRUE;
698 DEBUG_L(PerlIO_printf(Perl_debug_log,
699 "isdigit('%s') unexpectedly is %d\n",
700 name, cBOOL(isdigit(i))));
701 }
702 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
703 is_bad = TRUE;
704 DEBUG_L(PerlIO_printf(Perl_debug_log,
705 "isgraph('%s') unexpectedly is %d\n",
706 name, cBOOL(isgraph(i))));
707 }
708 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
709 is_bad = TRUE;
710 DEBUG_L(PerlIO_printf(Perl_debug_log,
711 "islower('%s') unexpectedly is %d\n",
712 name, cBOOL(islower(i))));
713 }
714 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
715 is_bad = TRUE;
716 DEBUG_L(PerlIO_printf(Perl_debug_log,
717 "isprint('%s') unexpectedly is %d\n",
718 name, cBOOL(isprint(i))));
719 }
720 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
721 is_bad = TRUE;
722 DEBUG_L(PerlIO_printf(Perl_debug_log,
723 "ispunct('%s') unexpectedly is %d\n",
724 name, cBOOL(ispunct(i))));
725 }
726 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
727 is_bad = TRUE;
728 DEBUG_L(PerlIO_printf(Perl_debug_log,
729 "isspace('%s') unexpectedly is %d\n",
730 name, cBOOL(isspace(i))));
731 }
732 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
733 is_bad = TRUE;
734 DEBUG_L(PerlIO_printf(Perl_debug_log,
735 "isupper('%s') unexpectedly is %d\n",
736 name, cBOOL(isupper(i))));
737 }
738 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
739 is_bad = TRUE;
740 DEBUG_L(PerlIO_printf(Perl_debug_log,
741 "isxdigit('%s') unexpectedly is %d\n",
742 name, cBOOL(isxdigit(i))));
743 }
744 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
745 is_bad = TRUE;
746 DEBUG_L(PerlIO_printf(Perl_debug_log,
747 "tolower('%s')=0x%x instead of the expected 0x%x\n",
748 name, tolower(i), (int) toLOWER_A(i)));
749 }
750 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
751 is_bad = TRUE;
752 DEBUG_L(PerlIO_printf(Perl_debug_log,
753 "toupper('%s')=0x%x instead of the expected 0x%x\n",
754 name, toupper(i), (int) toUPPER_A(i)));
755 }
756 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
757 is_bad = TRUE;
758 DEBUG_L(PerlIO_printf(Perl_debug_log,
759 "'\\n' (=%02X) is not a control\n", (int) i));
760 }
761
762 /* Add to the list; Separate multiple entries with a blank */
763 if (is_bad) {
764 if (bad_count) {
765 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 766 }
8b7358b9
KW
767 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
768 bad_count++;
8c6180a9
KW
769 }
770 }
771 }
772
7d4bcc4a
KW
773# ifdef MB_CUR_MAX
774
8c6180a9 775 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 776 * this locale requires more than one byte, there are going to be
8c6180a9 777 * problems. */
9c8a6dc2
KW
778 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
779 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
780 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
781
8b7358b9
KW
782 if ( check_for_problems && MB_CUR_MAX > 1
783 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
784
785 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
786 * locale. Just assume that the implementation for them (plus
787 * for POSIX) is correct and the > 1 value is spurious. (Since
788 * these are specially handled to never be considered UTF-8
789 * locales, as long as this is the only problem, everything
790 * should work fine */
791 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
792 {
8c6180a9
KW
793 multi_byte_locale = TRUE;
794 }
7d4bcc4a
KW
795
796# endif
8c6180a9 797
8b7358b9
KW
798 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
799 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
800 PL_warn_locale = Perl_newSVpvf(aTHX_
801 "Locale '%s' contains (at least) the following characters"
802 " which have\nnon-standard meanings: %s\nThe Perl program"
803 " will use the standard meanings",
804 newctype, bad_chars_list);
805
806 }
807 else {
808 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 809 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 810 newctype,
8c6180a9
KW
811 (multi_byte_locale)
812 ? " Some characters in it are not recognized by"
813 " Perl."
814 : "",
815 (bad_count)
816 ? "\nThe following characters (and maybe others)"
817 " may not have the same meaning as the Perl"
818 " program expects:\n"
819 : "",
820 (bad_count)
821 ? bad_chars_list
822 : ""
823 );
8b7358b9
KW
824 }
825
826 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
827
cc9eaeb0 828 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
829 * output the message now. If not in that scope, we save the
830 * message to be output at the first operation using this locale,
831 * if that actually happens. Most programs don't use locales, so
832 * they are immune to bad ones. */
cc9eaeb0 833 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 834
780fcc9f
KW
835 /* The '0' below suppresses a bogus gcc compiler warning */
836 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 837
bddebb56
KW
838 if (IN_LC(LC_CTYPE)) {
839 SvREFCNT_dec_NN(PL_warn_locale);
840 PL_warn_locale = NULL;
841 }
780fcc9f 842 }
baa60164 843 }
31f05a37 844 }
98994639
HS
845
846#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 847
98994639
HS
848}
849
98994639 850void
2726666d
KW
851Perl__warn_problematic_locale()
852{
2726666d
KW
853
854#ifdef USE_LOCALE_CTYPE
855
5f04a188
KW
856 dTHX;
857
858 /* Internal-to-core function that outputs the message in PL_warn_locale,
859 * and then NULLS it. Should be called only through the macro
860 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
861
2726666d 862 if (PL_warn_locale) {
2726666d
KW
863 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
864 SvPVX(PL_warn_locale),
865 0 /* dummy to avoid compiler warning */ );
2726666d
KW
866 SvREFCNT_dec_NN(PL_warn_locale);
867 PL_warn_locale = NULL;
868 }
869
870#endif
871
872}
873
a4f00dcc
KW
874STATIC void
875S_new_collate(pTHX_ const char *newcoll)
98994639 876{
7d4bcc4a
KW
877
878#ifndef USE_LOCALE_COLLATE
879
880 PERL_UNUSED_ARG(newcoll);
881 PERL_UNUSED_CONTEXT;
882
883#else
0d071d52 884
291a84fb 885 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
886 * core Perl this and that 'newcoll' is the name of the new locale.
887 *
d35fca5f
KW
888 * The design of locale collation is that every locale change is given an
889 * index 'PL_collation_ix'. The first time a string particpates in an
890 * operation that requires collation while locale collation is active, it
891 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
892 * magic includes the collation index, and the transformation of the string
893 * by strxfrm(), q.v. That transformation is used when doing comparisons,
894 * instead of the string itself. If a string changes, the magic is
895 * cleared. The next time the locale changes, the index is incremented,
896 * and so we know during a comparison that the transformation is not
897 * necessarily still valid, and so is recomputed. Note that if the locale
898 * changes enough times, the index could wrap (a U32), and it is possible
899 * that a transformation would improperly be considered valid, leading to
900 * an unlikely bug */
0d071d52 901
98994639
HS
902 if (! newcoll) {
903 if (PL_collation_name) {
904 ++PL_collation_ix;
905 Safefree(PL_collation_name);
906 PL_collation_name = NULL;
907 }
908 PL_collation_standard = TRUE;
00bf60ca 909 is_standard_collation:
98994639
HS
910 PL_collxfrm_base = 0;
911 PL_collxfrm_mult = 2;
165a1c52 912 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 913 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 914 PL_strxfrm_max_cp = 0;
98994639
HS
915 return;
916 }
917
d35fca5f 918 /* If this is not the same locale as currently, set the new one up */
98994639
HS
919 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
920 ++PL_collation_ix;
921 Safefree(PL_collation_name);
922 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 923 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
924 if (PL_collation_standard) {
925 goto is_standard_collation;
926 }
98994639 927
165a1c52 928 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 929 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 930 PL_strxfrm_max_cp = 0;
165a1c52 931
59c018b9
KW
932 /* A locale collation definition includes primary, secondary, tertiary,
933 * etc. weights for each character. To sort, the primary weights are
934 * used, and only if they compare equal, then the secondary weights are
935 * used, and only if they compare equal, then the tertiary, etc.
936 *
937 * strxfrm() works by taking the input string, say ABC, and creating an
938 * output transformed string consisting of first the primary weights,
939 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
940 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
941 * may not have weights at every level. In our example, let's say B
942 * doesn't have a tertiary weight, and A doesn't have a secondary
943 * weight. The constructed string is then going to be
944 * A¹B¹C¹ B²C² A³C³ ....
945 * This has the desired effect that strcmp() will look at the secondary
946 * or tertiary weights only if the strings compare equal at all higher
947 * priority weights. The spaces shown here, like in
c342d20e 948 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
949 * are not just for readability. In the general case, these must
950 * actually be bytes, which we will call here 'separator weights'; and
951 * they must be smaller than any other weight value, but since these
952 * are C strings, only the terminating one can be a NUL (some
953 * implementations may include a non-NUL separator weight just before
954 * the NUL). Implementations tend to reserve 01 for the separator
955 * weights. They are needed so that a shorter string's secondary
956 * weights won't be misconstrued as primary weights of a longer string,
957 * etc. By making them smaller than any other weight, the shorter
958 * string will sort first. (Actually, if all secondary weights are
959 * smaller than all primary ones, there is no need for a separator
960 * weight between those two levels, etc.)
961 *
962 * The length of the transformed string is roughly a linear function of
963 * the input string. It's not exactly linear because some characters
964 * don't have weights at all levels. When we call strxfrm() we have to
965 * allocate some memory to hold the transformed string. The
966 * calculations below try to find coefficients 'm' and 'b' for this
967 * locale so that m*x + b equals how much space we need, given the size
968 * of the input string in 'x'. If we calculate too small, we increase
969 * the size as needed, and call strxfrm() again, but it is better to
970 * get it right the first time to avoid wasted expensive string
971 * transformations. */
972
98994639 973 {
79f120c8
KW
974 /* We use the string below to find how long the tranformation of it
975 * is. Almost all locales are supersets of ASCII, or at least the
976 * ASCII letters. We use all of them, half upper half lower,
977 * because if we used fewer, we might hit just the ones that are
978 * outliers in a particular locale. Most of the strings being
979 * collated will contain a preponderance of letters, and even if
980 * they are above-ASCII, they are likely to have the same number of
981 * weight levels as the ASCII ones. It turns out that digits tend
982 * to have fewer levels, and some punctuation has more, but those
983 * are relatively sparse in text, and khw believes this gives a
984 * reasonable result, but it could be changed if experience so
985 * dictates. */
986 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
987 char * x_longer; /* Transformed 'longer' */
988 Size_t x_len_longer; /* Length of 'x_longer' */
989
990 char * x_shorter; /* We also transform a substring of 'longer' */
991 Size_t x_len_shorter;
992
a4a439fb 993 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
994 * are interested only in its length). It is used because it has
995 * the intelligence to handle all cases, but to work, it needs some
996 * values of 'm' and 'b' to get it started. For the purposes of
997 * this calculation we use a very conservative estimate of 'm' and
998 * 'b'. This assumes a weight can be multiple bytes, enough to
999 * hold any UV on the platform, and there are 5 levels, 4 weight
1000 * bytes, and a trailing NUL. */
1001 PL_collxfrm_base = 5;
1002 PL_collxfrm_mult = 5 * sizeof(UV);
1003
1004 /* Find out how long the transformation really is */
a4a439fb
KW
1005 x_longer = _mem_collxfrm(longer,
1006 sizeof(longer) - 1,
1007 &x_len_longer,
1008
1009 /* We avoid converting to UTF-8 in the
1010 * called function by telling it the
1011 * string is in UTF-8 if the locale is a
1012 * UTF-8 one. Since the string passed
1013 * here is invariant under UTF-8, we can
1014 * claim it's UTF-8 even though it isn't.
1015 * */
1016 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1017 Safefree(x_longer);
1018
1019 /* Find out how long the transformation of a substring of 'longer'
1020 * is. Together the lengths of these transformations are
1021 * sufficient to calculate 'm' and 'b'. The substring is all of
1022 * 'longer' except the first character. This minimizes the chances
1023 * of being swayed by outliers */
a4a439fb 1024 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1025 sizeof(longer) - 2,
a4a439fb
KW
1026 &x_len_shorter,
1027 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1028 Safefree(x_shorter);
1029
1030 /* If the results are nonsensical for this simple test, the whole
1031 * locale definition is suspect. Mark it so that locale collation
1032 * is not active at all for it. XXX Should we warn? */
1033 if ( x_len_shorter == 0
1034 || x_len_longer == 0
1035 || x_len_shorter >= x_len_longer)
1036 {
1037 PL_collxfrm_mult = 0;
1038 PL_collxfrm_base = 0;
1039 }
1040 else {
1041 SSize_t base; /* Temporary */
1042
1043 /* We have both: m * strlen(longer) + b = x_len_longer
1044 * m * strlen(shorter) + b = x_len_shorter;
1045 * subtracting yields:
1046 * m * (strlen(longer) - strlen(shorter))
1047 * = x_len_longer - x_len_shorter
1048 * But we have set things up so that 'shorter' is 1 byte smaller
1049 * than 'longer'. Hence:
1050 * m = x_len_longer - x_len_shorter
1051 *
1052 * But if something went wrong, make sure the multiplier is at
1053 * least 1.
1054 */
1055 if (x_len_longer > x_len_shorter) {
1056 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1057 }
1058 else {
1059 PL_collxfrm_mult = 1;
1060 }
1061
1062 /* mx + b = len
1063 * so: b = len - mx
1064 * but in case something has gone wrong, make sure it is
1065 * non-negative */
1066 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1067 if (base < 0) {
1068 base = 0;
1069 }
1070
1071 /* Add 1 for the trailing NUL */
1072 PL_collxfrm_base = base + 1;
1073 }
58eebef2 1074
7d4bcc4a
KW
1075# ifdef DEBUGGING
1076
58eebef2
KW
1077 if (DEBUG_L_TEST || debug_initialization) {
1078 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1079 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1080 "x_len_longer=%zu,"
1081 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1082 __FILE__, __LINE__,
1083 PL_in_utf8_COLLATE_locale,
1084 x_len_shorter, x_len_longer,
1085 PL_collxfrm_mult, PL_collxfrm_base);
1086 }
7d4bcc4a
KW
1087# endif
1088
98994639
HS
1089 }
1090 }
1091
1092#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1093
98994639
HS
1094}
1095
d2b24094 1096#ifdef WIN32
b8cc575c 1097
a4f00dcc 1098STATIC char *
b8cc575c 1099S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1100{
1101 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1102 * difference between the two unless the input locale is "", which normally
1103 * means on Windows to get the machine default, which is set via the
1104 * computer's "Regional and Language Options" (or its current equivalent).
1105 * In POSIX, it instead means to find the locale from the user's
1106 * environment. This routine changes the Windows behavior to first look in
1107 * the environment, and, if anything is found, use that instead of going to
1108 * the machine default. If there is no environment override, the machine
1109 * default is used, by calling the real setlocale() with "".
1110 *
1111 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1112 * use the particular category's variable if set; otherwise to use the LANG
1113 * variable. */
b385bb4d 1114
175c4cf9 1115 bool override_LC_ALL = FALSE;
89f7b9aa 1116 char * result;
e5f10d49 1117 unsigned int i;
89f7b9aa 1118
b385bb4d 1119 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1120
1121# ifdef LC_ALL
1122
b385bb4d
KW
1123 locale = PerlEnv_getenv("LC_ALL");
1124 if (! locale) {
e5f10d49
KW
1125 if (category == LC_ALL) {
1126 override_LC_ALL = TRUE;
1127 }
1128 else {
7d4bcc4a
KW
1129
1130# endif
7d4bcc4a 1131
e5f10d49
KW
1132 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1133 if (category == categories[i]) {
1134 locale = PerlEnv_getenv(category_names[i]);
1135 goto found_locale;
1136 }
1137 }
7d4bcc4a 1138
b385bb4d 1139 locale = PerlEnv_getenv("LANG");
481465ea 1140 if (! locale) {
b385bb4d
KW
1141 locale = "";
1142 }
e5f10d49
KW
1143
1144 found_locale: ;
7d4bcc4a
KW
1145
1146# ifdef LC_ALL
1147
e5f10d49 1148 }
b385bb4d 1149 }
7d4bcc4a
KW
1150
1151# endif
1152
b385bb4d
KW
1153 }
1154
89f7b9aa 1155 result = setlocale(category, locale);
bbc98134 1156 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
a4f00dcc 1157 setlocale_debug_string(category, locale, result)));
89f7b9aa 1158
481465ea 1159 if (! override_LC_ALL) {
89f7b9aa
KW
1160 return result;
1161 }
1162
dfd77d7a 1163 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
1164 * LANG variable or the system default if there is no LANG. But these have
1165 * lower priority than the other LC_foo variables, so override it for each
1166 * one that is set. (If they are set to "", it means to use the same thing
1167 * we just set LC_ALL to, so can skip) */
7d4bcc4a 1168
948523db 1169 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
1170 result = PerlEnv_getenv(category_names[i]);
1171 if (result && strNE(result, "")) {
1172 setlocale(categories[i], result);
1173 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1174 __FILE__, __LINE__,
1175 setlocale_debug_string(categories[i], result, "not captured")));
1176 }
89f7b9aa 1177 }
7d4bcc4a 1178
bbc98134
KW
1179 result = setlocale(LC_ALL, NULL);
1180 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1181 __FILE__, __LINE__,
a4f00dcc 1182 setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 1183
bbc98134 1184 return result;
b385bb4d
KW
1185}
1186
1187#endif
1188
a4f00dcc
KW
1189char *
1190Perl_setlocale(int category, const char * locale)
1191{
1192 /* This wraps POSIX::setlocale() */
1193
1194 char * retval;
1159483a 1195 char * newlocale;
a4f00dcc
KW
1196 dTHX;
1197
a4f00dcc
KW
1198#ifdef USE_LOCALE_NUMERIC
1199
291a84fb
KW
1200 /* A NULL locale means only query what the current one is. We have the
1201 * LC_NUMERIC name saved, because we are normally switched into the C
1202 * locale for it. For an LC_ALL query, switch back to get the correct
1203 * results. All other categories don't require special handling */
a4f00dcc
KW
1204 if (locale == NULL) {
1205 if (category == LC_NUMERIC) {
1206 return savepv(PL_numeric_name);
1207 }
1208
7d4bcc4a 1209# ifdef LC_ALL
a4f00dcc 1210
c0d737a8
KW
1211 else if (category == LC_ALL && ! PL_numeric_underlying) {
1212
a4f00dcc
KW
1213 SET_NUMERIC_UNDERLYING();
1214 }
1215
7d4bcc4a 1216# endif
a4f00dcc
KW
1217
1218 }
1219
1220#endif
1221
e5f10d49
KW
1222 /* Save retval since subsequent setlocale() calls may overwrite it. */
1223 retval = savepv(do_setlocale_r(category, locale));
a4f00dcc
KW
1224
1225 DEBUG_L(PerlIO_printf(Perl_debug_log,
1226 "%s:%d: %s\n", __FILE__, __LINE__,
1227 setlocale_debug_string(category, locale, retval)));
1228 if (! retval) {
1229 /* Should never happen that a query would return an error, but be
1230 * sure and reset to C locale */
1231 if (locale == 0) {
1232 SET_NUMERIC_STANDARD();
1233 }
7d4bcc4a 1234
a4f00dcc
KW
1235 return NULL;
1236 }
1237
a4f00dcc
KW
1238 /* If locale == NULL, we are just querying the state, but may have switched
1239 * to NUMERIC_UNDERLYING. Switch back before returning. */
1240 if (locale == NULL) {
1241 SET_NUMERIC_STANDARD();
1242 return retval;
1243 }
a4f00dcc 1244
1159483a
KW
1245 /* Now that have switched locales, we have to update our records to
1246 * correspond. */
a4f00dcc 1247
1159483a 1248 switch (category) {
a4f00dcc 1249
1159483a 1250#ifdef USE_LOCALE_CTYPE
a4f00dcc 1251
1159483a
KW
1252 case LC_CTYPE:
1253 new_ctype(retval);
1254 break;
a4f00dcc 1255
1159483a 1256#endif
a4f00dcc
KW
1257#ifdef USE_LOCALE_COLLATE
1258
1159483a
KW
1259 case LC_COLLATE:
1260 new_collate(retval);
1261 break;
a4f00dcc 1262
1159483a 1263#endif
a4f00dcc
KW
1264#ifdef USE_LOCALE_NUMERIC
1265
1159483a
KW
1266 case LC_NUMERIC:
1267 new_numeric(retval);
1268 break;
a4f00dcc 1269
1159483a
KW
1270#endif
1271#ifdef LC_ALL
a4f00dcc 1272
1159483a 1273 case LC_ALL:
a4f00dcc 1274
1159483a
KW
1275 /* LC_ALL updates all the things we care about. The values may not
1276 * be the same as 'retval', as the locale "" may have set things
1277 * individually */
a4f00dcc 1278
1159483a 1279# ifdef USE_LOCALE_CTYPE
a4f00dcc 1280
1159483a
KW
1281 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1282 new_ctype(newlocale);
a4f00dcc 1283
1159483a
KW
1284# endif /* USE_LOCALE_CTYPE */
1285# ifdef USE_LOCALE_COLLATE
1286
1287 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1288 new_collate(newlocale);
a4f00dcc 1289
7d4bcc4a 1290# endif
1159483a 1291# ifdef USE_LOCALE_NUMERIC
a4f00dcc 1292
1159483a
KW
1293 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1294 new_numeric(newlocale);
a4f00dcc 1295
1159483a
KW
1296# endif /* USE_LOCALE_NUMERIC */
1297#endif /* LC_ALL */
a4f00dcc 1298
1159483a
KW
1299 default:
1300 break;
a4f00dcc
KW
1301 }
1302
1303 return retval;
1304
f7416781
KW
1305
1306}
1307
1308PERL_STATIC_INLINE const char *
1309S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1310{
1311 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1312 * growing it if necessary */
1313
1314 const Size_t string_size = strlen(string) + offset + 1;
1315
1316 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1317
1318 if (*buf_size == 0) {
1319 Newx(*buf, string_size, char);
1320 *buf_size = string_size;
1321 }
1322 else if (string_size > *buf_size) {
1323 Renew(*buf, string_size, char);
1324 *buf_size = string_size;
1325 }
1326
1327 Copy(string, *buf + offset, string_size - offset, char);
1328 return *buf;
1329}
1330
1331/*
1332
1333=head1 Locale-related functions and macros
1334
1335=for apidoc Perl_langinfo
1336
7d4bcc4a 1337This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
1338taking the same C<item> parameter values, and returning the same information.
1339But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1340of Perl's locale handling from your code, and can be used on systems that lack
1341a native C<nl_langinfo>.
1342
1343Expanding on these:
1344
1345=over
1346
1347=item *
1348
1349It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1350without you having to write extra code. The reason for the extra code would be
1351because these are from the C<LC_NUMERIC> locale category, which is normally
1352kept set to the C locale by Perl, no matter what the underlying locale is
1353supposed to be, and so to get the expected results, you have to temporarily
1354toggle into the underlying locale, and later toggle back. (You could use
1355plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1356but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1357keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1358expecting the radix (decimal point) character to be a dot.)
1359
1360=item *
1361
1362Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1363makes your code more portable. Of the fifty-some possible items specified by
1364the POSIX 2008 standard,
1365L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1366only two are completely unimplemented. It uses various techniques to recover
1367the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1368both of which are specified in C89, so should be always be available. Later
1369C<strftime()> versions have additional capabilities; C<""> is returned for
1370those not available on your system.
1371
1372The details for those items which may differ from what this emulation returns
1373and what a native C<nl_langinfo()> would return are:
1374
1375=over
1376
1377=item C<CODESET>
1378
1379=item C<ERA>
1380
1381Unimplemented, so returns C<"">.
1382
1383=item C<YESEXPR>
1384
c1566110
KW
1385=item C<YESSTR>
1386
f7416781
KW
1387=item C<NOEXPR>
1388
c1566110
KW
1389=item C<NOSTR>
1390
1391Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
1392removed from POSIX 2008, and are retained for backwards compatibility. Your
1393platform's C<nl_langinfo> may not support them.
f7416781
KW
1394
1395=item C<D_FMT>
1396
1397Always evaluates to C<%x>, the locale's appropriate date representation.
1398
1399=item C<T_FMT>
1400
1401Always evaluates to C<%X>, the locale's appropriate time representation.
1402
1403=item C<D_T_FMT>
1404
1405Always evaluates to C<%c>, the locale's appropriate date and time
1406representation.
1407
1408=item C<CRNCYSTR>
1409
1410The return may be incorrect for those rare locales where the currency symbol
1411replaces the radix character.
1412Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1413to work differently.
1414
1415=item C<ALT_DIGITS>
1416
1417Currently this gives the same results as Linux does.
1418Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1419to work differently.
1420
1421=item C<ERA_D_FMT>
1422
1423=item C<ERA_T_FMT>
1424
1425=item C<ERA_D_T_FMT>
1426
1427=item C<T_FMT_AMPM>
1428
1429These are derived by using C<strftime()>, and not all versions of that function
1430know about them. C<""> is returned for these on such systems.
1431
1432=back
1433
1434When using C<Perl_langinfo> on systems that don't have a native
1435C<nl_langinfo()>, you must
1436
1437 #include "perl_langinfo.h"
1438
1439before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1440C<#include> with this one. (Doing it this way keeps out the symbols that plain
1441C<langinfo.h> imports into the namespace for code that doesn't need it.)
1442
1443You also should not use the bare C<langinfo.h> item names, but should preface
1444them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1445The C<PERL_I<foo>> versions will also work for this function on systems that do
1446have a native C<nl_langinfo>.
1447
1448=item *
1449
1450It is thread-friendly, returning its result in a buffer that won't be
1451overwritten by another thread, so you don't have to code for that possibility.
1452The buffer can be overwritten by the next call to C<nl_langinfo> or
1453C<Perl_langinfo> in the same thread.
1454
1455=item *
1456
7d4bcc4a 1457ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
1458*>>, but you are (only by documentation) forbidden to write into the buffer.
1459By declaring this C<const>, the compiler enforces this restriction. The extra
1460C<const> is why this isn't an unequivocal drop-in replacement for
1461C<nl_langinfo>.
1462
1463=back
1464
1465The original impetus for C<Perl_langinfo()> was so that code that needs to
1466find out the current currency symbol, floating point radix character, or digit
1467grouping separator can use, on all systems, the simpler and more
1468thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1469pain to make thread-friendly. For other fields returned by C<localeconv>, it
1470is better to use the methods given in L<perlcall> to call
1471L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1472
1473=cut
1474
1475*/
1476
1477const char *
1478#ifdef HAS_NL_LANGINFO
1479Perl_langinfo(const nl_item item)
1480#else
1481Perl_langinfo(const int item)
1482#endif
1483{
f61748ac
KW
1484 return my_nl_langinfo(item, TRUE);
1485}
1486
1487const char *
1488#ifdef HAS_NL_LANGINFO
1489S_my_nl_langinfo(const nl_item item, bool toggle)
1490#else
1491S_my_nl_langinfo(const int item, bool toggle)
1492#endif
1493{
ae74815b 1494 dTHX;
f7416781 1495
5a854ab3
KW
1496 /* We only need to toggle into the underlying LC_NUMERIC locale for these
1497 * two items, and only if not already there */
1498 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
1499 || PL_numeric_underlying))
1500 {
1501 toggle = FALSE;
1502 }
1503
ab340fff
KW
1504#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1505#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1506
ab340fff 1507 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1508 * for those items dependent on it. This must be copied to a buffer before
1509 * switching back, as some systems destroy the buffer when setlocale() is
1510 * called */
f7416781 1511
038d3702
KW
1512 {
1513 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
1514
b2fee59c 1515 if (toggle) {
038d3702 1516 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 1517 }
f7416781 1518
5acc4454
KW
1519 LOCALE_LOCK; /* Prevent interference from another thread executing
1520 this code section (the only call to nl_langinfo in
1521 the core) */
1522
b2fee59c
KW
1523 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf,
1524 &PL_langinfo_bufsize, 0);
f7416781 1525
5acc4454
KW
1526 LOCALE_UNLOCK;
1527
b2fee59c 1528 if (toggle) {
038d3702 1529 RESTORE_LC_NUMERIC();
b2fee59c 1530 }
f7416781
KW
1531 }
1532
ab340fff
KW
1533# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1534
5a854ab3 1535 {
b2fee59c
KW
1536 bool do_free = FALSE;
1537 locale_t cur = uselocale((locale_t) 0);
ab340fff 1538
b2fee59c
KW
1539 if (cur == LC_GLOBAL_LOCALE) {
1540 cur = duplocale(LC_GLOBAL_LOCALE);
1541 do_free = TRUE;
1542 }
ab340fff 1543
b2fee59c 1544 if (toggle) {
e1aa2579
KW
1545 if (PL_underlying_numeric_obj) {
1546 cur = PL_underlying_numeric_obj;
1547 }
1548 else {
1549 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1550 do_free = TRUE;
1551 }
b2fee59c 1552 }
ab340fff 1553
b2fee59c
KW
1554 save_to_buffer(nl_langinfo_l(item, cur),
1555 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1556 if (do_free) {
1557 freelocale(cur);
1558 }
5a854ab3 1559 }
ab340fff 1560
c1566110
KW
1561# endif
1562
1563 if (strEQ(PL_langinfo_buf, "")) {
1564 if (item == PERL_YESSTR) {
1565 return "yes";
1566 }
1567 if (item == PERL_NOSTR) {
1568 return "no";
1569 }
1570 }
1571
ab340fff
KW
1572 return PL_langinfo_buf;
1573
f7416781 1574#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
1575
1576 {
1577
f7416781
KW
1578# ifdef HAS_LOCALECONV
1579
43dd6b15 1580 const struct lconv* lc;
038d3702 1581 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
1582
1583# endif
1584# ifdef HAS_STRFTIME
1585
43dd6b15
KW
1586 struct tm tm;
1587 bool return_format = FALSE; /* Return the %format, not the value */
1588 const char * format;
f7416781
KW
1589
1590# endif
1591
43dd6b15
KW
1592 /* We copy the results to a per-thread buffer, even if not
1593 * multi-threaded. This is in part to simplify this code, and partly
1594 * because we need a buffer anyway for strftime(), and partly because a
1595 * call of localeconv() could otherwise wipe out the buffer, and the
1596 * programmer would not be expecting this, as this is a nl_langinfo()
1597 * substitute after all, so s/he might be thinking their localeconv()
1598 * is safe until another localeconv() call. */
f7416781 1599
43dd6b15
KW
1600 switch (item) {
1601 Size_t len;
1602 const char * retval;
f7416781 1603
43dd6b15
KW
1604 /* These 2 are unimplemented */
1605 case PERL_CODESET:
1606 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 1607
43dd6b15
KW
1608 default:
1609 return "";
f7416781 1610
43dd6b15
KW
1611 /* We use only an English set, since we don't know any more */
1612 case PERL_YESEXPR: return "^[+1yY]";
1613 case PERL_YESSTR: return "yes";
1614 case PERL_NOEXPR: return "^[-0nN]";
1615 case PERL_NOSTR: return "no";
f7416781
KW
1616
1617# ifdef HAS_LOCALECONV
1618
43dd6b15 1619 case PERL_CRNCYSTR:
f7416781 1620
43dd6b15
KW
1621 /* We don't bother with localeconv_l() because any system that
1622 * has it is likely to also have nl_langinfo() */
291a84fb 1623
5acc4454
KW
1624 LOCALE_LOCK; /* Prevent interference with other threads
1625 using localeconv() */
1626
43dd6b15
KW
1627 lc = localeconv();
1628 if ( ! lc
1629 || ! lc->currency_symbol
1630 || strEQ("", lc->currency_symbol))
1631 {
1632 LOCALE_UNLOCK;
1633 return "";
1634 }
f7416781 1635
43dd6b15
KW
1636 /* Leave the first spot empty to be filled in below */
1637 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1638 &PL_langinfo_bufsize, 1);
1639 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1640 { /* khw couldn't figure out how the localedef specifications
1641 would show that the $ should replace the radix; this is
1642 just a guess as to how it might work.*/
1643 *PL_langinfo_buf = '.';
1644 }
1645 else if (lc->p_cs_precedes) {
1646 *PL_langinfo_buf = '-';
1647 }
1648 else {
1649 *PL_langinfo_buf = '+';
1650 }
f7416781 1651
43dd6b15
KW
1652 LOCALE_UNLOCK;
1653 break;
f7416781 1654
43dd6b15
KW
1655 case PERL_RADIXCHAR:
1656 case PERL_THOUSEP:
f7416781 1657
43dd6b15 1658 if (toggle) {
038d3702 1659 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 1660 }
f7416781 1661
5acc4454
KW
1662 LOCALE_LOCK; /* Prevent interference with other threads
1663 using localeconv() */
1664
43dd6b15
KW
1665 lc = localeconv();
1666 if (! lc) {
33394adc
KW
1667 retval = "";
1668 }
43dd6b15
KW
1669 else {
1670 retval = (item == PERL_RADIXCHAR)
1671 ? lc->decimal_point
1672 : lc->thousands_sep;
1673 if (! retval) {
1674 retval = "";
1675 }
1676 }
f7416781 1677
43dd6b15
KW
1678 save_to_buffer(retval, &PL_langinfo_buf,
1679 &PL_langinfo_bufsize, 0);
f7416781 1680
5acc4454
KW
1681 LOCALE_UNLOCK;
1682
43dd6b15 1683 if (toggle) {
038d3702 1684 RESTORE_LC_NUMERIC();
43dd6b15 1685 }
f7416781 1686
43dd6b15 1687 break;
f7416781
KW
1688
1689# endif
1690# ifdef HAS_STRFTIME
1691
43dd6b15
KW
1692 /* These are defined by C89, so we assume that strftime supports
1693 * them, and so are returned unconditionally; they may not be what
1694 * the locale actually says, but should give good enough results
1695 * for someone using them as formats (as opposed to trying to parse
1696 * them to figure out what the locale says). The other format
1697 * items are actually tested to verify they work on the platform */
1698 case PERL_D_FMT: return "%x";
1699 case PERL_T_FMT: return "%X";
1700 case PERL_D_T_FMT: return "%c";
1701
1702 /* These formats are only available in later strfmtime's */
1703 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1704 case PERL_T_FMT_AMPM:
1705
1706 /* The rest can be gotten from most versions of strftime(). */
1707 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1708 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1709 case PERL_ABDAY_7:
1710 case PERL_ALT_DIGITS:
1711 case PERL_AM_STR: case PERL_PM_STR:
1712 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1713 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1714 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1715 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1716 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1717 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1718 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1719 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1720 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
1721 case PERL_MON_12:
1722
1723 LOCALE_LOCK;
1724
1725 init_tm(&tm); /* Precaution against core dumps */
1726 tm.tm_sec = 30;
1727 tm.tm_min = 30;
1728 tm.tm_hour = 6;
1729 tm.tm_year = 2017 - 1900;
1730 tm.tm_wday = 0;
1731 tm.tm_mon = 0;
1732 switch (item) {
1733 default:
1734 LOCALE_UNLOCK;
1735 Perl_croak(aTHX_
1736 "panic: %s: %d: switch case: %d problem",
1737 __FILE__, __LINE__, item);
1738 NOT_REACHED; /* NOTREACHED */
1739
1740 case PERL_PM_STR: tm.tm_hour = 18;
1741 case PERL_AM_STR:
1742 format = "%p";
1743 break;
1744
1745 case PERL_ABDAY_7: tm.tm_wday++;
1746 case PERL_ABDAY_6: tm.tm_wday++;
1747 case PERL_ABDAY_5: tm.tm_wday++;
1748 case PERL_ABDAY_4: tm.tm_wday++;
1749 case PERL_ABDAY_3: tm.tm_wday++;
1750 case PERL_ABDAY_2: tm.tm_wday++;
1751 case PERL_ABDAY_1:
1752 format = "%a";
1753 break;
1754
1755 case PERL_DAY_7: tm.tm_wday++;
1756 case PERL_DAY_6: tm.tm_wday++;
1757 case PERL_DAY_5: tm.tm_wday++;
1758 case PERL_DAY_4: tm.tm_wday++;
1759 case PERL_DAY_3: tm.tm_wday++;
1760 case PERL_DAY_2: tm.tm_wday++;
1761 case PERL_DAY_1:
1762 format = "%A";
1763 break;
1764
1765 case PERL_ABMON_12: tm.tm_mon++;
1766 case PERL_ABMON_11: tm.tm_mon++;
1767 case PERL_ABMON_10: tm.tm_mon++;
1768 case PERL_ABMON_9: tm.tm_mon++;
1769 case PERL_ABMON_8: tm.tm_mon++;
1770 case PERL_ABMON_7: tm.tm_mon++;
1771 case PERL_ABMON_6: tm.tm_mon++;
1772 case PERL_ABMON_5: tm.tm_mon++;
1773 case PERL_ABMON_4: tm.tm_mon++;
1774 case PERL_ABMON_3: tm.tm_mon++;
1775 case PERL_ABMON_2: tm.tm_mon++;
1776 case PERL_ABMON_1:
1777 format = "%b";
1778 break;
1779
1780 case PERL_MON_12: tm.tm_mon++;
1781 case PERL_MON_11: tm.tm_mon++;
1782 case PERL_MON_10: tm.tm_mon++;
1783 case PERL_MON_9: tm.tm_mon++;
1784 case PERL_MON_8: tm.tm_mon++;
1785 case PERL_MON_7: tm.tm_mon++;
1786 case PERL_MON_6: tm.tm_mon++;
1787 case PERL_MON_5: tm.tm_mon++;
1788 case PERL_MON_4: tm.tm_mon++;
1789 case PERL_MON_3: tm.tm_mon++;
1790 case PERL_MON_2: tm.tm_mon++;
1791 case PERL_MON_1:
1792 format = "%B";
1793 break;
1794
1795 case PERL_T_FMT_AMPM:
1796 format = "%r";
1797 return_format = TRUE;
1798 break;
1799
1800 case PERL_ERA_D_FMT:
1801 format = "%Ex";
1802 return_format = TRUE;
1803 break;
1804
1805 case PERL_ERA_T_FMT:
1806 format = "%EX";
1807 return_format = TRUE;
1808 break;
1809
1810 case PERL_ERA_D_T_FMT:
1811 format = "%Ec";
1812 return_format = TRUE;
1813 break;
1814
1815 case PERL_ALT_DIGITS:
1816 tm.tm_wday = 0;
1817 format = "%Ow"; /* Find the alternate digit for 0 */
1818 break;
1819 }
f7416781 1820
43dd6b15
KW
1821 /* We can't use my_strftime() because it doesn't look at
1822 * tm_wday */
1823 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1824 format, &tm))
1825 {
1826 /* A zero return means one of:
1827 * a) there wasn't enough space in PL_langinfo_buf
1828 * b) the format, like a plain %p, returns empty
1829 * c) it was an illegal format, though some
1830 * implementations of strftime will just return the
1831 * illegal format as a plain character sequence.
1832 *
1833 * To quickly test for case 'b)', try again but precede
1834 * the format with a plain character. If that result is
1835 * still empty, the problem is either 'a)' or 'c)' */
1836
1837 Size_t format_size = strlen(format) + 1;
1838 Size_t mod_size = format_size + 1;
1839 char * mod_format;
1840 char * temp_result;
1841
1842 Newx(mod_format, mod_size, char);
1843 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 1844 *mod_format = ' ';
43dd6b15
KW
1845 my_strlcpy(mod_format + 1, format, mod_size);
1846 len = strftime(temp_result,
1847 PL_langinfo_bufsize,
1848 mod_format, &tm);
1849 Safefree(mod_format);
1850 Safefree(temp_result);
1851
1852 /* If 'len' is non-zero, it means that we had a case like
1853 * %p which means the current locale doesn't use a.m. or
1854 * p.m., and that is valid */
1855 if (len == 0) {
1856
1857 /* Here, still didn't work. If we get well beyond a
1858 * reasonable size, bail out to prevent an infinite
1859 * loop. */
1860
1861 if (PL_langinfo_bufsize > 100 * format_size) {
1862 *PL_langinfo_buf = '\0';
1863 }
1864 else {
1865 /* Double the buffer size to retry; Add 1 in case
1866 * original was 0, so we aren't stuck at 0. */
1867 PL_langinfo_bufsize *= 2;
1868 PL_langinfo_bufsize++;
1869 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1870 continue;
1871 }
1872 }
f7416781 1873
f7416781 1874 break;
43dd6b15 1875 }
f7416781 1876
43dd6b15
KW
1877 /* Here, we got a result.
1878 *
1879 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1880 * alternate format for wday 0. If the value is the same as
1881 * the normal 0, there isn't an alternate, so clear the buffer.
1882 * */
1883 if ( item == PERL_ALT_DIGITS
1884 && strEQ(PL_langinfo_buf, "0"))
1885 {
1886 *PL_langinfo_buf = '\0';
1887 }
f7416781 1888
43dd6b15
KW
1889 /* ALT_DIGITS is problematic. Experiments on it showed that
1890 * strftime() did not always work properly when going from
1891 * alt-9 to alt-10. Only a few locales have this item defined,
1892 * and in all of them on Linux that khw was able to find,
1893 * nl_langinfo() merely returned the alt-0 character, possibly
1894 * doubled. Most Unicode digits are in blocks of 10
1895 * consecutive code points, so that is sufficient information
1896 * for those scripts, as we can infer alt-1, alt-2, .... But
1897 * for a Japanese locale, a CJK ideographic 0 is returned, and
1898 * the CJK digits are not in code point order, so you can't
1899 * really infer anything. The localedef for this locale did
1900 * specify the succeeding digits, so that strftime() works
1901 * properly on them, without needing to infer anything. But
1902 * the nl_langinfo() return did not give sufficient information
1903 * for the caller to understand what's going on. So until
1904 * there is evidence that it should work differently, this
1905 * returns the alt-0 string for ALT_DIGITS.
1906 *
1907 * wday was chosen because its range is all a single digit.
1908 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 1909
43dd6b15 1910 LOCALE_UNLOCK;
f7416781 1911
43dd6b15
KW
1912 /* If to return the format, not the value, overwrite the buffer
1913 * with it. But some strftime()s will keep the original format
1914 * if illegal, so change those to "" */
1915 if (return_format) {
1916 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
1917 *PL_langinfo_buf = '\0';
1918 }
43dd6b15
KW
1919 else {
1920 save_to_buffer(format, &PL_langinfo_buf,
1921 &PL_langinfo_bufsize, 0);
f7416781
KW
1922 }
1923 }
1924
1925 break;
f7416781
KW
1926
1927# endif
1928
43dd6b15 1929 }
f7416781
KW
1930 }
1931
1932 return PL_langinfo_buf;
1933
1934#endif
1935
a4f00dcc 1936}
b385bb4d 1937
98994639
HS
1938/*
1939 * Initialize locale awareness.
1940 */
1941int
1942Perl_init_i18nl10n(pTHX_ int printwarn)
1943{
0e92a118
KW
1944 /* printwarn is
1945 *
1946 * 0 if not to output warning when setup locale is bad
1947 * 1 if to output warning based on value of PERL_BADLANG
1948 * >1 if to output regardless of PERL_BADLANG
1949 *
1950 * returns
98994639 1951 * 1 = set ok or not applicable,
0e92a118
KW
1952 * 0 = fallback to a locale of lower priority
1953 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1954 *
1955 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1956 * set, debugging information is output.
1957 *
1958 * This looks more complicated than it is, mainly due to the #ifdefs.
1959 *
1960 * We try to set LC_ALL to the value determined by the environment. If
1961 * there is no LC_ALL on this platform, we try the individual categories we
1962 * know about. If this works, we are done.
1963 *
1964 * But if it doesn't work, we have to do something else. We search the
1965 * environment variables ourselves instead of relying on the system to do
1966 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1967 * think there is one), and the ultimate fallback "C". This is all done in
1968 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
1969 * more complex. The 'trial_locales' array is initialized with just one
1970 * element; it causes the behavior described in the paragraph above this to
1971 * happen. If that fails, we add elements to 'trial_locales', and do extra
1972 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
1973 *
1974 * On Ultrix, the locale MUST come from the environment, so there is
1975 * preliminary code to set it. I (khw) am not sure that it is necessary,
1976 * and that this couldn't be folded into the loop, but barring any real
1977 * platforms to test on, it's staying as-is
1978 *
1979 * A slight complication is that in embedded Perls, the locale may already
1980 * be set-up, and we don't want to get it from the normal environment
1981 * variables. This is handled by having a special environment variable
1982 * indicate we're in this situation. We simply set setlocale's 2nd
1983 * parameter to be a NULL instead of "". That indicates to setlocale that
1984 * it is not to change anything, but to return the current value,
1985 * effectively initializing perl's db to what the locale already is.
1986 *
1987 * We play the same trick with NULL if a LC_ALL succeeds. We call
1988 * setlocale() on the individual categores with NULL to get their existing
1989 * values for our db, instead of trying to change them.
1990 * */
98994639 1991
0e92a118
KW
1992 int ok = 1;
1993
7d4bcc4a
KW
1994#ifndef USE_LOCALE
1995
1996 PERL_UNUSED_ARG(printwarn);
1997
1998#else /* USE_LOCALE */
7d4bcc4a
KW
1999# ifdef __GLIBC__
2000
175c4cf9 2001 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2002
2003# endif
65ebb059 2004
ccd65d51
KW
2005 /* NULL uses the existing already set up locale */
2006 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2007 ? NULL
2008 : "";
c3fcd832
KW
2009 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2010 unsigned int trial_locales_count;
175c4cf9
KW
2011 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2012 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2013 bool setlocale_failure = FALSE;
65ebb059 2014 unsigned int i;
175c4cf9
KW
2015
2016 /* A later getenv() could zap this, so only use here */
2017 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2018
2019 const bool locwarn = (printwarn > 1
e5f10d49
KW
2020 || ( printwarn
2021 && ( ! bad_lang_use_once
22ff3130 2022 || (
e5f10d49
KW
2023 /* disallow with "" or "0" */
2024 *bad_lang_use_once
2025 && strNE("0", bad_lang_use_once)))));
ea92aad8 2026
291a84fb 2027 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2028 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2029
2030 /* current locale for given category; should have been copied so aren't
2031 * volatile */
8de4332b 2032 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2033
7d4bcc4a
KW
2034# ifdef WIN32
2035
6bce99ee
JH
2036 /* In some systems you can find out the system default locale
2037 * and use that as the fallback locale. */
7d4bcc4a
KW
2038# define SYSTEM_DEFAULT_LOCALE
2039# endif
2040# ifdef SYSTEM_DEFAULT_LOCALE
2041
65ebb059 2042 const char *system_default_locale = NULL;
98994639 2043
7d4bcc4a 2044# endif
948523db
KW
2045
2046# ifndef DEBUGGING
2047# define DEBUG_LOCALE_INIT(a,b,c)
2048# else
7d4bcc4a 2049
8298454c 2050 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2051
2052# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2053 STMT_START { \
2054 if (debug_initialization) { \
2055 PerlIO_printf(Perl_debug_log, \
2056 "%s:%d: %s\n", \
2057 __FILE__, __LINE__, \
a4f00dcc 2058 setlocale_debug_string(category, \
2fcc0ca9
KW
2059 locale, \
2060 result)); \
2061 } \
2062 } STMT_END
2fcc0ca9 2063
948523db
KW
2064/* Make sure the parallel arrays are properly set up */
2065# ifdef USE_LOCALE_NUMERIC
2066 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2067 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
2068# endif
2069# ifdef USE_LOCALE_CTYPE
2070 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2071 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
2072# endif
2073# ifdef USE_LOCALE_COLLATE
2074 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2075 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
2076# endif
2077# ifdef USE_LOCALE_TIME
2078 assert(categories[LC_TIME_INDEX] == LC_TIME);
2079 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2080# endif
2081# ifdef USE_LOCALE_MESSAGES
2082 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2083 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
2084# endif
2085# ifdef USE_LOCALE_MONETARY
2086 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2087 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
2088# endif
9821811f
KW
2089# ifdef USE_LOCALE_ADDRESS
2090 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
2091 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
2092# endif
2093# ifdef USE_LOCALE_IDENTIFICATION
2094 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
2095 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
2096# endif
2097# ifdef USE_LOCALE_MEASUREMENT
2098 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
2099 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
2100# endif
2101# ifdef USE_LOCALE_PAPER
2102 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
2103 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
2104# endif
2105# ifdef USE_LOCALE_TELEPHONE
2106 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
2107 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
2108# endif
948523db
KW
2109# ifdef LC_ALL
2110 assert(categories[LC_ALL_INDEX] == LC_ALL);
2111 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
2112 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
2113# endif
2114# endif /* DEBUGGING */
47280b20
KW
2115
2116 /* Initialize the cache of the program's UTF-8ness for the always known
2117 * locales C and POSIX */
2118 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
2119 sizeof(PL_locale_utf8ness));
2120
3ca88433
KW
2121 PL_numeric_radix_sv = newSVpvs(".");
2122
b56c4436 2123# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
2124
2125 /*
2126 * Ultrix setlocale(..., "") fails if there are no environment
2127 * variables from which to get a locale name.
2128 */
2129
b56c4436
KW
2130# ifndef LC_ALL
2131# error Ultrix without LC_ALL not implemented
2132# else
7d4bcc4a 2133
b56c4436
KW
2134 {
2135 bool done = FALSE;
ea92aad8
KW
2136 if (lang) {
2137 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
2138 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
2139 if (sl_result[LC_ALL_INDEX])
2140 done = TRUE;
2141 else
e5f10d49 2142 setlocale_failure = TRUE;
ea92aad8
KW
2143 }
2144 if (! setlocale_failure) {
2145 const char * locale_param;
2146 for (i = 0; i < LC_ALL_INDEX; i++) {
2147 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
2148 ? setlocale_init
2149 : NULL;
2150 sl_result[i] = do_setlocale_r(categories[i], locale_param);
2151 if (! sl_result[i]) {
2152 setlocale_failure = TRUE;
2153 }
2154 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 2155 }
c835d6be 2156 }
7d4bcc4a
KW
2157 }
2158
2159# endif /* LC_ALL */
e5f10d49 2160# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 2161
65ebb059 2162 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
2163 * the list. Normally the loop is executed just once. But if setting the
2164 * locale fails, inside the loop we add fallback trials to the array and so
2165 * will execute the loop multiple times */
c3fcd832
KW
2166 trial_locales[0] = setlocale_init;
2167 trial_locales_count = 1;
7d4bcc4a 2168
65ebb059
KW
2169 for (i= 0; i < trial_locales_count; i++) {
2170 const char * trial_locale = trial_locales[i];
2171
2172 if (i > 0) {
2173
2174 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
2175 * when i==0, but I (khw) don't think that behavior makes much
2176 * sense */
2177 setlocale_failure = FALSE;
2178
7d4bcc4a 2179# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 2180# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 2181
65ebb059
KW
2182 /* On Windows machines, an entry of "" after the 0th means to use
2183 * the system default locale, which we now proceed to get. */
2184 if (strEQ(trial_locale, "")) {
2185 unsigned int j;
2186
2187 /* Note that this may change the locale, but we are going to do
2188 * that anyway just below */
837ce802 2189 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 2190 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 2191
7d4bcc4a 2192 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
2193 * try */
2194 if (! system_default_locale) {
2195 goto next_iteration;
2196 }
2197 for (j = 0; j < trial_locales_count; j++) {
2198 if (strEQ(system_default_locale, trial_locales[j])) {
2199 goto next_iteration;
2200 }
2201 }
2202
2203 trial_locale = system_default_locale;
2204 }
ec0202b5
KW
2205# else
2206# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
2207# endif
7d4bcc4a 2208# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
2209
2210 } /* For i > 0 */
65ebb059 2211
7d4bcc4a
KW
2212# ifdef LC_ALL
2213
948523db
KW
2214 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
2215 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
2216 if (! sl_result[LC_ALL_INDEX]) {
49c85077 2217 setlocale_failure = TRUE;
7cd8b568
KW
2218 }
2219 else {
2220 /* Since LC_ALL succeeded, it should have changed all the other
2221 * categories it can to its value; so we massage things so that the
2222 * setlocales below just return their category's current values.
2223 * This adequately handles the case in NetBSD where LC_COLLATE may
2224 * not be defined for a locale, and setting it individually will
7d4bcc4a 2225 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
2226 * the POSIX locale. */
2227 trial_locale = NULL;
2228 }
7d4bcc4a
KW
2229
2230# endif /* LC_ALL */
98994639 2231
e5f10d49
KW
2232 if (! setlocale_failure) {
2233 unsigned int j;
2234 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2235 curlocales[j]
2236 = savepv(do_setlocale_r(categories[j], trial_locale));
2237 if (! curlocales[j]) {
2238 setlocale_failure = TRUE;
2239 }
2240 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2241 }
c835d6be 2242
e5f10d49
KW
2243 if (! setlocale_failure) { /* All succeeded */
2244 break; /* Exit trial_locales loop */
49c85077 2245 }
65ebb059 2246 }
98994639 2247
49c85077
KW
2248 /* Here, something failed; will need to try a fallback. */
2249 ok = 0;
65ebb059 2250
49c85077
KW
2251 if (i == 0) {
2252 unsigned int j;
98994639 2253
65ebb059 2254 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2255
2256# ifdef LC_ALL
98994639 2257
49c85077
KW
2258 PerlIO_printf(Perl_error_log,
2259 "perl: warning: Setting locale failed.\n");
98994639 2260
7d4bcc4a 2261# else /* !LC_ALL */
98994639 2262
49c85077
KW
2263 PerlIO_printf(Perl_error_log,
2264 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2265
e5f10d49
KW
2266 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2267 if (! curlocales[j]) {
2268 PerlIO_printf(Perl_error_log, category_names[j]);
2269 }
2270 else {
2271 Safefree(curlocales[j]);
2272 }
2273 }
7d4bcc4a 2274
7d4bcc4a 2275# endif /* LC_ALL */
98994639 2276
49c85077
KW
2277 PerlIO_printf(Perl_error_log,
2278 "perl: warning: Please check that your locale settings:\n");
98994639 2279
7d4bcc4a
KW
2280# ifdef __GLIBC__
2281
49c85077
KW
2282 PerlIO_printf(Perl_error_log,
2283 "\tLANGUAGE = %c%s%c,\n",
2284 language ? '"' : '(',
2285 language ? language : "unset",
2286 language ? '"' : ')');
7d4bcc4a 2287# endif
98994639 2288
49c85077
KW
2289 PerlIO_printf(Perl_error_log,
2290 "\tLC_ALL = %c%s%c,\n",
2291 lc_all ? '"' : '(',
2292 lc_all ? lc_all : "unset",
2293 lc_all ? '"' : ')');
98994639 2294
7d4bcc4a
KW
2295# if defined(USE_ENVIRON_ARRAY)
2296
49c85077 2297 {
cd999af9 2298 char **e;
d5e32b93
KW
2299
2300 /* Look through the environment for any variables of the
2301 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2302 * already handled above. These are assumed to be locale
2303 * settings. Output them and their values. */
cd999af9 2304 for (e = environ; *e; e++) {
d5e32b93
KW
2305 const STRLEN prefix_len = sizeof("LC_") - 1;
2306 STRLEN uppers_len;
2307
cd999af9 2308 if ( strBEGINs(*e, "LC_")
c8b388b0 2309 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2310 && (uppers_len = strspn(*e + prefix_len,
2311 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2312 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2313 {
2314 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2315 (int) (prefix_len + uppers_len), *e,
2316 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2317 }
2318 }
49c85077 2319 }
7d4bcc4a
KW
2320
2321# else
2322
49c85077
KW
2323 PerlIO_printf(Perl_error_log,
2324 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2325
2326# endif
98994639 2327
49c85077
KW
2328 PerlIO_printf(Perl_error_log,
2329 "\tLANG = %c%s%c\n",
2330 lang ? '"' : '(',
2331 lang ? lang : "unset",
2332 lang ? '"' : ')');
98994639 2333
49c85077
KW
2334 PerlIO_printf(Perl_error_log,
2335 " are supported and installed on your system.\n");
2336 }
98994639 2337
65ebb059 2338 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2339 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2340 * usually change the upper bound of the loop we are in.
2341 *
2342 * Since the system's default way of setting the locale has not
2343 * found one that works, We use Perl's defined ordering: LC_ALL,
2344 * LANG, and the C locale. We don't try the same locale twice, so
2345 * don't add to the list if already there. (On POSIX systems, the
2346 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2347 * but there's no harm done by doing it explicitly.
2348 *
2349 * Note that this tries the LC_ALL environment variable even on
2350 * systems which have no LC_ALL locale setting. This may or may
2351 * not have been originally intentional, but there's no real need
2352 * to change the behavior. */
65ebb059
KW
2353 if (lc_all) {
2354 for (j = 0; j < trial_locales_count; j++) {
2355 if (strEQ(lc_all, trial_locales[j])) {
2356 goto done_lc_all;
2357 }
2358 }
2359 trial_locales[trial_locales_count++] = lc_all;
2360 }
2361 done_lc_all:
98994639 2362
65ebb059
KW
2363 if (lang) {
2364 for (j = 0; j < trial_locales_count; j++) {
2365 if (strEQ(lang, trial_locales[j])) {
2366 goto done_lang;
2367 }
2368 }
2369 trial_locales[trial_locales_count++] = lang;
2370 }
2371 done_lang:
2372
7d4bcc4a
KW
2373# if defined(WIN32) && defined(LC_ALL)
2374
65ebb059
KW
2375 /* For Windows, we also try the system default locale before "C".
2376 * (If there exists a Windows without LC_ALL we skip this because
2377 * it gets too complicated. For those, the "C" is the next
2378 * fallback possibility). The "" is the same as the 0th element of
2379 * the array, but the code at the loop above knows to treat it
2380 * differently when not the 0th */
2381 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2382
2383# endif
65ebb059
KW
2384
2385 for (j = 0; j < trial_locales_count; j++) {
2386 if (strEQ("C", trial_locales[j])) {
2387 goto done_C;
2388 }
2389 }
2390 trial_locales[trial_locales_count++] = "C";
98994639 2391
65ebb059
KW
2392 done_C: ;
2393 } /* end of first time through the loop */
98994639 2394
7d4bcc4a
KW
2395# ifdef WIN32
2396
65ebb059 2397 next_iteration: ;
7d4bcc4a
KW
2398
2399# endif
65ebb059
KW
2400
2401 } /* end of looping through the trial locales */
2402
2403 if (ok < 1) { /* If we tried to fallback */
2404 const char* msg;
2405 if (! setlocale_failure) { /* fallback succeeded */
2406 msg = "Falling back to";
2407 }
2408 else { /* fallback failed */
e5f10d49 2409 unsigned int j;
98994639 2410
65ebb059
KW
2411 /* We dropped off the end of the loop, so have to decrement i to
2412 * get back to the value the last time through */
2413 i--;
98994639 2414
65ebb059
KW
2415 ok = -1;
2416 msg = "Failed to fall back to";
2417
2418 /* To continue, we should use whatever values we've got */
7d4bcc4a 2419
e5f10d49
KW
2420 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2421 Safefree(curlocales[j]);
2422 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2423 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2424 }
65ebb059
KW
2425 }
2426
2427 if (locwarn) {
2428 const char * description;
2429 const char * name = "";
2430 if (strEQ(trial_locales[i], "C")) {
2431 description = "the standard locale";
2432 name = "C";
2433 }
7d4bcc4a
KW
2434
2435# ifdef SYSTEM_DEFAULT_LOCALE
2436
65ebb059
KW
2437 else if (strEQ(trial_locales[i], "")) {
2438 description = "the system default locale";
2439 if (system_default_locale) {
2440 name = system_default_locale;
2441 }
2442 }
7d4bcc4a
KW
2443
2444# endif /* SYSTEM_DEFAULT_LOCALE */
2445
65ebb059
KW
2446 else {
2447 description = "a fallback locale";
2448 name = trial_locales[i];
2449 }
2450 if (name && strNE(name, "")) {
2451 PerlIO_printf(Perl_error_log,
2452 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2453 }
2454 else {
2455 PerlIO_printf(Perl_error_log,
2456 "perl: warning: %s %s.\n", msg, description);
2457 }
2458 }
2459 } /* End of tried to fallback */
98994639 2460
e5f10d49
KW
2461 /* Done with finding the locales; update our records */
2462
7d4bcc4a
KW
2463# ifdef USE_LOCALE_CTYPE
2464
948523db 2465 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2466
e5f10d49 2467# endif
7d4bcc4a
KW
2468# ifdef USE_LOCALE_COLLATE
2469
948523db 2470 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2471
e5f10d49 2472# endif
7d4bcc4a
KW
2473# ifdef USE_LOCALE_NUMERIC
2474
948523db 2475 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2476
2477# endif
2478
948523db 2479 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20
KW
2480
2481# if defined(USE_ITHREADS)
2482
2483 /* This caches whether each category's locale is UTF-8 or not. This
2484 * may involve changing the locale. It is ok to do this at
2485 * initialization time before any threads have started, but not later.
2486 * Caching means that if the program heeds our dictate not to change
2487 * locales in threaded applications, this data will remain valid, and
2488 * it may get queried without changing locales. If the environment is
2489 * such that all categories have the same locale, this isn't needed, as
2490 * the code will not change the locale; but this handles the uncommon
2491 * case where the environment has disparate locales for the categories
2492 * */
2493 (void) _is_cur_LC_category_utf8(categories[i]);
2494
2495# endif
2496
e5f10d49
KW
2497 Safefree(curlocales[i]);
2498 }
b310b053 2499
7d4bcc4a
KW
2500# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2501
49c85077 2502 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
2503 * locale is UTF-8. The call to new_ctype() just above has already
2504 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
2505 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
2506 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
2507 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
2508 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 2509
a05d7ebb 2510 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2511 This is an alternative to using the -C command line switch
2512 (the -C if present will override this). */
2513 {
dd374669 2514 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2515 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2516 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2517 PL_utf8cache = -1;
b310b053
JH
2518 }
2519
7d4bcc4a 2520# endif
7d4bcc4a
KW
2521# ifdef __GLIBC__
2522
175c4cf9 2523 Safefree(language);
7d4bcc4a
KW
2524
2525# endif
175c4cf9
KW
2526
2527 Safefree(lc_all);
2528 Safefree(lang);
2529
e3305790 2530#endif /* USE_LOCALE */
2fcc0ca9 2531#ifdef DEBUGGING
7d4bcc4a 2532
2fcc0ca9 2533 /* So won't continue to output stuff */
27cdc72e 2534 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2535
2fcc0ca9
KW
2536#endif
2537
98994639
HS
2538 return ok;
2539}
2540
98994639
HS
2541#ifdef USE_LOCALE_COLLATE
2542
a4a439fb 2543char *
a4a439fb
KW
2544Perl__mem_collxfrm(pTHX_ const char *input_string,
2545 STRLEN len, /* Length of 'input_string' */
2546 STRLEN *xlen, /* Set to length of returned string
2547 (not including the collation index
2548 prefix) */
2549 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2550 )
98994639 2551{
a4a439fb
KW
2552
2553 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2554 * differences. First, it handles embedded NULs. Second, it allocates a bit
2555 * more memory than needed for the transformed data itself. The real
55e5378d 2556 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2557 * the length of that, and doesn't include the collation index size.
2558 * Please see sv_collxfrm() to see how this is used. */
2559
55e5378d
KW
2560#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2561
6696cfa7
KW
2562 char * s = (char *) input_string;
2563 STRLEN s_strlen = strlen(input_string);
79f120c8 2564 char *xbuf = NULL;
55e5378d 2565 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2566 STRLEN length_in_chars;
c664130f 2567 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2568
a4a439fb
KW
2569 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2570
2571 /* Must be NUL-terminated */
2572 assert(*(input_string + len) == '\0');
7918f24d 2573
79f120c8
KW
2574 /* If this locale has defective collation, skip */
2575 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2576 DEBUG_L(PerlIO_printf(Perl_debug_log,
2577 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2578 goto bad;
2579 }
2580
6696cfa7
KW
2581 /* Replace any embedded NULs with the control that sorts before any others.
2582 * This will give as good as possible results on strings that don't
2583 * otherwise contain that character, but otherwise there may be
2584 * less-than-perfect results with that character and NUL. This is
fdc080f3 2585 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2586 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2587 NUL */
6696cfa7
KW
2588 char * e = s + len;
2589 char * sans_nuls;
fdc080f3 2590 STRLEN sans_nuls_len;
94762aa0 2591 int try_non_controls;
afc4976f
KW
2592 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2593 making sure 2nd byte is NUL.
2594 */
2595 STRLEN this_replacement_len;
2596
1e4c9676
KW
2597 /* If we don't know what non-NUL control character sorts lowest for
2598 * this locale, find it */
f28f4d2a 2599 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2600 int j;
afc4976f 2601 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2602 includes the collation index
2603 prefixed. */
2604
91c0e2e0 2605 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2606
2607 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2608 * NUL, in which case we instead look for any character. Controls
2609 * are preferred because collation order is, in general, context
2610 * sensitive, with adjoining characters affecting the order, and
2611 * controls are less likely to have such interactions, allowing the
2612 * NUL-replacement to stand on its own. (Another way to look at it
2613 * is to imagine what would happen if the NUL were replaced by a
2614 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2615 for (try_non_controls = 0;
2616 try_non_controls < 2;
2617 try_non_controls++)
2618 {
d4ff9586
KW
2619 /* Look through all legal code points (NUL isn't) */
2620 for (j = 1; j < 256; j++) {
2621 char * x; /* j's xfrm plus collation index */
2622 STRLEN x_len; /* length of 'x' */
2623 STRLEN trial_len = 1;
736a4fed 2624 char cur_source[] = { '\0', '\0' };
d4ff9586 2625
736a4fed
KW
2626 /* Skip non-controls the first time through the loop. The
2627 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2628 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2629 ? ! isCNTRL_L1(j)
2630 : ! isCNTRL_LC(j))
2631 {
d4ff9586 2632 continue;
6696cfa7 2633 }
6696cfa7 2634
736a4fed
KW
2635 /* Create a 1-char string of the current code point */
2636 cur_source[0] = (char) j;
2637
d4ff9586
KW
2638 /* Then transform it */
2639 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2640 0 /* The string is not in UTF-8 */);
6696cfa7 2641
1e4c9676 2642 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2643 * */
2644 if (! x) {
2645 continue;
2646 }
6696cfa7 2647
d4ff9586
KW
2648 /* If this character's transformation is lower than
2649 * the current lowest, this one becomes the lowest */
2650 if ( cur_min_x == NULL
2651 || strLT(x + COLLXFRM_HDR_LEN,
2652 cur_min_x + COLLXFRM_HDR_LEN))
2653 {
f28f4d2a 2654 PL_strxfrm_NUL_replacement = j;
d4ff9586 2655 cur_min_x = x;
d4ff9586
KW
2656 }
2657 else {
2658 Safefree(x);
2659 }
1e4c9676 2660 } /* end of loop through all 255 characters */
6696cfa7 2661
1e4c9676 2662 /* Stop looking if found */
94762aa0
KW
2663 if (cur_min_x) {
2664 break;
2665 }
2666
2667 /* Unlikely, but possible, if there aren't any controls that
2668 * work in the locale, repeat the loop, looking for any
2669 * character that works */
2670 DEBUG_L(PerlIO_printf(Perl_debug_log,
2671 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2672 } /* End of loop to try first the controls, then any char */
6696cfa7 2673
94762aa0
KW
2674 if (! cur_min_x) {
2675 DEBUG_L(PerlIO_printf(Perl_debug_log,
2676 "_mem_collxfrm: Couldn't find any character to replace"
2677 " embedded NULs in locale %s with", PL_collation_name));
2678 goto bad;
58eebef2
KW
2679 }
2680
94762aa0
KW
2681 DEBUG_L(PerlIO_printf(Perl_debug_log,
2682 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2683 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2684
6696cfa7 2685 Safefree(cur_min_x);
1e4c9676 2686 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2687
2688 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2689 * UTF8-ness of the original */
f28f4d2a
KW
2690 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2691 this_replacement_char[0] =
2692 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2693 this_replacement_char[1] =
2694 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2695 this_replacement_len = 2;
2696 }
2697 else {
f28f4d2a 2698 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2699 /* this_replacement_char[1] = '\0' was done at initialization */
2700 this_replacement_len = 1;
6696cfa7
KW
2701 }
2702
2703 /* The worst case length for the replaced string would be if every
2704 * character in it is NUL. Multiply that by the length of each
2705 * replacement, and allow for a trailing NUL */
afc4976f 2706 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2707 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2708 *sans_nuls = '\0';
2709
6696cfa7
KW
2710 /* Replace each NUL with the lowest collating control. Loop until have
2711 * exhausted all the NULs */
2712 while (s + s_strlen < e) {
6069d6c5 2713 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2714
2715 /* Do the actual replacement */
6069d6c5 2716 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2717
2718 /* Move past the input NUL */
2719 s += s_strlen + 1;
2720 s_strlen = strlen(s);
2721 }
2722
2723 /* And add anything that trails the final NUL */
6069d6c5 2724 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2725
2726 /* Switch so below we transform this modified string */
2727 s = sans_nuls;
2728 len = strlen(s);
1e4c9676 2729 } /* End of replacing NULs */
6696cfa7 2730
a4a439fb
KW
2731 /* Make sure the UTF8ness of the string and locale match */
2732 if (utf8 != PL_in_utf8_COLLATE_locale) {
2733 const char * const t = s; /* Temporary so we can later find where the
2734 input was */
2735
2736 /* Here they don't match. Change the string's to be what the locale is
2737 * expecting */
2738
2739 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2740 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2741 utf8 = TRUE;
2742 }
2743 else { /* locale is not UTF-8; but input is; downgrade the input */
2744
2745 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2746
2747 /* If the downgrade was successful we are done, but if the input
2748 * contains things that require UTF-8 to represent, have to do
2749 * damage control ... */
2750 if (UNLIKELY(utf8)) {
2751
2752 /* What we do is construct a non-UTF-8 string with
2753 * 1) the characters representable by a single byte converted
2754 * to be so (if necessary);
2755 * 2) and the rest converted to collate the same as the
2756 * highest collating representable character. That makes
2757 * them collate at the end. This is similar to how we
2758 * handle embedded NULs, but we use the highest collating
2759 * code point instead of the smallest. Like the NUL case,
2760 * this isn't perfect, but is the best we can reasonably
2761 * do. Every above-255 code point will sort the same as
2762 * the highest-sorting 0-255 code point. If that code
2763 * point can combine in a sequence with some other code
2764 * points for weight calculations, us changing something to
2765 * be it can adversely affect the results. But in most
2766 * cases, it should work reasonably. And note that this is
2767 * really an illegal situation: using code points above 255
2768 * on a locale where only 0-255 are valid. If two strings
2769 * sort entirely equal, then the sort order for the
2770 * above-255 code points will be in code point order. */
2771
2772 utf8 = FALSE;
2773
2774 /* If we haven't calculated the code point with the maximum
2775 * collating order for this locale, do so now */
2776 if (! PL_strxfrm_max_cp) {
2777 int j;
2778
2779 /* The current transformed string that collates the
2780 * highest (except it also includes the prefixed collation
2781 * index. */
2782 char * cur_max_x = NULL;
2783
2784 /* Look through all legal code points (NUL isn't) */
2785 for (j = 1; j < 256; j++) {
2786 char * x;
2787 STRLEN x_len;
736a4fed 2788 char cur_source[] = { '\0', '\0' };
a4a439fb 2789
736a4fed
KW
2790 /* Create a 1-char string of the current code point */
2791 cur_source[0] = (char) j;
a4a439fb
KW
2792
2793 /* Then transform it */
2794 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2795
2796 /* If something went wrong (which it shouldn't), just
2797 * ignore this code point */
94762aa0 2798 if (! x) {
a4a439fb
KW
2799 continue;
2800 }
2801
2802 /* If this character's transformation is higher than
2803 * the current highest, this one becomes the highest */
2804 if ( cur_max_x == NULL
55e5378d
KW
2805 || strGT(x + COLLXFRM_HDR_LEN,
2806 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2807 {
2808 PL_strxfrm_max_cp = j;
2809 cur_max_x = x;
2810 }
2811 else {
2812 Safefree(x);
2813 }
2814 }
2815
94762aa0
KW
2816 if (! cur_max_x) {
2817 DEBUG_L(PerlIO_printf(Perl_debug_log,
2818 "_mem_collxfrm: Couldn't find any character to"
2819 " replace above-Latin1 chars in locale %s with",
2820 PL_collation_name));
2821 goto bad;
2822 }
2823
58eebef2
KW
2824 DEBUG_L(PerlIO_printf(Perl_debug_log,
2825 "_mem_collxfrm: highest 1-byte collating character"
2826 " in locale %s is 0x%02X\n",
2827 PL_collation_name,
2828 PL_strxfrm_max_cp));
58eebef2 2829
a4a439fb
KW
2830 Safefree(cur_max_x);
2831 }
2832
2833 /* Here we know which legal code point collates the highest.
2834 * We are ready to construct the non-UTF-8 string. The length
2835 * will be at least 1 byte smaller than the input string
2836 * (because we changed at least one 2-byte character into a
2837 * single byte), but that is eaten up by the trailing NUL */
2838 Newx(s, len, char);
2839
2840 {
2841 STRLEN i;
2842 STRLEN d= 0;
042d9e50 2843 char * e = (char *) t + len;
a4a439fb
KW
2844
2845 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2846 U8 cur_char = t[i];
2847 if (UTF8_IS_INVARIANT(cur_char)) {
2848 s[d++] = cur_char;
2849 }
042d9e50 2850 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2851 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2852 }
2853 else { /* Replace illegal cp with highest collating
2854 one */
2855 s[d++] = PL_strxfrm_max_cp;
2856 }
2857 }
2858 s[d++] = '\0';
2859 Renew(s, d, char); /* Free up unused space */
2860 }
2861 }
2862 }
2863
2864 /* Here, we have constructed a modified version of the input. It could
2865 * be that we already had a modified copy before we did this version.
2866 * If so, that copy is no longer needed */
2867 if (t != input_string) {
2868 Safefree(t);
2869 }
2870 }
2871
17f41037
KW
2872 length_in_chars = (utf8)
2873 ? utf8_length((U8 *) s, (U8 *) s + len)
2874 : len;
2875
59c018b9
KW
2876 /* The first element in the output is the collation id, used by
2877 * sv_collxfrm(); then comes the space for the transformed string. The
2878 * equation should give us a good estimate as to how much is needed */
55e5378d 2879 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2880 + PL_collxfrm_base
17f41037 2881 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2882 Newx(xbuf, xAlloc, char);
c7202dee
KW
2883 if (UNLIKELY(! xbuf)) {
2884 DEBUG_L(PerlIO_printf(Perl_debug_log,
2885 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2886 goto bad;
c7202dee 2887 }
98994639 2888
d35fca5f 2889 /* Store the collation id */
98994639 2890 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2891
2892 /* Then the transformation of the input. We loop until successful, or we
2893 * give up */
4ebeff16 2894 for (;;) {
1adab0a7 2895
55e5378d 2896 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2897
2898 /* If the transformed string occupies less space than we told strxfrm()
2899 * was available, it means it successfully transformed the whole
2900 * string. */
55e5378d 2901 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2902
1adab0a7
KW
2903 /* Some systems include a trailing NUL in the returned length.
2904 * Ignore it, using a loop in case multiple trailing NULs are
2905 * returned. */
2906 while ( (*xlen) > 0
2907 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2908 {
2909 (*xlen)--;
2910 }
2911
17f41037
KW
2912 /* If the first try didn't get it, it means our prediction was low.
2913 * Modify the coefficients so that we predict a larger value in any
2914 * future transformations */
2915 if (! first_time) {
2916 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2917 STRLEN computed_guess = PL_collxfrm_base
2918 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2919
2920 /* On zero-length input, just keep current slope instead of
2921 * dividing by 0 */
2922 const STRLEN new_m = (length_in_chars != 0)
2923 ? needed / length_in_chars
2924 : PL_collxfrm_mult;
17f41037
KW
2925
2926 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2927 "%s: %d: initial size of %zu bytes for a length "
2928 "%zu string was insufficient, %zu needed\n",
17f41037 2929 __FILE__, __LINE__,
b07929e4 2930 computed_guess, length_in_chars, needed));
17f41037
KW
2931
2932 /* If slope increased, use it, but discard this result for
2933 * length 1 strings, as we can't be sure that it's a real slope
2934 * change */
2935 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
2936
2937# ifdef DEBUGGING
2938
17f41037
KW
2939 STRLEN old_m = PL_collxfrm_mult;
2940 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
2941
2942# endif
2943
17f41037
KW
2944 PL_collxfrm_mult = new_m;
2945 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2946 computed_guess = PL_collxfrm_base
2947 + (PL_collxfrm_mult * length_in_chars);
2948 if (computed_guess < needed) {
2949 PL_collxfrm_base += needed - computed_guess;
2950 }
2951
2952 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2953 "%s: %d: slope is now %zu; was %zu, base "
2954 "is now %zu; was %zu\n",
17f41037 2955 __FILE__, __LINE__,
b07929e4
KW
2956 PL_collxfrm_mult, old_m,
2957 PL_collxfrm_base, old_b));
17f41037
KW
2958 }
2959 else { /* Slope didn't change, but 'b' did */
2960 const STRLEN new_b = needed
2961 - computed_guess
2962 + PL_collxfrm_base;
2963 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2964 "%s: %d: base is now %zu; was %zu\n",
17f41037 2965 __FILE__, __LINE__,
b07929e4 2966 new_b, PL_collxfrm_base));
17f41037
KW
2967 PL_collxfrm_base = new_b;
2968 }
2969 }
2970
4ebeff16
KW
2971 break;
2972 }
bb0f664e 2973
c7202dee
KW
2974 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2975 DEBUG_L(PerlIO_printf(Perl_debug_log,
2976 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2977 *xlen, PERL_INT_MAX));
4ebeff16 2978 goto bad;
c7202dee 2979 }
d35fca5f 2980
c664130f 2981 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
2982 * (usually not including the trailing NUL) when it fails due to not
2983 * enough space being provided. Assume that this is the case unless
2984 * it's been proven otherwise */
c664130f 2985 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 2986 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
2987 }
2988 else { /* Here, either:
2989 * 1) The strxfrm() has previously shown bad behavior; or
2990 * 2) It isn't the first time through the loop, which means
2991 * that the strxfrm() is now showing bad behavior, because
2992 * we gave it what it said was needed in the previous
2993 * iteration, and it came back saying it needed still more.
2994 * (Many versions of cygwin fit this. When the buffer size
2995 * isn't sufficient, they return the input size instead of
2996 * how much is needed.)
d4ff9586
KW
2997 * Increase the buffer size by a fixed percentage and try again.
2998 * */
6ddd902c 2999 xAlloc += (xAlloc / 4) + 1;
c664130f 3000 PL_strxfrm_is_behaved = FALSE;
c664130f 3001
7d4bcc4a
KW
3002# ifdef DEBUGGING
3003
58eebef2
KW
3004 if (DEBUG_Lv_TEST || debug_initialization) {
3005 PerlIO_printf(Perl_debug_log,
3006 "_mem_collxfrm required more space than previously calculated"
b07929e4 3007 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 3008 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 3009 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 3010 }
7d4bcc4a
KW
3011
3012# endif
3013
58eebef2 3014 }
c664130f 3015
4ebeff16 3016 Renew(xbuf, xAlloc, char);
c7202dee
KW
3017 if (UNLIKELY(! xbuf)) {
3018 DEBUG_L(PerlIO_printf(Perl_debug_log,
3019 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 3020 goto bad;
c7202dee 3021 }
c664130f
KW
3022
3023 first_time = FALSE;
4ebeff16 3024 }
98994639 3025
6696cfa7 3026
7d4bcc4a
KW
3027# ifdef DEBUGGING
3028
58eebef2 3029 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
3030
3031 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
3032 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
3033 PerlIO_printf(Perl_debug_log, "%s\n",
3034 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
3035 *xlen, 1));
58eebef2 3036 }
7d4bcc4a
KW
3037
3038# endif
58eebef2 3039
3c5f993e 3040 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 3041 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 3042
6696cfa7
KW
3043 if (s != input_string) {
3044 Safefree(s);
98994639
HS
3045 }
3046
98994639
HS
3047 return xbuf;
3048
3049 bad:
3050 Safefree(xbuf);
6696cfa7
KW
3051 if (s != input_string) {
3052 Safefree(s);
3053 }
98994639 3054 *xlen = 0;
7d4bcc4a
KW
3055
3056# ifdef DEBUGGING
3057
58eebef2 3058 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 3059 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 3060 }
7d4bcc4a
KW
3061
3062# endif
3063
98994639
HS
3064 return NULL;
3065}
3066
7d4bcc4a 3067# ifdef DEBUGGING
c7202dee 3068
4cbaac56 3069STATIC void
c7202dee
KW
3070S_print_collxfrm_input_and_return(pTHX_
3071 const char * const s,
3072 const char * const e,
3073 const STRLEN * const xlen,
3074 const bool is_utf8)
3075{
c7202dee
KW
3076
3077 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
3078
511e4ff7
DM
3079 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
3080 (UV)PL_collation_ix);
c7202dee 3081 if (xlen) {
08b6dc1d 3082 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
3083 }
3084 else {
3085 PerlIO_printf(Perl_debug_log, "NULL");
3086 }
3087 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
3088 PL_collation_name);
9c8a6dc2
KW
3089 print_bytes_for_locale(s, e, is_utf8);
3090
3091 PerlIO_printf(Perl_debug_log, "'\n");
3092}
3093
3094STATIC void
3095S_print_bytes_for_locale(pTHX_
3096 const char * const s,
3097 const char * const e,
3098 const bool is_utf8)
3099{
3100 const char * t = s;
3101 bool prev_was_printable = TRUE;
3102 bool first_time = TRUE;
3103
3104 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
3105
3106 while (t < e) {
3107 UV cp = (is_utf8)
3108 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
3109 : * (U8 *) t;
3110 if (isPRINT(cp)) {
3111 if (! prev_was_printable) {
3112 PerlIO_printf(Perl_debug_log, " ");
3113 }
3114 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
3115 prev_was_printable = TRUE;
3116 }
3117 else {
3118 if (! first_time) {
3119 PerlIO_printf(Perl_debug_log, " ");
3120 }
147e3846 3121 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
3122 prev_was_printable = FALSE;
3123 }
3124 t += (is_utf8) ? UTF8SKIP(t) : 1;
3125 first_time = FALSE;
3126 }
c7202dee
KW
3127}
3128
7d4bcc4a 3129# endif /* #ifdef DEBUGGING */
98994639 3130#endif /* USE_LOCALE_COLLATE */
58eebef2 3131
8ef6e574
KW
3132#ifdef USE_LOCALE
3133
962aa53f
KW
3134STATIC const char *
3135S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
3136{
3137 /* Changes the locale for LC_'switch_category" to that of
3138 * LC_'template_category', if they aren't already the same. If not NULL,
3139 * 'template_locale' is the locale that 'template_category' is in.
3140 *
3141 * Returns the original locale for 'switch_category' so can be switched
3142 * back to with the companion function restore_switched_locale(), (NULL if
3143 * no restoral is necessary.) */
3144
3145 char * restore_to_locale = NULL;
3146
3147 if (switch_category == template_category) { /* No changes needed */
3148 return NULL;
3149 }
3150
3151 /* Find the original locale of the category we may need to change, so that
3152 * it can be restored to later */
d6189047
KW
3153 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
3154 NULL)));
962aa53f
KW
3155 if (! restore_to_locale) {
3156 Perl_croak(aTHX_
3157 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3158 __FILE__, __LINE__, category_name(switch_category), errno);
3159 }
962aa53f
KW
3160
3161 /* If the locale of the template category wasn't passed in, find it now */
3162 if (template_locale == NULL) {
3163 template_locale = do_setlocale_r(template_category, NULL);
3164 if (! template_locale) {
3165 Perl_croak(aTHX_
3166 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3167 __FILE__, __LINE__, category_name(template_category), errno);
3168 }
3169 }
3170
3171 /* It the locales are the same, there's nothing to do */
3172 if (strEQ(restore_to_locale, template_locale)) {
3173 Safefree(restore_to_locale);
3174
3175 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
3176 category_name(switch_category), restore_to_locale));
3177
3178 return NULL;
3179 }
3180
3181 /* Finally, change the locale to the template one */
3182 if (! do_setlocale_r(switch_category, template_locale)) {
3183 Perl_croak(aTHX_
3184 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
3185 __FILE__, __LINE__, category_name(switch_category),
3186 template_locale, errno);
3187 }
3188
3189 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
3190 category_name(switch_category), template_locale));
3191
3192 return restore_to_locale;
3193}
3194
3195STATIC void
3196S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
3197{
3198 /* Restores the locale for LC_'category' to 'original_locale', or do
3199 * nothing if the latter parameter is NULL */
3200
3201 if (original_locale == NULL) {
3202 return;
3203 }
3204
3205 if (! do_setlocale_r(category, original_locale)) {
3206 Perl_croak(aTHX_
3207 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
3208 __FILE__, __LINE__,
3209 category_name(category), original_locale, errno);
3210 }
3211
3212 Safefree(original_locale);
3213}
3214
c1284011
KW
3215bool
3216Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
3217{
3218 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
3219 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 3220 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
3221 * could give the wrong result. The result will very likely be correct for
3222 * languages that have commonly used non-ASCII characters, but for notably
3223 * English, it comes down to if the locale's name ends in something like
3224 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61 3225
47280b20 3226 /* Name of current locale corresponding to the input category */
8de4332b 3227 const char *save_input_locale = NULL;
47280b20
KW
3228
3229 bool is_utf8 = FALSE; /* The return value */
7d74bb61 3230
47280b20
KW
3231 /* The variables below are for the cache of previous lookups using this
3232 * function. The cache is a C string, described at the definition for
3233 * 'C_and_POSIX_utf8ness'.
3234 *
3235 * The first part of the cache is fixed, for the C and POSIX locales. The
3236 * varying part starts just after them. */
3237 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
3238
3239 Size_t utf8ness_cache_size; /* Size of the varying portion */
3240 Size_t input_name_len; /* Length in bytes of save_input_locale */
3241 Size_t input_name_len_with_overhead; /* plus extra chars used to store
3242 the name in the cache */
3243 char * delimited; /* The name plus the delimiters used to store
3244 it in the cache */
3245 char * name_pos; /* position of 'delimited' in the cache, or 0
3246 if not there */
3247
3248
7d4bcc4a
KW
3249# ifdef LC_ALL
3250
7d74bb61 3251 assert(category != LC_ALL);
7d4bcc4a
KW
3252
3253# endif
7d74bb61 3254
47280b20 3255 /* Get the desired category's locale */
d6189047 3256 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
7d74bb61 3257 if (! save_input_locale) {
47280b20 3258 Perl_croak(aTHX_
d707d779
KW
3259 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3260 __FILE__, __LINE__, category_name(category), errno);
7d74bb61 3261 }
47280b20 3262
47280b20
KW
3263 DEBUG_L(PerlIO_printf(Perl_debug_log,
3264 "Current locale for %s is %s\n",
3265 category_name(category), save_input_locale));
3266
3267 input_name_len = strlen(save_input_locale);
3268
3269 /* In our cache, each name is accompanied by two delimiters and a single
3270 * utf8ness digit */
3271 input_name_len_with_overhead = input_name_len + 3;
3272
3273 /* Allocate and populate space for a copy of the name surrounded by the
3274 * delimiters */
3275 Newx(delimited, input_name_len_with_overhead, char);
3276 delimited[0] = UTF8NESS_SEP[0];
3277 Copy(save_input_locale, delimited + 1, input_name_len, char);
3278 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
3279 delimited[input_name_len+2] = '\0';
3280
3281 /* And see if that is in the cache */
3282 name_pos = instr(PL_locale_utf8ness, delimited);
3283 if (name_pos) {
3284 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
3285
3286# ifdef DEBUGGING
3287
3288 if (DEBUG_Lv_TEST || debug_initialization) {
3289 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
3290 save_input_locale, is_utf8);
3291 }
3292
3293# endif
3294
3295 /* And, if not already in that position, move it to the beginning of
3296 * the non-constant portion of the list, since it is the most recently
3297 * used. (We don't have to worry about overflow, since just moving
3298 * existing names around) */
3299 if (name_pos > utf8ness_cache) {
3300 Move(utf8ness_cache,
3301 utf8ness_cache + input_name_len_with_overhead,
3302 name_pos - utf8ness_cache, char);
3303 Copy(delimited,
3304 utf8ness_cache,
3305 input_name_len_with_overhead - 1, char);
3306 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3307 }
3308
3309 Safefree(delimited);
b07fffd1 3310 Safefree(save_input_locale);
47280b20 3311 return is_utf8;
7d74bb61
KW
3312 }
3313
47280b20
KW
3314 /* Here we don't have stored the utf8ness for the input locale. We have to
3315 * calculate it */
3316
94646a69
KW
3317# if defined(USE_LOCALE_CTYPE) \
3318 && ( (defined(HAS_NL_LANGINFO) && defined(CODESET)) \
3319 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
7d74bb61 3320
0dec74cd 3321 {
962aa53f
KW
3322 const char *original_ctype_locale
3323 = switch_category_locale_to_template(LC_CTYPE,
3324 category,
3325 save_input_locale);
69014004 3326
7d74bb61 3327 /* Here the current LC_CTYPE is set to the locale of the category whose
0dec74cd 3328 * information is desired. This means that nl_langinfo() and mbtowc()
1d958db2 3329 * should give the correct results */
119ee68b 3330
0dec74cd
KW
3331# ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
3332 calling the functions if we have this */
3333
3334 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3335 * Unicode code point. */
3336
3337 DEBUG_L(PerlIO_printf(Perl_debug_log, "\tMB_CUR_MAX=%d\n",
3338 (int) MB_CUR_MAX));
3339 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
3340 is_utf8 = FALSE;
b5b2847c
KW
3341 restore_switched_locale(LC_CTYPE, original_ctype_locale);
3342 goto finish_and_return;
0dec74cd
KW
3343 }
3344
3345# endif
7d4bcc4a
KW
3346# if defined(HAS_NL_LANGINFO) && defined(CODESET)
3347
0dec74cd
KW
3348 { /* The task is easiest if the platform has this POSIX 2001 function.
3349 Except on some platforms it can wrongly return "", so have to have
3350 a fallback. And it can return that it's UTF-8, even if there are
3351 variances from that. For example, Turkish locales may use the
3352 alternate dotted I rules, and sometimes it appears to be a
3353 defective locale definition. XXX We should probably check for
3354 these in the Latin1 range and warn */
c70a3e68
KW
3355 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
3356 /* FALSE => already in dest locale */
119ee68b 3357
c70a3e68
KW
3358 DEBUG_L(PerlIO_printf(Perl_debug_log,
3359 "\tnllanginfo returned CODESET '%s'\n", codeset));
3360
3361 if (codeset && strNE(codeset, "")) {
1d958db2 3362
1d075173
KW
3363 /* If the implementation of foldEQ() somehow were
3364 * to change to not go byte-by-byte, this could
3365 * read past end of string, as only one length is
3366 * checked. But currently, a premature NUL will
3367 * compare false, and it will stop there */
3368 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
3369 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
1d958db2 3370
69014004
KW
3371 DEBUG_L(PerlIO_printf(Perl_debug_log,
3372 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
3373 codeset, is_utf8));
b5b2847c
KW
3374 restore_switched_locale(LC_CTYPE, original_ctype_locale);
3375 goto finish_and_return;
1d958db2 3376 }
119ee68b
KW
3377 }
3378
7d4bcc4a 3379# endif
94646a69 3380# if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
0dec74cd
KW
3381 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
3382 * late adder to C89, so very likely to have it. However, testing has
3383 * shown that, like nl_langinfo() above, there are locales that are not
3384 * strictly UTF-8 that this will return that they are */
69014004 3385
0dec74cd 3386 {
119ee68b 3387 wchar_t wc;
51fc4b19
KW
3388 int len;
3389
94646a69
KW
3390# if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
3391
3392 mbstate_t ps;
3393
3394# endif
3395
3396 /* mbrtowc() and mbtowc() convert a byte string to a wide
3397 * character. Feed a byte string to one of them and check that the
3398 * result is the expected Unicode code point */
3399
3400# if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
3401 /* Prefer this function if available, as it's reentrant */
3402
3403 memset(&ps, 0, sizeof(ps));;
3404 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
3405 state */
3406 errno = 0;
3407 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
3408
3409# else
0dec74cd 3410
856b881c 3411 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 3412 errno = 0;
b1d4925c 3413 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
51fc4b19 3414
94646a69
KW
3415# endif
3416
0dec74cd
KW
3417 DEBUG_L(PerlIO_printf(Perl_debug_log,
3418 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
3419 len, (unsigned int) wc, errno));
b1d4925c 3420
0dec74cd
KW
3421 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
3422 && wc == (wchar_t) UNICODE_REPLACEMENT);
119ee68b 3423 }
7d4bcc4a 3424
962aa53f 3425 restore_switched_locale(LC_CTYPE, original_ctype_locale);
47280b20 3426 goto finish_and_return;
7d74bb61 3427 }
119ee68b 3428
0dec74cd
KW
3429# endif
3430# else
7d74bb61 3431
0dec74cd
KW
3432 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
3433 * try looking at the currency symbol to see if it disambiguates
3434 * things. Often that will be in the native script, and if the symbol
3435 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
3436 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
3437 * string being valid UTF-8 are quite small */
fa9b773e 3438
660ee27b
KW
3439# ifdef USE_LOCALE_MONETARY
3440
3441 /* If have LC_MONETARY, we can look at the currency symbol. Often that
3442 * will be in the native script. We do this one first because there is
3443 * just one string to examine, so potentially avoids work */
7d4bcc4a 3444
9db16864
KW
3445 {
3446 const char *original_monetary_locale
660ee27b
KW
3447 = switch_category_locale_to_template(LC_MONETARY,
3448 category,
3449 save_input_locale);
9db16864 3450 bool only_ascii = FALSE;
660ee27b
KW
3451 const U8 * currency_string
3452 = (const U8 *) my_nl_langinfo(PERL_CRNCYSTR, FALSE);
3453 /* 2nd param not relevant for this item */
3454 const U8 * first_variant;
fa9b773e 3455
660ee27b
KW
3456 assert( *currency_string == '-'
3457 || *currency_string == '+'
3458 || *currency_string == '.');
97f4de96 3459
660ee27b 3460 currency_string++;
fa9b773e 3461
660ee27b 3462 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
9db16864
KW
3463 {
3464 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));
3465 only_ascii = TRUE;
3466 }
3467 else {
660ee27b 3468 is_utf8 = is_strict_utf8_string(first_variant, 0);
9db16864 3469 }
fa9b773e 3470
9db16864 3471 restore_switched_locale(LC_MONETARY, original_monetary_locale);
fa9b773e 3472
9db16864 3473 if (! only_ascii) {
fa9b773e 3474
9db16864
KW
3475 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3476 * otherwise assume the locale is UTF-8 if and only if the symbol
3477 * is non-ascii UTF-8. */
3478 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
3479 save_input_locale, is_utf8));
3480 goto finish_and_return;
3481 }
13542a67 3482 }
fa9b773e 3483
660ee27b 3484# endif /* USE_LOCALE_MONETARY */
7d4bcc4a 3485# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e 3486
9db16864
KW
3487 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3488 * the names of the months and weekdays, timezone, and am/pm indicator */
3489 {
3490 const char *original_time_locale
3491 = switch_category_locale_to_template(LC_TIME,
3492 category,
3493 save_input_locale);
3494 int hour = 10;
3495 bool is_dst = FALSE;
3496 int dom = 1;
3497 int month = 0;
3498 int i;
3499 char * formatted_time;
3500
3501 /* Here the current LC_TIME is set to the locale of the category
3502 * whose information is desired. Look at all the days of the week and
3503 * month names, and the timezone and am/pm indicator for UTF-8 variant
3504 * characters. The first such a one found will tell us if the locale
3505 * is UTF-8 or not */
3506
3507 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3508 formatted_time = my_strftime("%A %B %Z %p",
3509 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
3510 if ( ! formatted_time
3511 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3512 {
15f7e74e 3513
9db16864
KW
3514 /* Here, we didn't find a non-ASCII. Try the next time through
3515 * with the complemented dst and am/pm, and try with the next
3516 * weekday. After we have gotten all weekdays, try the next
3517 * month */
3518 is_dst = ! is_dst;
3519 hour = (hour + 12) % 24;
3520 dom++;
3521 if (i > 6) {
3522 month++;
3523 }
3524 continue;
15f7e74e 3525 }
9db16864
KW
3526
3527 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3528 * false otherwise. But first, restore LC_TIME to its original
3529 * locale if we changed it */
3530 restore_switched_locale(LC_TIME, original_time_locale);
3531
3532 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3533 save_input_locale,
3534 is_utf8_string((U8 *) formatted_time, 0)));
3535 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
3536 goto finish_and_return;
15f7e74e
KW
3537 }
3538
9db16864
KW
3539 /* Falling off the end of the loop indicates all the names were just
3540 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3541 * to its original locale */
962aa53f 3542 restore_switched_locale(LC_TIME, original_time_locale);
9db16864 3543 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));
15f7e74e
KW
3544 }
3545
7d4bcc4a 3546# endif
15f7e74e 3547
7d4bcc4a 3548# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93 3549
9db16864
KW
3550 /* This code is ifdefd out because it was found to not be necessary in testing
3551 * on our dromedary test machine, which has over 700 locales. There, this
3552 * added no value to looking at the currency symbol and the time strings. I
3553 * left it in so as to avoid rewriting it if real-world experience indicates
3554 * that dromedary is an outlier. Essentially, instead of returning abpve if we
3555 * haven't found illegal utf8, we continue on and examine all the strerror()
3556 * messages on the platform for utf8ness. If all are ASCII, we still don't
3557 * know the answer; but otherwise we have a pretty good indication of the
3558 * utf8ness. The reason this doesn't help much is that the messages may not
3559 * have been translated into the locale. The currency symbol and time strings
3560 * are much more likely to have been translated. */
3561 {
3562 int e;
3563 bool non_ascii = FALSE;
3564 const char *original_messages_locale
3565 = switch_category_locale_to_template(LC_MESSAGES,
3566 category,
3567 save_input_locale);
3568 const char * errmsg = NULL;
3569
3570 /* Here the current LC_MESSAGES is set to the locale of the category
3571 * whose information is desired. Look through all the messages. We
3572 * can't use Strerror() here because it may expand to code that
3573 * segfaults in miniperl */
3574
3575 for (e = 0; e <= sys_nerr; e++) {
3576 errno = 0;
3577 errmsg = sys_errlist[e];
3578 if (errno || !errmsg) {
3579 break;
3580 }
3581 errmsg = savepv(errmsg);
3582 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
3583 non_ascii = TRUE;
3584 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3585 break;
3586 }
855aeb93 3587 }
9db16864 3588 Safefree(errmsg);
855aeb93 3589
9db16864 3590 restore_switched_locale(LC_MESSAGES, original_messages_locale);
855aeb93 3591
9db16864 3592 if (non_ascii) {
5857e934 3593
9db16864
KW
3594 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3595 * any non-ascii means it is one; otherwise we assume it isn't */
3596 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3597 save_input_locale,
3598 is_utf8));
3599 goto finish_and_return;
3600 }
855aeb93 3601
9db16864
KW
3602 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));
3603 }
855aeb93 3604
7d4bcc4a 3605# endif
0dec74cd 3606# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3607 UTF-8 locale */
7d4bcc4a 3608
97f4de96
KW
3609 /* As a last resort, look at the locale name to see if it matches
3610 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3611 * return of setlocale(), is actually defined to be opaque, so we can't
3612 * really rely on the absence of various substrings in the name to indicate
3613 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3614 * be a UTF-8 locale. Similarly for the other common names */
3615
0dec74cd 3616 {
c36d8df8 3617 const Size_t final_pos = strlen(save_input_locale) - 1;
0dec74cd 3618
c36d8df8
KW
3619 if (final_pos >= 3) {
3620 const char *name = save_input_locale;
97f4de96 3621
c36d8df8
KW
3622 /* Find next 'U' or 'u' and look from there */
3623 while ((name += strcspn(name, "Uu") + 1)
3624 <= save_input_locale + final_pos - 2)
97f4de96 3625 {
c36d8df8
KW
3626 if ( isALPHA_FOLD_NE(*name, 't')
3627 || isALPHA_FOLD_NE(*(name + 1), 'f'))
3628 {
3629 continue;
3630 }
3631 name += 2;
3632 if (*(name) == '-') {
3633 if ((name > save_input_locale + final_pos - 1)) {
3634 break;
3635 }
3636 name++;
3637 }
3638 if (*(name) == '8') {
3639 DEBUG_L(PerlIO_printf(Perl_debug_log,
3640 "Locale %s ends with UTF-8 in name\n",
3641 save_input_locale));
3642 is_utf8 = TRUE;
3643 goto finish_and_return;
97f4de96 3644 }
97f4de96 3645 }
c36d8df8
KW
3646 DEBUG_L(PerlIO_printf(Perl_debug_log,
3647 "Locale %s doesn't end with UTF-8 in name\n",
3648 save_input_locale));
97f4de96 3649 }
97f4de96 3650
4d8d465a 3651# ifdef WIN32
7d4bcc4a 3652
c36d8df8
KW
3653 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
3654 if (memENDs(save_input_locale, final_pos, "65001")) {
3655 DEBUG_L(PerlIO_printf(Perl_debug_log,
8a0832a1 3656 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
97f4de96 3657 save_input_locale));
c36d8df8
KW
3658 is_utf8 = TRUE;
3659 goto finish_and_return;
3660 }
0dec74cd 3661 }
7d4bcc4a 3662
4d8d465a
KW
3663# endif
3664# endif
97f4de96
KW
3665
3666 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3667 * since we are about to return FALSE anyway, there is no point in doing
3668 * this extra work */
7d4bcc4a 3669
0dec74cd 3670# if 0
97f4de96
KW
3671 if (instr(save_input_locale, "8859")) {
3672 DEBUG_L(PerlIO_printf(Perl_debug_log,
3673 "Locale %s has 8859 in name, not UTF-8 locale\n",
3674 save_input_locale));
47280b20
KW
3675 is_utf8 = FALSE;
3676 goto finish_and_return;
97f4de96 3677 }
0dec74cd 3678# endif
97f4de96 3679
69014004
KW
3680 DEBUG_L(PerlIO_printf(Perl_debug_log,
3681 "Assuming locale %s is not a UTF-8 locale\n",
3682 save_input_locale));
47280b20
KW
3683 is_utf8 = FALSE;
3684
0dec74cd
KW
3685# endif /* the code that is compiled when no modern LC_CTYPE */
3686
47280b20
KW
3687 finish_and_return:
3688
3689 /* Cache this result so we don't have to go through all this next time. */
3690 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
3691 - (utf8ness_cache - PL_locale_utf8ness);
3692
3693 /* But we can't save it if it is too large for the total space available */
3694 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
3695 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
3696
3697 /* Here it can fit, but we may need to clear out the oldest cached
3698 * result(s) to do so. Check */
3699 if (utf8ness_cache_len + input_name_len_with_overhead
3700 >= utf8ness_cache_size)
3701 {
3702 /* Here we have to clear something out to make room for this.
3703 * Start looking at the rightmost place where it could fit and find
3704 * the beginning of the entry that extends past that. */
3705 char * cutoff = (char *) my_memrchr(utf8ness_cache,
3706 UTF8NESS_SEP[0],
3707 utf8ness_cache_size
3708 - input_name_len_with_overhead);
3709
3710 assert(cutoff);
3711 assert(cutoff >= utf8ness_cache);
3712
3713 /* This and all subsequent entries must be removed */
3714 *cutoff = '\0';
3715 utf8ness_cache_len = strlen(utf8ness_cache);
3716 }
3717
3718 /* Make space for the new entry */
3719 Move(utf8ness_cache,
3720 utf8ness_cache + input_name_len_with_overhead,
3721 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
3722
3723 /* And insert it */
3724 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
3725 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3726
3727 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
3728 & (PERL_UINTMAX_T) ~1) != '0')
3729 {
3730 Perl_croak(aTHX_
7fcf9272
KW
3731 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu,"
3732 " inserted_name=%s, its_len=%zu\n",
47280b20
KW
3733 __FILE__, __LINE__,
3734 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
3735 delimited, input_name_len_with_overhead);
3736 }
3737 }
3738
3739# ifdef DEBUGGING
3740
3741 if (DEBUG_Lv_TEST || debug_initialization) {
3742 PerlIO_printf(Perl_debug_log,
3743 "PL_locale_utf8ness is now %s; returning %d\n",
3744 PL_locale_utf8ness, is_utf8);
3745 }
3746
3747# endif
3748
3749 Safefree(delimited);
fa9b773e 3750 Safefree(save_input_locale);
47280b20 3751 return is_utf8;
7d74bb61
KW
3752}
3753
8ef6e574 3754#endif
7d74bb61 3755
d6ded950
KW
3756bool
3757Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3758{
1a4f13e1 3759 dVAR;