This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perldelta for dTHX_DEBUGGING
[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
48015184
KW
59
60/* Returns the Unix errno portion; ignoring any others. This is a macro here
61 * instead of putting it into perl.h, because unclear to khw what should be
62 * done generally. */
63#define GET_ERRNO saved_errno
64
291a84fb
KW
65/* strlen() of a literal string constant. We might want this more general,
66 * but using it in just this file for now. A problem with more generality is
67 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
68#define STRLENs(s) (sizeof("" s "") - 1)
69
63e5b0d7
KW
70/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
71 * return of setlocale(), then this is extremely likely to be the C or POSIX
72 * locale. However, the output of setlocale() is documented to be opaque, but
73 * the odds are extremely small that it would return these two strings for some
74 * other locale. Note that VMS in these two locales includes many non-ASCII
75 * characters as controls and punctuation (below are hex bytes):
76 * cntrl: 84-97 9B-9F
77 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
78 * Oddly, none there are listed as alphas, though some represent alphabetics
79 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
80#define isNAME_C_OR_POSIX(name) \
81 ( (name) != NULL \
82 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
83 || strEQ((name), "POSIX")))
84
8ef6e574
KW
85#ifdef USE_LOCALE
86
47280b20
KW
87/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
88 * looked up. This is in the form of a C string: */
89
90#define UTF8NESS_SEP "\v"
91#define UTF8NESS_PREFIX "\f"
92
93/* So, the string looks like:
98994639 94 *
47280b20 95 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 96 *
47280b20
KW
97 * where the digit 0 after the \a indicates that the locale starting just
98 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
99
100STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
101STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
102
103#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
104 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
105
106/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
107 * kept there always. The remining portion of the cache is LRU, with the
108 * oldest looked-up locale at the tail end */
109
98994639
HS
110STATIC char *
111S_stdize_locale(pTHX_ char *locs)
112{
47280b20
KW
113 /* Standardize the locale name from a string returned by 'setlocale',
114 * possibly modifying that string.
115 *
116 * The typical return value of setlocale() is either
117 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
118 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
119 * (the space-separated values represent the various sublocales,
120 * in some unspecified order). This is not handled by this function.
121 *
122 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
123 * which is harmful for further use of the string in setlocale(). This
124 * function removes the trailing new line and everything up through the '='
125 * */
126
7452cf6a 127 const char * const s = strchr(locs, '=');
98994639
HS
128 bool okay = TRUE;
129
7918f24d
NC
130 PERL_ARGS_ASSERT_STDIZE_LOCALE;
131
8772537c
AL
132 if (s) {
133 const char * const t = strchr(s, '.');
98994639 134 okay = FALSE;
8772537c
AL
135 if (t) {
136 const char * const u = strchr(t, '\n');
137 if (u && (u[1] == 0)) {
138 const STRLEN len = u - s;
139 Move(s + 1, locs, len, char);
140 locs[len] = 0;
141 okay = TRUE;
98994639
HS
142 }
143 }
144 }
145
146 if (!okay)
147 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
148
149 return locs;
150}
151
e5f10d49
KW
152/* Two parallel arrays; first the locale categories Perl uses on this system;
153 * the second array is their names. These arrays are in mostly arbitrary
154 * order. */
155
156const int categories[] = {
157
158# ifdef USE_LOCALE_NUMERIC
159 LC_NUMERIC,
160# endif
161# ifdef USE_LOCALE_CTYPE
162 LC_CTYPE,
163# endif
164# ifdef USE_LOCALE_COLLATE
165 LC_COLLATE,
166# endif
167# ifdef USE_LOCALE_TIME
168 LC_TIME,
169# endif
170# ifdef USE_LOCALE_MESSAGES
171 LC_MESSAGES,
172# endif
173# ifdef USE_LOCALE_MONETARY
174 LC_MONETARY,
175# endif
9821811f
KW
176# ifdef USE_LOCALE_ADDRESS
177 LC_ADDRESS,
178# endif
179# ifdef USE_LOCALE_IDENTIFICATION
180 LC_IDENTIFICATION,
181# endif
182# ifdef USE_LOCALE_MEASUREMENT
183 LC_MEASUREMENT,
184# endif
185# ifdef USE_LOCALE_PAPER
186 LC_PAPER,
187# endif
188# ifdef USE_LOCALE_TELEPHONE
189 LC_TELEPHONE,
190# endif
e5f10d49
KW
191# ifdef LC_ALL
192 LC_ALL,
193# endif
194 -1 /* Placeholder because C doesn't allow a
195 trailing comma, and it would get complicated
196 with all the #ifdef's */
197};
198
199/* The top-most real element is LC_ALL */
200
201const char * category_names[] = {
202
203# ifdef USE_LOCALE_NUMERIC
204 "LC_NUMERIC",
205# endif
206# ifdef USE_LOCALE_CTYPE
207 "LC_CTYPE",
208# endif
209# ifdef USE_LOCALE_COLLATE
210 "LC_COLLATE",
211# endif
212# ifdef USE_LOCALE_TIME
213 "LC_TIME",
214# endif
215# ifdef USE_LOCALE_MESSAGES
216 "LC_MESSAGES",
217# endif
218# ifdef USE_LOCALE_MONETARY
219 "LC_MONETARY",
220# endif
9821811f
KW
221# ifdef USE_LOCALE_ADDRESS
222 "LC_ADDRESS",
223# endif
224# ifdef USE_LOCALE_IDENTIFICATION
225 "LC_IDENTIFICATION",
226# endif
227# ifdef USE_LOCALE_MEASUREMENT
228 "LC_MEASUREMENT",
229# endif
230# ifdef USE_LOCALE_PAPER
231 "LC_PAPER",
232# endif
233# ifdef USE_LOCALE_TELEPHONE
234 "LC_TELEPHONE",
235# endif
e5f10d49
KW
236# ifdef LC_ALL
237 "LC_ALL",
238# endif
239 NULL /* Placeholder */
240 };
241
242# ifdef LC_ALL
243
244 /* On systems with LC_ALL, it is kept in the highest index position. (-2
245 * to account for the final unused placeholder element.) */
246# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
247
248# else
249
250 /* On systems without LC_ALL, we pretend it is there, one beyond the real
251 * top element, hence in the unused placeholder element. */
252# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
253
254# endif
255
256/* Pretending there is an LC_ALL element just above allows us to avoid most
257 * special cases. Most loops through these arrays in the code below are
258 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
259 * on either type of system. But the code must be written to not access the
948523db
KW
260 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
261 * checked for at compile time by using the #define LC_ALL_INDEX which is only
262 * defined if we do have LC_ALL. */
e5f10d49 263
b09aaf40
KW
264STATIC const char *
265S_category_name(const int category)
266{
267 unsigned int i;
268
269#ifdef LC_ALL
270
271 if (category == LC_ALL) {
272 return "LC_ALL";
273 }
274
275#endif
276
277 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
278 if (category == categories[i]) {
279 return category_names[i];
280 }
281 }
282
283 {
284 const char suffix[] = " (unknown)";
285 int temp = category;
286 Size_t length = sizeof(suffix) + 1;
287 char * unknown;
288 dTHX;
289
290 if (temp < 0) {
291 length++;
292 temp = - temp;
293 }
294
295 /* Calculate the number of digits */
296 while (temp >= 10) {
297 temp /= 10;
298 length++;
299 }
300
301 Newx(unknown, length, char);
302 my_snprintf(unknown, length, "%d%s", category, suffix);
303 SAVEFREEPV(unknown);
304 return unknown;
305 }
306}
307
948523db
KW
308/* Now create LC_foo_INDEX #defines for just those categories on this system */
309# ifdef USE_LOCALE_NUMERIC
310# define LC_NUMERIC_INDEX 0
311# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
312# else
313# define _DUMMY_NUMERIC -1
314# endif
315# ifdef USE_LOCALE_CTYPE
316# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
317# define _DUMMY_CTYPE LC_CTYPE_INDEX
318# else
319# define _DUMMY_CTYPE _DUMMY_NUMERIC
320# endif
321# ifdef USE_LOCALE_COLLATE
322# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
323# define _DUMMY_COLLATE LC_COLLATE_INDEX
324# else
325# define _DUMMY_COLLATE _DUMMY_COLLATE
326# endif
327# ifdef USE_LOCALE_TIME
328# define LC_TIME_INDEX _DUMMY_COLLATE + 1
329# define _DUMMY_TIME LC_TIME_INDEX
330# else
331# define _DUMMY_TIME _DUMMY_COLLATE
332# endif
333# ifdef USE_LOCALE_MESSAGES
334# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
335# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
336# else
337# define _DUMMY_MESSAGES _DUMMY_TIME
338# endif
339# ifdef USE_LOCALE_MONETARY
340# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
341# define _DUMMY_MONETARY LC_MONETARY_INDEX
342# else
343# define _DUMMY_MONETARY _DUMMY_MESSAGES
344# endif
9821811f
KW
345# ifdef USE_LOCALE_ADDRESS
346# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
347# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
348# else
349# define _DUMMY_ADDRESS _DUMMY_MONETARY
350# endif
351# ifdef USE_LOCALE_IDENTIFICATION
352# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
353# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
354# else
355# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
356# endif
357# ifdef USE_LOCALE_MEASUREMENT
358# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
359# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
360# else
361# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
362# endif
363# ifdef USE_LOCALE_PAPER
364# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
365# define _DUMMY_PAPER LC_PAPER_INDEX
366# else
367# define _DUMMY_PAPER _DUMMY_MEASUREMENT
368# endif
369# ifdef USE_LOCALE_TELEPHONE
370# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
371# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
372# else
373# define _DUMMY_TELEPHONE _DUMMY_PAPER
374# endif
948523db 375# ifdef LC_ALL
9821811f 376# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
377# endif
378#endif /* ifdef USE_LOCALE */
8ef6e574 379
d2b24094
KW
380/* Windows requres a customized base-level setlocale() */
381# ifdef WIN32
382# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
383# else
384# define my_setlocale(cat, locale) setlocale(cat, locale)
385# endif
386
837ce802
KW
387/* Just placeholders for now. "_c" is intended to be called when the category
388 * is a constant known at compile time; "_r", not known until run time */
d2b24094
KW
389# define do_setlocale_c(category, locale) my_setlocale(category, locale)
390# define do_setlocale_r(category, locale) my_setlocale(category, locale)
837ce802 391
a4f00dcc 392STATIC void
86799d2d 393S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 394{
86799d2d
KW
395 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
396 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 397
86799d2d
KW
398#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
399 || defined(HAS_NL_LANGINFO))
98994639 400
87f8e8e7
KW
401 const char * radix = (use_locale)
402 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
403 /* FALSE => already in dest locale */
404 : ".";
2213a3be 405
87f8e8e7 406 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 407
87f8e8e7
KW
408 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
409 * a UTF-8 locale, then mark the radix as being in UTF-8 */
410 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 411 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
412 && _is_cur_LC_category_utf8(LC_NUMERIC))
413 {
414 SvUTF8_on(PL_numeric_radix_sv);
415 }
86799d2d
KW
416
417# ifdef DEBUGGING
7d4bcc4a 418
2fcc0ca9
KW
419 if (DEBUG_L_TEST || debug_initialization) {
420 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
421 SvPVX(PL_numeric_radix_sv),
422 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 423 }
69014004 424
86799d2d
KW
425# endif
426#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 427
98994639
HS
428}
429
a39edc4c 430
98994639 431void
8772537c 432Perl_new_numeric(pTHX_ const char *newnum)
98994639 433{
7d4bcc4a
KW
434
435#ifndef USE_LOCALE_NUMERIC
436
437 PERL_UNUSED_ARG(newnum);
438
439#else
0d071d52 440
291a84fb 441 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
442 * core Perl this and that 'newnum' is the name of the new locale.
443 * It installs this locale as the current underlying default.
444 *
445 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
446 * set_numeric_underlying() and set_numeric_standard() functions, which
447 * should probably not be called directly, but only via macros like
0d071d52
KW
448 * SET_NUMERIC_STANDARD() in perl.h.
449 *
450 * The toggling is necessary mainly so that a non-dot radix decimal point
451 * character can be output, while allowing internal calculations to use a
452 * dot.
453 *
454 * This sets several interpreter-level variables:
bb304765 455 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 456 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
457 * that the current locale is the program's underlying
458 * locale
459 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
460 * that the current locale is the C locale or
461 * indistinguishable from the C locale. If non-zero, it
462 * is in C; if > 1, it means it may not be toggled away
7738054c 463 * from C.
4c68b815
KW
464 * PL_numeric_underlying_is_standard A bool kept by this function
465 * indicating that the underlying locale and the standard
466 * C locale are indistinguishable for the purposes of
467 * LC_NUMERIC. This happens when both of the above two
468 * variables are true at the same time. (Toggling is a
469 * no-op under these circumstances.) This variable is
470 * used to avoid having to recalculate.
0d071d52
KW
471 * Any code changing the locale (outside this file) should use
472 * POSIX::setlocale, which calls this function. Therefore this function
473 * should be called directly only from this file and from
474 * POSIX::setlocale() */
475
b03f34cf 476 char *save_newnum;
98994639
HS
477
478 if (! newnum) {
43c5f42d
NC
479 Safefree(PL_numeric_name);
480 PL_numeric_name = NULL;
98994639 481 PL_numeric_standard = TRUE;
892e6465 482 PL_numeric_underlying = TRUE;
4c68b815 483 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
484 return;
485 }
486
b03f34cf 487 save_newnum = stdize_locale(savepv(newnum));
892e6465 488 PL_numeric_underlying = TRUE;
4c68b815
KW
489 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
490
491 /* If its name isn't C nor POSIX, it could still be indistinguishable from
492 * them */
493 if (! PL_numeric_standard) {
494 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
495 FALSE /* Don't toggle locale */ ))
496 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
497 FALSE)));
498 }
abe1abcf 499
4c68b815 500 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 501 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 502 Safefree(PL_numeric_name);
b03f34cf 503 PL_numeric_name = save_newnum;
b03f34cf 504 }
abe1abcf
KW
505 else {
506 Safefree(save_newnum);
507 }
4c28b29c 508
4c68b815
KW
509 PL_numeric_underlying_is_standard = PL_numeric_standard;
510
7e5377f7 511# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
512
513 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
514 PL_numeric_name,
515 PL_underlying_numeric_obj);
516
517#endif
518
4c68b815
KW
519 if (DEBUG_L_TEST || debug_initialization) {
520 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
521 }
522
4c28b29c
KW
523 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
524 * have to worry about the radix being a non-dot. (Core operations that
525 * need the underlying locale change to it temporarily). */
526 set_numeric_standard();
527
98994639 528#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 529
98994639
HS
530}
531
532void
533Perl_set_numeric_standard(pTHX)
534{
7d4bcc4a 535
98994639 536#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 537
28c1bf33
KW
538 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
539 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
540 * macro avoids calling this routine if toggling isn't necessary according
541 * to our records (which could be wrong if some XS code has changed the
542 * locale behind our back) */
0d071d52 543
837ce802 544 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 545 PL_numeric_standard = TRUE;
4c68b815 546 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 547 set_numeric_radix(0);
7d4bcc4a
KW
548
549# ifdef DEBUGGING
550
2fcc0ca9
KW
551 if (DEBUG_L_TEST || debug_initialization) {
552 PerlIO_printf(Perl_debug_log,
58e4a467 553 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 554 }
98994639 555
7d4bcc4a 556# endif
98994639 557#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 558
98994639
HS
559}
560
561void
5792c642 562Perl_set_numeric_underlying(pTHX)
98994639 563{
7d4bcc4a 564
98994639 565#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 566
28c1bf33 567 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
568 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
569 * instead of calling this directly. The macro avoids calling this routine
570 * if toggling isn't necessary according to our records (which could be
571 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 572
837ce802 573 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 574 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 575 PL_numeric_underlying = TRUE;
3c62fd46 576 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
577
578# ifdef DEBUGGING
579
2fcc0ca9
KW
580 if (DEBUG_L_TEST || debug_initialization) {
581 PerlIO_printf(Perl_debug_log,
58e4a467 582 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
583 PL_numeric_name);
584 }
98994639 585
7d4bcc4a 586# endif
98994639 587#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 588
98994639
HS
589}
590
591/*
592 * Set up for a new ctype locale.
593 */
a4f00dcc
KW
594STATIC void
595S_new_ctype(pTHX_ const char *newctype)
98994639 596{
7d4bcc4a
KW
597
598#ifndef USE_LOCALE_CTYPE
599
600 PERL_ARGS_ASSERT_NEW_CTYPE;
601 PERL_UNUSED_ARG(newctype);
602 PERL_UNUSED_CONTEXT;
603
604#else
0d071d52 605
291a84fb 606 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
607 * core Perl this and that 'newctype' is the name of the new locale.
608 *
609 * This function sets up the folding arrays for all 256 bytes, assuming
610 * that tofold() is tolc() since fold case is not a concept in POSIX,
611 *
612 * Any code changing the locale (outside this file) should use
613 * POSIX::setlocale, which calls this function. Therefore this function
614 * should be called directly only from this file and from
615 * POSIX::setlocale() */
616
27da23d5 617 dVAR;
68067e4e 618 UV i;
98994639 619
8b7358b9
KW
620 /* Don't check for problems if we are suppressing the warnings */
621 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
622
7918f24d
NC
623 PERL_ARGS_ASSERT_NEW_CTYPE;
624
215c5139
KW
625 /* We will replace any bad locale warning with 1) nothing if the new one is
626 * ok; or 2) a new warning for the bad new locale */
627 if (PL_warn_locale) {
628 SvREFCNT_dec_NN(PL_warn_locale);
629 PL_warn_locale = NULL;
630 }
631
c1284011 632 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
633
634 /* A UTF-8 locale gets standard rules. But note that code still has to
635 * handle this specially because of the three problematic code points */
636 if (PL_in_utf8_CTYPE_locale) {
637 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
638 }
8b7358b9
KW
639
640 /* We don't populate the other lists if a UTF-8 locale, but do check that
641 * everything works as expected, unless checking turned off */
642 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
643 /* Assume enough space for every character being bad. 4 spaces each
644 * for the 94 printable characters that are output like "'x' "; and 5
645 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
646 * NUL */
8b7358b9 647 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
648 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
649 to start */
650 unsigned int bad_count = 0; /* Count of bad characters */
651
baa60164 652 for (i = 0; i < 256; i++) {
8b7358b9 653 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
654 if (isupper(i))
655 PL_fold_locale[i] = (U8) tolower(i);
656 else if (islower(i))
657 PL_fold_locale[i] = (U8) toupper(i);
658 else
659 PL_fold_locale[i] = (U8) i;
8b7358b9 660 }
8c6180a9
KW
661
662 /* If checking for locale problems, see if the native ASCII-range
663 * printables plus \n and \t are in their expected categories in
664 * the new locale. If not, this could mean big trouble, upending
665 * Perl's and most programs' assumptions, like having a
666 * metacharacter with special meaning become a \w. Fortunately,
667 * it's very rare to find locales that aren't supersets of ASCII
668 * nowadays. It isn't a problem for most controls to be changed
669 * into something else; we check only \n and \t, though perhaps \r
670 * could be an issue as well. */
7d4bcc4a 671 if ( check_for_problems
8c6180a9
KW
672 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
673 {
8b7358b9
KW
674 bool is_bad = FALSE;
675 char name[3] = { '\0' };
676
677 /* Convert the name into a string */
678 if (isPRINT_A(i)) {
679 name[0] = i;
680 name[1] = '\0';
681 }
682 else if (i == '\n') {
683 my_strlcpy(name, "\n", sizeof(name));
684 }
685 else {
686 my_strlcpy(name, "\t", sizeof(name));
687 }
688
689 /* Check each possibe class */
690 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
691 is_bad = TRUE;
692 DEBUG_L(PerlIO_printf(Perl_debug_log,
693 "isalnum('%s') unexpectedly is %d\n",
694 name, cBOOL(isalnum(i))));
695 }
696 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
697 is_bad = TRUE;
698 DEBUG_L(PerlIO_printf(Perl_debug_log,
699 "isalpha('%s') unexpectedly is %d\n",
700 name, cBOOL(isalpha(i))));
701 }
702 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
703 is_bad = TRUE;
704 DEBUG_L(PerlIO_printf(Perl_debug_log,
705 "isdigit('%s') unexpectedly is %d\n",
706 name, cBOOL(isdigit(i))));
707 }
708 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
709 is_bad = TRUE;
710 DEBUG_L(PerlIO_printf(Perl_debug_log,
711 "isgraph('%s') unexpectedly is %d\n",
712 name, cBOOL(isgraph(i))));
713 }
714 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
715 is_bad = TRUE;
716 DEBUG_L(PerlIO_printf(Perl_debug_log,
717 "islower('%s') unexpectedly is %d\n",
718 name, cBOOL(islower(i))));
719 }
720 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
721 is_bad = TRUE;
722 DEBUG_L(PerlIO_printf(Perl_debug_log,
723 "isprint('%s') unexpectedly is %d\n",
724 name, cBOOL(isprint(i))));
725 }
726 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
727 is_bad = TRUE;
728 DEBUG_L(PerlIO_printf(Perl_debug_log,
729 "ispunct('%s') unexpectedly is %d\n",
730 name, cBOOL(ispunct(i))));
731 }
732 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
733 is_bad = TRUE;
734 DEBUG_L(PerlIO_printf(Perl_debug_log,
735 "isspace('%s') unexpectedly is %d\n",
736 name, cBOOL(isspace(i))));
737 }
738 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
739 is_bad = TRUE;
740 DEBUG_L(PerlIO_printf(Perl_debug_log,
741 "isupper('%s') unexpectedly is %d\n",
742 name, cBOOL(isupper(i))));
743 }
744 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
745 is_bad = TRUE;
746 DEBUG_L(PerlIO_printf(Perl_debug_log,
747 "isxdigit('%s') unexpectedly is %d\n",
748 name, cBOOL(isxdigit(i))));
749 }
750 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
751 is_bad = TRUE;
752 DEBUG_L(PerlIO_printf(Perl_debug_log,
753 "tolower('%s')=0x%x instead of the expected 0x%x\n",
754 name, tolower(i), (int) toLOWER_A(i)));
755 }
756 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
757 is_bad = TRUE;
758 DEBUG_L(PerlIO_printf(Perl_debug_log,
759 "toupper('%s')=0x%x instead of the expected 0x%x\n",
760 name, toupper(i), (int) toUPPER_A(i)));
761 }
762 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
763 is_bad = TRUE;
764 DEBUG_L(PerlIO_printf(Perl_debug_log,
765 "'\\n' (=%02X) is not a control\n", (int) i));
766 }
767
768 /* Add to the list; Separate multiple entries with a blank */
769 if (is_bad) {
770 if (bad_count) {
771 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 772 }
8b7358b9
KW
773 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
774 bad_count++;
8c6180a9
KW
775 }
776 }
777 }
778
7d4bcc4a
KW
779# ifdef MB_CUR_MAX
780
8c6180a9 781 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 782 * this locale requires more than one byte, there are going to be
8c6180a9 783 * problems. */
9c8a6dc2
KW
784 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
785 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
786 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
787
8b7358b9
KW
788 if ( check_for_problems && MB_CUR_MAX > 1
789 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
790
791 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
792 * locale. Just assume that the implementation for them (plus
793 * for POSIX) is correct and the > 1 value is spurious. (Since
794 * these are specially handled to never be considered UTF-8
795 * locales, as long as this is the only problem, everything
796 * should work fine */
797 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
798 {
8c6180a9
KW
799 multi_byte_locale = TRUE;
800 }
7d4bcc4a
KW
801
802# endif
8c6180a9 803
8b7358b9
KW
804 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
805 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
806 PL_warn_locale = Perl_newSVpvf(aTHX_
807 "Locale '%s' contains (at least) the following characters"
808 " which have\nnon-standard meanings: %s\nThe Perl program"
809 " will use the standard meanings",
810 newctype, bad_chars_list);
811
812 }
813 else {
814 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 815 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 816 newctype,
8c6180a9
KW
817 (multi_byte_locale)
818 ? " Some characters in it are not recognized by"
819 " Perl."
820 : "",
821 (bad_count)
822 ? "\nThe following characters (and maybe others)"
823 " may not have the same meaning as the Perl"
824 " program expects:\n"
825 : "",
826 (bad_count)
827 ? bad_chars_list
828 : ""
829 );
8b7358b9
KW
830 }
831
b119c2be
KW
832# ifdef HAS_NL_LANGINFO
833
834 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
835 /* parameter FALSE is a don't care here */
836 my_nl_langinfo(PERL_CODESET, FALSE));
837
838# endif
839
8b7358b9
KW
840 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
841
cc9eaeb0 842 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
843 * output the message now. If not in that scope, we save the
844 * message to be output at the first operation using this locale,
845 * if that actually happens. Most programs don't use locales, so
846 * they are immune to bad ones. */
cc9eaeb0 847 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 848
780fcc9f
KW
849 /* The '0' below suppresses a bogus gcc compiler warning */
850 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 851
bddebb56
KW
852 if (IN_LC(LC_CTYPE)) {
853 SvREFCNT_dec_NN(PL_warn_locale);
854 PL_warn_locale = NULL;
855 }
780fcc9f 856 }
baa60164 857 }
31f05a37 858 }
98994639
HS
859
860#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 861
98994639
HS
862}
863
98994639 864void
2726666d
KW
865Perl__warn_problematic_locale()
866{
2726666d
KW
867
868#ifdef USE_LOCALE_CTYPE
869
5f04a188
KW
870 dTHX;
871
872 /* Internal-to-core function that outputs the message in PL_warn_locale,
873 * and then NULLS it. Should be called only through the macro
874 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
875
2726666d 876 if (PL_warn_locale) {
2726666d
KW
877 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
878 SvPVX(PL_warn_locale),
879 0 /* dummy to avoid compiler warning */ );
2726666d
KW
880 SvREFCNT_dec_NN(PL_warn_locale);
881 PL_warn_locale = NULL;
882 }
883
884#endif
885
886}
887
a4f00dcc
KW
888STATIC void
889S_new_collate(pTHX_ const char *newcoll)
98994639 890{
7d4bcc4a
KW
891
892#ifndef USE_LOCALE_COLLATE
893
894 PERL_UNUSED_ARG(newcoll);
895 PERL_UNUSED_CONTEXT;
896
897#else
0d071d52 898
291a84fb 899 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
900 * core Perl this and that 'newcoll' is the name of the new locale.
901 *
d35fca5f
KW
902 * The design of locale collation is that every locale change is given an
903 * index 'PL_collation_ix'. The first time a string particpates in an
904 * operation that requires collation while locale collation is active, it
905 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
906 * magic includes the collation index, and the transformation of the string
907 * by strxfrm(), q.v. That transformation is used when doing comparisons,
908 * instead of the string itself. If a string changes, the magic is
909 * cleared. The next time the locale changes, the index is incremented,
910 * and so we know during a comparison that the transformation is not
911 * necessarily still valid, and so is recomputed. Note that if the locale
912 * changes enough times, the index could wrap (a U32), and it is possible
913 * that a transformation would improperly be considered valid, leading to
914 * an unlikely bug */
0d071d52 915
98994639
HS
916 if (! newcoll) {
917 if (PL_collation_name) {
918 ++PL_collation_ix;
919 Safefree(PL_collation_name);
920 PL_collation_name = NULL;
921 }
922 PL_collation_standard = TRUE;
00bf60ca 923 is_standard_collation:
98994639
HS
924 PL_collxfrm_base = 0;
925 PL_collxfrm_mult = 2;
165a1c52 926 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 927 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 928 PL_strxfrm_max_cp = 0;
98994639
HS
929 return;
930 }
931
d35fca5f 932 /* If this is not the same locale as currently, set the new one up */
98994639
HS
933 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
934 ++PL_collation_ix;
935 Safefree(PL_collation_name);
936 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 937 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
938 if (PL_collation_standard) {
939 goto is_standard_collation;
940 }
98994639 941
165a1c52 942 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 943 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 944 PL_strxfrm_max_cp = 0;
165a1c52 945
59c018b9
KW
946 /* A locale collation definition includes primary, secondary, tertiary,
947 * etc. weights for each character. To sort, the primary weights are
948 * used, and only if they compare equal, then the secondary weights are
949 * used, and only if they compare equal, then the tertiary, etc.
950 *
951 * strxfrm() works by taking the input string, say ABC, and creating an
952 * output transformed string consisting of first the primary weights,
953 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
954 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
955 * may not have weights at every level. In our example, let's say B
956 * doesn't have a tertiary weight, and A doesn't have a secondary
957 * weight. The constructed string is then going to be
958 * A¹B¹C¹ B²C² A³C³ ....
959 * This has the desired effect that strcmp() will look at the secondary
960 * or tertiary weights only if the strings compare equal at all higher
961 * priority weights. The spaces shown here, like in
c342d20e 962 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
963 * are not just for readability. In the general case, these must
964 * actually be bytes, which we will call here 'separator weights'; and
965 * they must be smaller than any other weight value, but since these
966 * are C strings, only the terminating one can be a NUL (some
967 * implementations may include a non-NUL separator weight just before
968 * the NUL). Implementations tend to reserve 01 for the separator
969 * weights. They are needed so that a shorter string's secondary
970 * weights won't be misconstrued as primary weights of a longer string,
971 * etc. By making them smaller than any other weight, the shorter
972 * string will sort first. (Actually, if all secondary weights are
973 * smaller than all primary ones, there is no need for a separator
974 * weight between those two levels, etc.)
975 *
976 * The length of the transformed string is roughly a linear function of
977 * the input string. It's not exactly linear because some characters
978 * don't have weights at all levels. When we call strxfrm() we have to
979 * allocate some memory to hold the transformed string. The
980 * calculations below try to find coefficients 'm' and 'b' for this
981 * locale so that m*x + b equals how much space we need, given the size
982 * of the input string in 'x'. If we calculate too small, we increase
983 * the size as needed, and call strxfrm() again, but it is better to
984 * get it right the first time to avoid wasted expensive string
985 * transformations. */
986
98994639 987 {
79f120c8
KW
988 /* We use the string below to find how long the tranformation of it
989 * is. Almost all locales are supersets of ASCII, or at least the
990 * ASCII letters. We use all of them, half upper half lower,
991 * because if we used fewer, we might hit just the ones that are
992 * outliers in a particular locale. Most of the strings being
993 * collated will contain a preponderance of letters, and even if
994 * they are above-ASCII, they are likely to have the same number of
995 * weight levels as the ASCII ones. It turns out that digits tend
996 * to have fewer levels, and some punctuation has more, but those
997 * are relatively sparse in text, and khw believes this gives a
998 * reasonable result, but it could be changed if experience so
999 * dictates. */
1000 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1001 char * x_longer; /* Transformed 'longer' */
1002 Size_t x_len_longer; /* Length of 'x_longer' */
1003
1004 char * x_shorter; /* We also transform a substring of 'longer' */
1005 Size_t x_len_shorter;
1006
a4a439fb 1007 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1008 * are interested only in its length). It is used because it has
1009 * the intelligence to handle all cases, but to work, it needs some
1010 * values of 'm' and 'b' to get it started. For the purposes of
1011 * this calculation we use a very conservative estimate of 'm' and
1012 * 'b'. This assumes a weight can be multiple bytes, enough to
1013 * hold any UV on the platform, and there are 5 levels, 4 weight
1014 * bytes, and a trailing NUL. */
1015 PL_collxfrm_base = 5;
1016 PL_collxfrm_mult = 5 * sizeof(UV);
1017
1018 /* Find out how long the transformation really is */
a4a439fb
KW
1019 x_longer = _mem_collxfrm(longer,
1020 sizeof(longer) - 1,
1021 &x_len_longer,
1022
1023 /* We avoid converting to UTF-8 in the
1024 * called function by telling it the
1025 * string is in UTF-8 if the locale is a
1026 * UTF-8 one. Since the string passed
1027 * here is invariant under UTF-8, we can
1028 * claim it's UTF-8 even though it isn't.
1029 * */
1030 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1031 Safefree(x_longer);
1032
1033 /* Find out how long the transformation of a substring of 'longer'
1034 * is. Together the lengths of these transformations are
1035 * sufficient to calculate 'm' and 'b'. The substring is all of
1036 * 'longer' except the first character. This minimizes the chances
1037 * of being swayed by outliers */
a4a439fb 1038 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1039 sizeof(longer) - 2,
a4a439fb
KW
1040 &x_len_shorter,
1041 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1042 Safefree(x_shorter);
1043
1044 /* If the results are nonsensical for this simple test, the whole
1045 * locale definition is suspect. Mark it so that locale collation
1046 * is not active at all for it. XXX Should we warn? */
1047 if ( x_len_shorter == 0
1048 || x_len_longer == 0
1049 || x_len_shorter >= x_len_longer)
1050 {
1051 PL_collxfrm_mult = 0;
1052 PL_collxfrm_base = 0;
1053 }
1054 else {
1055 SSize_t base; /* Temporary */
1056
1057 /* We have both: m * strlen(longer) + b = x_len_longer
1058 * m * strlen(shorter) + b = x_len_shorter;
1059 * subtracting yields:
1060 * m * (strlen(longer) - strlen(shorter))
1061 * = x_len_longer - x_len_shorter
1062 * But we have set things up so that 'shorter' is 1 byte smaller
1063 * than 'longer'. Hence:
1064 * m = x_len_longer - x_len_shorter
1065 *
1066 * But if something went wrong, make sure the multiplier is at
1067 * least 1.
1068 */
1069 if (x_len_longer > x_len_shorter) {
1070 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1071 }
1072 else {
1073 PL_collxfrm_mult = 1;
1074 }
1075
1076 /* mx + b = len
1077 * so: b = len - mx
1078 * but in case something has gone wrong, make sure it is
1079 * non-negative */
1080 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1081 if (base < 0) {
1082 base = 0;
1083 }
1084
1085 /* Add 1 for the trailing NUL */
1086 PL_collxfrm_base = base + 1;
1087 }
58eebef2 1088
7d4bcc4a
KW
1089# ifdef DEBUGGING
1090
58eebef2
KW
1091 if (DEBUG_L_TEST || debug_initialization) {
1092 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1093 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1094 "x_len_longer=%zu,"
1095 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1096 __FILE__, __LINE__,
1097 PL_in_utf8_COLLATE_locale,
1098 x_len_shorter, x_len_longer,
1099 PL_collxfrm_mult, PL_collxfrm_base);
1100 }
7d4bcc4a
KW
1101# endif
1102
98994639
HS
1103 }
1104 }
1105
1106#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1107
98994639
HS
1108}
1109
d2b24094 1110#ifdef WIN32
b8cc575c 1111
a4f00dcc 1112STATIC char *
b8cc575c 1113S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1114{
1115 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1116 * difference between the two unless the input locale is "", which normally
1117 * means on Windows to get the machine default, which is set via the
1118 * computer's "Regional and Language Options" (or its current equivalent).
1119 * In POSIX, it instead means to find the locale from the user's
1120 * environment. This routine changes the Windows behavior to first look in
1121 * the environment, and, if anything is found, use that instead of going to
1122 * the machine default. If there is no environment override, the machine
1123 * default is used, by calling the real setlocale() with "".
1124 *
1125 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1126 * use the particular category's variable if set; otherwise to use the LANG
1127 * variable. */
b385bb4d 1128
175c4cf9 1129 bool override_LC_ALL = FALSE;
89f7b9aa 1130 char * result;
e5f10d49 1131 unsigned int i;
89f7b9aa 1132
b385bb4d 1133 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1134
1135# ifdef LC_ALL
1136
b385bb4d
KW
1137 locale = PerlEnv_getenv("LC_ALL");
1138 if (! locale) {
e5f10d49
KW
1139 if (category == LC_ALL) {
1140 override_LC_ALL = TRUE;
1141 }
1142 else {
7d4bcc4a
KW
1143
1144# endif
7d4bcc4a 1145
e5f10d49
KW
1146 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1147 if (category == categories[i]) {
1148 locale = PerlEnv_getenv(category_names[i]);
1149 goto found_locale;
1150 }
1151 }
7d4bcc4a 1152
b385bb4d 1153 locale = PerlEnv_getenv("LANG");
481465ea 1154 if (! locale) {
b385bb4d
KW
1155 locale = "";
1156 }
e5f10d49
KW
1157
1158 found_locale: ;
7d4bcc4a
KW
1159
1160# ifdef LC_ALL
1161
e5f10d49 1162 }
b385bb4d 1163 }
7d4bcc4a
KW
1164
1165# endif
1166
b385bb4d
KW
1167 }
1168
89f7b9aa 1169 result = setlocale(category, locale);
48015184
KW
1170 DEBUG_L(STMT_START {
1171 dSAVE_ERRNO;
1172 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
1173 setlocale_debug_string(category, locale, result));
1174 RESTORE_ERRNO;
1175 } STMT_END);
89f7b9aa 1176
481465ea 1177 if (! override_LC_ALL) {
89f7b9aa
KW
1178 return result;
1179 }
1180
dfd77d7a 1181 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
1182 * LANG variable or the system default if there is no LANG. But these have
1183 * lower priority than the other LC_foo variables, so override it for each
1184 * one that is set. (If they are set to "", it means to use the same thing
1185 * we just set LC_ALL to, so can skip) */
7d4bcc4a 1186
948523db 1187 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
1188 result = PerlEnv_getenv(category_names[i]);
1189 if (result && strNE(result, "")) {
1190 setlocale(categories[i], result);
1191 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1192 __FILE__, __LINE__,
1193 setlocale_debug_string(categories[i], result, "not captured")));
1194 }
89f7b9aa 1195 }
7d4bcc4a 1196
bbc98134 1197 result = setlocale(LC_ALL, NULL);
48015184
KW
1198 DEBUG_L(STMT_START {
1199 dSAVE_ERRNO;
1200 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 1201 __FILE__, __LINE__,
48015184
KW
1202 setlocale_debug_string(LC_ALL, NULL, result));
1203 RESTORE_ERRNO;
1204 } STMT_END);
89f7b9aa 1205
bbc98134 1206 return result;
b385bb4d
KW
1207}
1208
1209#endif
1210
a4f00dcc
KW
1211char *
1212Perl_setlocale(int category, const char * locale)
1213{
1214 /* This wraps POSIX::setlocale() */
1215
1216 char * retval;
1159483a 1217 char * newlocale;
a4f00dcc
KW
1218 dTHX;
1219
a4f00dcc
KW
1220#ifdef USE_LOCALE_NUMERIC
1221
291a84fb
KW
1222 /* A NULL locale means only query what the current one is. We have the
1223 * LC_NUMERIC name saved, because we are normally switched into the C
1224 * locale for it. For an LC_ALL query, switch back to get the correct
1225 * results. All other categories don't require special handling */
a4f00dcc
KW
1226 if (locale == NULL) {
1227 if (category == LC_NUMERIC) {
1228 return savepv(PL_numeric_name);
1229 }
1230
7d4bcc4a 1231# ifdef LC_ALL
a4f00dcc 1232
c0d737a8
KW
1233 else if (category == LC_ALL && ! PL_numeric_underlying) {
1234
a4f00dcc
KW
1235 SET_NUMERIC_UNDERLYING();
1236 }
1237
7d4bcc4a 1238# endif
a4f00dcc
KW
1239
1240 }
1241
1242#endif
1243
e5f10d49
KW
1244 /* Save retval since subsequent setlocale() calls may overwrite it. */
1245 retval = savepv(do_setlocale_r(category, locale));
a4f00dcc
KW
1246
1247 DEBUG_L(PerlIO_printf(Perl_debug_log,
1248 "%s:%d: %s\n", __FILE__, __LINE__,
1249 setlocale_debug_string(category, locale, retval)));
1250 if (! retval) {
1251 /* Should never happen that a query would return an error, but be
1252 * sure and reset to C locale */
1253 if (locale == 0) {
1254 SET_NUMERIC_STANDARD();
1255 }
7d4bcc4a 1256
a4f00dcc
KW
1257 return NULL;
1258 }
1259
a4f00dcc
KW
1260 /* If locale == NULL, we are just querying the state, but may have switched
1261 * to NUMERIC_UNDERLYING. Switch back before returning. */
1262 if (locale == NULL) {
1263 SET_NUMERIC_STANDARD();
1264 return retval;
1265 }
a4f00dcc 1266
1159483a
KW
1267 /* Now that have switched locales, we have to update our records to
1268 * correspond. */
a4f00dcc 1269
1159483a 1270 switch (category) {
a4f00dcc 1271
1159483a 1272#ifdef USE_LOCALE_CTYPE
a4f00dcc 1273
1159483a
KW
1274 case LC_CTYPE:
1275 new_ctype(retval);
1276 break;
a4f00dcc 1277
1159483a 1278#endif
a4f00dcc
KW
1279#ifdef USE_LOCALE_COLLATE
1280
1159483a
KW
1281 case LC_COLLATE:
1282 new_collate(retval);
1283 break;
a4f00dcc 1284
1159483a 1285#endif
a4f00dcc
KW
1286#ifdef USE_LOCALE_NUMERIC
1287
1159483a
KW
1288 case LC_NUMERIC:
1289 new_numeric(retval);
1290 break;
a4f00dcc 1291
1159483a
KW
1292#endif
1293#ifdef LC_ALL
a4f00dcc 1294
1159483a 1295 case LC_ALL:
a4f00dcc 1296
1159483a
KW
1297 /* LC_ALL updates all the things we care about. The values may not
1298 * be the same as 'retval', as the locale "" may have set things
1299 * individually */
a4f00dcc 1300
1159483a 1301# ifdef USE_LOCALE_CTYPE
a4f00dcc 1302
1159483a
KW
1303 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1304 new_ctype(newlocale);
a4f00dcc 1305
1159483a
KW
1306# endif /* USE_LOCALE_CTYPE */
1307# ifdef USE_LOCALE_COLLATE
1308
1309 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1310 new_collate(newlocale);
a4f00dcc 1311
7d4bcc4a 1312# endif
1159483a 1313# ifdef USE_LOCALE_NUMERIC
a4f00dcc 1314
1159483a
KW
1315 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1316 new_numeric(newlocale);
a4f00dcc 1317
1159483a
KW
1318# endif /* USE_LOCALE_NUMERIC */
1319#endif /* LC_ALL */
a4f00dcc 1320
1159483a
KW
1321 default:
1322 break;
a4f00dcc
KW
1323 }
1324
1325 return retval;
1326
f7416781
KW
1327
1328}
1329
1330PERL_STATIC_INLINE const char *
1331S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1332{
1333 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1334 * growing it if necessary */
1335
1336 const Size_t string_size = strlen(string) + offset + 1;
1337
1338 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1339
1340 if (*buf_size == 0) {
1341 Newx(*buf, string_size, char);
1342 *buf_size = string_size;
1343 }
1344 else if (string_size > *buf_size) {
1345 Renew(*buf, string_size, char);
1346 *buf_size = string_size;
1347 }
1348
1349 Copy(string, *buf + offset, string_size - offset, char);
1350 return *buf;
1351}
1352
1353/*
1354
1355=head1 Locale-related functions and macros
1356
1357=for apidoc Perl_langinfo
1358
7d4bcc4a 1359This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
1360taking the same C<item> parameter values, and returning the same information.
1361But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1362of Perl's locale handling from your code, and can be used on systems that lack
1363a native C<nl_langinfo>.
1364
1365Expanding on these:
1366
1367=over
1368
1369=item *
1370
1371It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1372without you having to write extra code. The reason for the extra code would be
1373because these are from the C<LC_NUMERIC> locale category, which is normally
1374kept set to the C locale by Perl, no matter what the underlying locale is
1375supposed to be, and so to get the expected results, you have to temporarily
1376toggle into the underlying locale, and later toggle back. (You could use
1377plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1378but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1379keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1380expecting the radix (decimal point) character to be a dot.)
1381
1382=item *
1383
1384Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1385makes your code more portable. Of the fifty-some possible items specified by
1386the POSIX 2008 standard,
1387L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1388only two are completely unimplemented. It uses various techniques to recover
1389the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1390both of which are specified in C89, so should be always be available. Later
1391C<strftime()> versions have additional capabilities; C<""> is returned for
1392those not available on your system.
1393
9f0bc911
KW
1394It is important to note that on such systems, this calls C<localeconv>, and so
1395overwrites the static buffer returned from previous explicit calls to that
1396function. Thus, if the program doesn't use or save the information from an
1397explicit C<localeconv> call (which good practice suggests should be done
1398anyway), use of this function can break it.
1399
f7416781
KW
1400The details for those items which may differ from what this emulation returns
1401and what a native C<nl_langinfo()> would return are:
1402
1403=over
1404
1405=item C<CODESET>
1406
1407=item C<ERA>
1408
1409Unimplemented, so returns C<"">.
1410
1411=item C<YESEXPR>
1412
c1566110
KW
1413=item C<YESSTR>
1414
f7416781
KW
1415=item C<NOEXPR>
1416
c1566110
KW
1417=item C<NOSTR>
1418
1419Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
1420removed from POSIX 2008, and are retained for backwards compatibility. Your
1421platform's C<nl_langinfo> may not support them.
f7416781
KW
1422
1423=item C<D_FMT>
1424
1425Always evaluates to C<%x>, the locale's appropriate date representation.
1426
1427=item C<T_FMT>
1428
1429Always evaluates to C<%X>, the locale's appropriate time representation.
1430
1431=item C<D_T_FMT>
1432
1433Always evaluates to C<%c>, the locale's appropriate date and time
1434representation.
1435
1436=item C<CRNCYSTR>
1437
1438The return may be incorrect for those rare locales where the currency symbol
1439replaces the radix character.
1440Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1441to work differently.
1442
1443=item C<ALT_DIGITS>
1444
1445Currently this gives the same results as Linux does.
1446Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1447to work differently.
1448
1449=item C<ERA_D_FMT>
1450
1451=item C<ERA_T_FMT>
1452
1453=item C<ERA_D_T_FMT>
1454
1455=item C<T_FMT_AMPM>
1456
1457These are derived by using C<strftime()>, and not all versions of that function
1458know about them. C<""> is returned for these on such systems.
1459
1460=back
1461
1462When using C<Perl_langinfo> on systems that don't have a native
1463C<nl_langinfo()>, you must
1464
1465 #include "perl_langinfo.h"
1466
1467before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1468C<#include> with this one. (Doing it this way keeps out the symbols that plain
1469C<langinfo.h> imports into the namespace for code that doesn't need it.)
1470
1471You also should not use the bare C<langinfo.h> item names, but should preface
1472them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1473The C<PERL_I<foo>> versions will also work for this function on systems that do
1474have a native C<nl_langinfo>.
1475
1476=item *
1477
1478It is thread-friendly, returning its result in a buffer that won't be
1479overwritten by another thread, so you don't have to code for that possibility.
1480The buffer can be overwritten by the next call to C<nl_langinfo> or
1481C<Perl_langinfo> in the same thread.
1482
1483=item *
1484
7d4bcc4a 1485ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
1486*>>, but you are (only by documentation) forbidden to write into the buffer.
1487By declaring this C<const>, the compiler enforces this restriction. The extra
1488C<const> is why this isn't an unequivocal drop-in replacement for
1489C<nl_langinfo>.
1490
1491=back
1492
1493The original impetus for C<Perl_langinfo()> was so that code that needs to
1494find out the current currency symbol, floating point radix character, or digit
1495grouping separator can use, on all systems, the simpler and more
1496thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1497pain to make thread-friendly. For other fields returned by C<localeconv>, it
1498is better to use the methods given in L<perlcall> to call
1499L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1500
1501=cut
1502
1503*/
1504
1505const char *
1506#ifdef HAS_NL_LANGINFO
1507Perl_langinfo(const nl_item item)
1508#else
1509Perl_langinfo(const int item)
1510#endif
1511{
f61748ac
KW
1512 return my_nl_langinfo(item, TRUE);
1513}
1514
1515const char *
1516#ifdef HAS_NL_LANGINFO
1517S_my_nl_langinfo(const nl_item item, bool toggle)
1518#else
1519S_my_nl_langinfo(const int item, bool toggle)
1520#endif
1521{
ae74815b 1522 dTHX;
f7416781 1523
5a854ab3
KW
1524 /* We only need to toggle into the underlying LC_NUMERIC locale for these
1525 * two items, and only if not already there */
1526 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
1527 || PL_numeric_underlying))
1528 {
1529 toggle = FALSE;
1530 }
1531
ab340fff
KW
1532#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1533#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1534
ab340fff 1535 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1536 * for those items dependent on it. This must be copied to a buffer before
1537 * switching back, as some systems destroy the buffer when setlocale() is
1538 * called */
f7416781 1539
038d3702
KW
1540 {
1541 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
1542
b2fee59c 1543 if (toggle) {
038d3702 1544 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 1545 }
f7416781 1546
5acc4454
KW
1547 LOCALE_LOCK; /* Prevent interference from another thread executing
1548 this code section (the only call to nl_langinfo in
1549 the core) */
1550
b2fee59c
KW
1551 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf,
1552 &PL_langinfo_bufsize, 0);
f7416781 1553
5acc4454
KW
1554 LOCALE_UNLOCK;
1555
b2fee59c 1556 if (toggle) {
038d3702 1557 RESTORE_LC_NUMERIC();
b2fee59c 1558 }
f7416781
KW
1559 }
1560
ab340fff
KW
1561# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1562
5a854ab3 1563 {
b2fee59c
KW
1564 bool do_free = FALSE;
1565 locale_t cur = uselocale((locale_t) 0);
ab340fff 1566
b2fee59c
KW
1567 if (cur == LC_GLOBAL_LOCALE) {
1568 cur = duplocale(LC_GLOBAL_LOCALE);
1569 do_free = TRUE;
1570 }
ab340fff 1571
b2fee59c 1572 if (toggle) {
e1aa2579
KW
1573 if (PL_underlying_numeric_obj) {
1574 cur = PL_underlying_numeric_obj;
1575 }
1576 else {
1577 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1578 do_free = TRUE;
1579 }
b2fee59c 1580 }
ab340fff 1581
b2fee59c
KW
1582 save_to_buffer(nl_langinfo_l(item, cur),
1583 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1584 if (do_free) {
1585 freelocale(cur);
1586 }
5a854ab3 1587 }
ab340fff 1588
c1566110
KW
1589# endif
1590
1591 if (strEQ(PL_langinfo_buf, "")) {
1592 if (item == PERL_YESSTR) {
1593 return "yes";
1594 }
1595 if (item == PERL_NOSTR) {
1596 return "no";
1597 }
1598 }
1599
ab340fff
KW
1600 return PL_langinfo_buf;
1601
f7416781 1602#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
1603
1604 {
1605
f7416781
KW
1606# ifdef HAS_LOCALECONV
1607
43dd6b15 1608 const struct lconv* lc;
038d3702 1609 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
1610
1611# endif
1612# ifdef HAS_STRFTIME
1613
43dd6b15
KW
1614 struct tm tm;
1615 bool return_format = FALSE; /* Return the %format, not the value */
1616 const char * format;
f7416781
KW
1617
1618# endif
1619
43dd6b15
KW
1620 /* We copy the results to a per-thread buffer, even if not
1621 * multi-threaded. This is in part to simplify this code, and partly
1622 * because we need a buffer anyway for strftime(), and partly because a
1623 * call of localeconv() could otherwise wipe out the buffer, and the
1624 * programmer would not be expecting this, as this is a nl_langinfo()
1625 * substitute after all, so s/he might be thinking their localeconv()
1626 * is safe until another localeconv() call. */
f7416781 1627
43dd6b15
KW
1628 switch (item) {
1629 Size_t len;
1630 const char * retval;
f7416781 1631
43dd6b15
KW
1632 /* These 2 are unimplemented */
1633 case PERL_CODESET:
1634 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 1635
43dd6b15
KW
1636 default:
1637 return "";
f7416781 1638
43dd6b15
KW
1639 /* We use only an English set, since we don't know any more */
1640 case PERL_YESEXPR: return "^[+1yY]";
1641 case PERL_YESSTR: return "yes";
1642 case PERL_NOEXPR: return "^[-0nN]";
1643 case PERL_NOSTR: return "no";
f7416781
KW
1644
1645# ifdef HAS_LOCALECONV
1646
43dd6b15 1647 case PERL_CRNCYSTR:
f7416781 1648
43dd6b15
KW
1649 /* We don't bother with localeconv_l() because any system that
1650 * has it is likely to also have nl_langinfo() */
291a84fb 1651
5acc4454
KW
1652 LOCALE_LOCK; /* Prevent interference with other threads
1653 using localeconv() */
1654
43dd6b15
KW
1655 lc = localeconv();
1656 if ( ! lc
1657 || ! lc->currency_symbol
1658 || strEQ("", lc->currency_symbol))
1659 {
1660 LOCALE_UNLOCK;
1661 return "";
1662 }
f7416781 1663
43dd6b15
KW
1664 /* Leave the first spot empty to be filled in below */
1665 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1666 &PL_langinfo_bufsize, 1);
1667 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1668 { /* khw couldn't figure out how the localedef specifications
1669 would show that the $ should replace the radix; this is
1670 just a guess as to how it might work.*/
1671 *PL_langinfo_buf = '.';
1672 }
1673 else if (lc->p_cs_precedes) {
1674 *PL_langinfo_buf = '-';
1675 }
1676 else {
1677 *PL_langinfo_buf = '+';
1678 }
f7416781 1679
43dd6b15
KW
1680 LOCALE_UNLOCK;
1681 break;
f7416781 1682
43dd6b15
KW
1683 case PERL_RADIXCHAR:
1684 case PERL_THOUSEP:
f7416781 1685
43dd6b15 1686 if (toggle) {
038d3702 1687 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 1688 }
f7416781 1689
5acc4454
KW
1690 LOCALE_LOCK; /* Prevent interference with other threads
1691 using localeconv() */
1692
43dd6b15
KW
1693 lc = localeconv();
1694 if (! lc) {
33394adc
KW
1695 retval = "";
1696 }
43dd6b15
KW
1697 else {
1698 retval = (item == PERL_RADIXCHAR)
1699 ? lc->decimal_point
1700 : lc->thousands_sep;
1701 if (! retval) {
1702 retval = "";
1703 }
1704 }
f7416781 1705
43dd6b15
KW
1706 save_to_buffer(retval, &PL_langinfo_buf,
1707 &PL_langinfo_bufsize, 0);
f7416781 1708
5acc4454
KW
1709 LOCALE_UNLOCK;
1710
43dd6b15 1711 if (toggle) {
038d3702 1712 RESTORE_LC_NUMERIC();
43dd6b15 1713 }
f7416781 1714
43dd6b15 1715 break;
f7416781
KW
1716
1717# endif
1718# ifdef HAS_STRFTIME
1719
43dd6b15
KW
1720 /* These are defined by C89, so we assume that strftime supports
1721 * them, and so are returned unconditionally; they may not be what
1722 * the locale actually says, but should give good enough results
1723 * for someone using them as formats (as opposed to trying to parse
1724 * them to figure out what the locale says). The other format
1725 * items are actually tested to verify they work on the platform */
1726 case PERL_D_FMT: return "%x";
1727 case PERL_T_FMT: return "%X";
1728 case PERL_D_T_FMT: return "%c";
1729
1730 /* These formats are only available in later strfmtime's */
1731 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1732 case PERL_T_FMT_AMPM:
1733
1734 /* The rest can be gotten from most versions of strftime(). */
1735 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1736 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1737 case PERL_ABDAY_7:
1738 case PERL_ALT_DIGITS:
1739 case PERL_AM_STR: case PERL_PM_STR:
1740 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1741 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1742 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1743 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1744 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1745 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1746 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1747 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1748 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
1749 case PERL_MON_12:
1750
1751 LOCALE_LOCK;
1752
1753 init_tm(&tm); /* Precaution against core dumps */
1754 tm.tm_sec = 30;
1755 tm.tm_min = 30;
1756 tm.tm_hour = 6;
1757 tm.tm_year = 2017 - 1900;
1758 tm.tm_wday = 0;
1759 tm.tm_mon = 0;
1760 switch (item) {
1761 default:
1762 LOCALE_UNLOCK;
1763 Perl_croak(aTHX_
1764 "panic: %s: %d: switch case: %d problem",
1765 __FILE__, __LINE__, item);
1766 NOT_REACHED; /* NOTREACHED */
1767
1768 case PERL_PM_STR: tm.tm_hour = 18;
1769 case PERL_AM_STR:
1770 format = "%p";
1771 break;
1772
1773 case PERL_ABDAY_7: tm.tm_wday++;
1774 case PERL_ABDAY_6: tm.tm_wday++;
1775 case PERL_ABDAY_5: tm.tm_wday++;
1776 case PERL_ABDAY_4: tm.tm_wday++;
1777 case PERL_ABDAY_3: tm.tm_wday++;
1778 case PERL_ABDAY_2: tm.tm_wday++;
1779 case PERL_ABDAY_1:
1780 format = "%a";
1781 break;
1782
1783 case PERL_DAY_7: tm.tm_wday++;
1784 case PERL_DAY_6: tm.tm_wday++;
1785 case PERL_DAY_5: tm.tm_wday++;
1786 case PERL_DAY_4: tm.tm_wday++;
1787 case PERL_DAY_3: tm.tm_wday++;
1788 case PERL_DAY_2: tm.tm_wday++;
1789 case PERL_DAY_1:
1790 format = "%A";
1791 break;
1792
1793 case PERL_ABMON_12: tm.tm_mon++;
1794 case PERL_ABMON_11: tm.tm_mon++;
1795 case PERL_ABMON_10: tm.tm_mon++;
1796 case PERL_ABMON_9: tm.tm_mon++;
1797 case PERL_ABMON_8: tm.tm_mon++;
1798 case PERL_ABMON_7: tm.tm_mon++;
1799 case PERL_ABMON_6: tm.tm_mon++;
1800 case PERL_ABMON_5: tm.tm_mon++;
1801 case PERL_ABMON_4: tm.tm_mon++;
1802 case PERL_ABMON_3: tm.tm_mon++;
1803 case PERL_ABMON_2: tm.tm_mon++;
1804 case PERL_ABMON_1:
1805 format = "%b";
1806 break;
1807
1808 case PERL_MON_12: tm.tm_mon++;
1809 case PERL_MON_11: tm.tm_mon++;
1810 case PERL_MON_10: tm.tm_mon++;
1811 case PERL_MON_9: tm.tm_mon++;
1812 case PERL_MON_8: tm.tm_mon++;
1813 case PERL_MON_7: tm.tm_mon++;
1814 case PERL_MON_6: tm.tm_mon++;
1815 case PERL_MON_5: tm.tm_mon++;
1816 case PERL_MON_4: tm.tm_mon++;
1817 case PERL_MON_3: tm.tm_mon++;
1818 case PERL_MON_2: tm.tm_mon++;
1819 case PERL_MON_1:
1820 format = "%B";
1821 break;
1822
1823 case PERL_T_FMT_AMPM:
1824 format = "%r";
1825 return_format = TRUE;
1826 break;
1827
1828 case PERL_ERA_D_FMT:
1829 format = "%Ex";
1830 return_format = TRUE;
1831 break;
1832
1833 case PERL_ERA_T_FMT:
1834 format = "%EX";
1835 return_format = TRUE;
1836 break;
1837
1838 case PERL_ERA_D_T_FMT:
1839 format = "%Ec";
1840 return_format = TRUE;
1841 break;
1842
1843 case PERL_ALT_DIGITS:
1844 tm.tm_wday = 0;
1845 format = "%Ow"; /* Find the alternate digit for 0 */
1846 break;
1847 }
f7416781 1848
43dd6b15
KW
1849 /* We can't use my_strftime() because it doesn't look at
1850 * tm_wday */
1851 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1852 format, &tm))
1853 {
1854 /* A zero return means one of:
1855 * a) there wasn't enough space in PL_langinfo_buf
1856 * b) the format, like a plain %p, returns empty
1857 * c) it was an illegal format, though some
1858 * implementations of strftime will just return the
1859 * illegal format as a plain character sequence.
1860 *
1861 * To quickly test for case 'b)', try again but precede
1862 * the format with a plain character. If that result is
1863 * still empty, the problem is either 'a)' or 'c)' */
1864
1865 Size_t format_size = strlen(format) + 1;
1866 Size_t mod_size = format_size + 1;
1867 char * mod_format;
1868 char * temp_result;
1869
1870 Newx(mod_format, mod_size, char);
1871 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 1872 *mod_format = ' ';
43dd6b15
KW
1873 my_strlcpy(mod_format + 1, format, mod_size);
1874 len = strftime(temp_result,
1875 PL_langinfo_bufsize,
1876 mod_format, &tm);
1877 Safefree(mod_format);
1878 Safefree(temp_result);
1879
1880 /* If 'len' is non-zero, it means that we had a case like
1881 * %p which means the current locale doesn't use a.m. or
1882 * p.m., and that is valid */
1883 if (len == 0) {
1884
1885 /* Here, still didn't work. If we get well beyond a
1886 * reasonable size, bail out to prevent an infinite
1887 * loop. */
1888
1889 if (PL_langinfo_bufsize > 100 * format_size) {
1890 *PL_langinfo_buf = '\0';
1891 }
1892 else {
1893 /* Double the buffer size to retry; Add 1 in case
1894 * original was 0, so we aren't stuck at 0. */
1895 PL_langinfo_bufsize *= 2;
1896 PL_langinfo_bufsize++;
1897 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1898 continue;
1899 }
1900 }
f7416781 1901
f7416781 1902 break;
43dd6b15 1903 }
f7416781 1904
43dd6b15
KW
1905 /* Here, we got a result.
1906 *
1907 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1908 * alternate format for wday 0. If the value is the same as
1909 * the normal 0, there isn't an alternate, so clear the buffer.
1910 * */
1911 if ( item == PERL_ALT_DIGITS
1912 && strEQ(PL_langinfo_buf, "0"))
1913 {
1914 *PL_langinfo_buf = '\0';
1915 }
f7416781 1916
43dd6b15
KW
1917 /* ALT_DIGITS is problematic. Experiments on it showed that
1918 * strftime() did not always work properly when going from
1919 * alt-9 to alt-10. Only a few locales have this item defined,
1920 * and in all of them on Linux that khw was able to find,
1921 * nl_langinfo() merely returned the alt-0 character, possibly
1922 * doubled. Most Unicode digits are in blocks of 10
1923 * consecutive code points, so that is sufficient information
1924 * for those scripts, as we can infer alt-1, alt-2, .... But
1925 * for a Japanese locale, a CJK ideographic 0 is returned, and
1926 * the CJK digits are not in code point order, so you can't
1927 * really infer anything. The localedef for this locale did
1928 * specify the succeeding digits, so that strftime() works
1929 * properly on them, without needing to infer anything. But
1930 * the nl_langinfo() return did not give sufficient information
1931 * for the caller to understand what's going on. So until
1932 * there is evidence that it should work differently, this
1933 * returns the alt-0 string for ALT_DIGITS.
1934 *
1935 * wday was chosen because its range is all a single digit.
1936 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 1937
43dd6b15 1938 LOCALE_UNLOCK;
f7416781 1939
43dd6b15
KW
1940 /* If to return the format, not the value, overwrite the buffer
1941 * with it. But some strftime()s will keep the original format
1942 * if illegal, so change those to "" */
1943 if (return_format) {
1944 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
1945 *PL_langinfo_buf = '\0';
1946 }
43dd6b15
KW
1947 else {
1948 save_to_buffer(format, &PL_langinfo_buf,
1949 &PL_langinfo_bufsize, 0);
f7416781
KW
1950 }
1951 }
1952
1953 break;
f7416781
KW
1954
1955# endif
1956
43dd6b15 1957 }
f7416781
KW
1958 }
1959
1960 return PL_langinfo_buf;
1961
1962#endif
1963
a4f00dcc 1964}
b385bb4d 1965
98994639
HS
1966/*
1967 * Initialize locale awareness.
1968 */
1969int
1970Perl_init_i18nl10n(pTHX_ int printwarn)
1971{
0e92a118
KW
1972 /* printwarn is
1973 *
1974 * 0 if not to output warning when setup locale is bad
1975 * 1 if to output warning based on value of PERL_BADLANG
1976 * >1 if to output regardless of PERL_BADLANG
1977 *
1978 * returns
98994639 1979 * 1 = set ok or not applicable,
0e92a118
KW
1980 * 0 = fallback to a locale of lower priority
1981 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1982 *
1983 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1984 * set, debugging information is output.
1985 *
1986 * This looks more complicated than it is, mainly due to the #ifdefs.
1987 *
1988 * We try to set LC_ALL to the value determined by the environment. If
1989 * there is no LC_ALL on this platform, we try the individual categories we
1990 * know about. If this works, we are done.
1991 *
1992 * But if it doesn't work, we have to do something else. We search the
1993 * environment variables ourselves instead of relying on the system to do
1994 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1995 * think there is one), and the ultimate fallback "C". This is all done in
1996 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
1997 * more complex. The 'trial_locales' array is initialized with just one
1998 * element; it causes the behavior described in the paragraph above this to
1999 * happen. If that fails, we add elements to 'trial_locales', and do extra
2000 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2001 *
2002 * On Ultrix, the locale MUST come from the environment, so there is
2003 * preliminary code to set it. I (khw) am not sure that it is necessary,
2004 * and that this couldn't be folded into the loop, but barring any real
2005 * platforms to test on, it's staying as-is
2006 *
2007 * A slight complication is that in embedded Perls, the locale may already
2008 * be set-up, and we don't want to get it from the normal environment
2009 * variables. This is handled by having a special environment variable
2010 * indicate we're in this situation. We simply set setlocale's 2nd
2011 * parameter to be a NULL instead of "". That indicates to setlocale that
2012 * it is not to change anything, but to return the current value,
2013 * effectively initializing perl's db to what the locale already is.
2014 *
2015 * We play the same trick with NULL if a LC_ALL succeeds. We call
2016 * setlocale() on the individual categores with NULL to get their existing
2017 * values for our db, instead of trying to change them.
2018 * */
98994639 2019
0e92a118
KW
2020 int ok = 1;
2021
7d4bcc4a
KW
2022#ifndef USE_LOCALE
2023
2024 PERL_UNUSED_ARG(printwarn);
2025
2026#else /* USE_LOCALE */
7d4bcc4a
KW
2027# ifdef __GLIBC__
2028
175c4cf9 2029 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2030
2031# endif
65ebb059 2032
ccd65d51
KW
2033 /* NULL uses the existing already set up locale */
2034 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2035 ? NULL
2036 : "";
c3fcd832
KW
2037 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2038 unsigned int trial_locales_count;
175c4cf9
KW
2039 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2040 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2041 bool setlocale_failure = FALSE;
65ebb059 2042 unsigned int i;
175c4cf9
KW
2043
2044 /* A later getenv() could zap this, so only use here */
2045 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2046
2047 const bool locwarn = (printwarn > 1
e5f10d49
KW
2048 || ( printwarn
2049 && ( ! bad_lang_use_once
22ff3130 2050 || (
e5f10d49
KW
2051 /* disallow with "" or "0" */
2052 *bad_lang_use_once
2053 && strNE("0", bad_lang_use_once)))));
ea92aad8 2054
291a84fb 2055 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2056 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2057
2058 /* current locale for given category; should have been copied so aren't
2059 * volatile */
8de4332b 2060 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2061
7d4bcc4a
KW
2062# ifdef WIN32
2063
6bce99ee
JH
2064 /* In some systems you can find out the system default locale
2065 * and use that as the fallback locale. */
7d4bcc4a
KW
2066# define SYSTEM_DEFAULT_LOCALE
2067# endif
2068# ifdef SYSTEM_DEFAULT_LOCALE
2069
65ebb059 2070 const char *system_default_locale = NULL;
98994639 2071
7d4bcc4a 2072# endif
948523db
KW
2073
2074# ifndef DEBUGGING
2075# define DEBUG_LOCALE_INIT(a,b,c)
2076# else
7d4bcc4a 2077
8298454c 2078 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2079
2080# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2081 STMT_START { \
2082 if (debug_initialization) { \
2083 PerlIO_printf(Perl_debug_log, \
2084 "%s:%d: %s\n", \
2085 __FILE__, __LINE__, \
a4f00dcc 2086 setlocale_debug_string(category, \
2fcc0ca9
KW
2087 locale, \
2088 result)); \
2089 } \
2090 } STMT_END
2fcc0ca9 2091
948523db
KW
2092/* Make sure the parallel arrays are properly set up */
2093# ifdef USE_LOCALE_NUMERIC
2094 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2095 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
2096# endif
2097# ifdef USE_LOCALE_CTYPE
2098 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2099 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
2100# endif
2101# ifdef USE_LOCALE_COLLATE
2102 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2103 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
2104# endif
2105# ifdef USE_LOCALE_TIME
2106 assert(categories[LC_TIME_INDEX] == LC_TIME);
2107 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2108# endif
2109# ifdef USE_LOCALE_MESSAGES
2110 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2111 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
2112# endif
2113# ifdef USE_LOCALE_MONETARY
2114 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2115 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
2116# endif
9821811f
KW
2117# ifdef USE_LOCALE_ADDRESS
2118 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
2119 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
2120# endif
2121# ifdef USE_LOCALE_IDENTIFICATION
2122 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
2123 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
2124# endif
2125# ifdef USE_LOCALE_MEASUREMENT
2126 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
2127 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
2128# endif
2129# ifdef USE_LOCALE_PAPER
2130 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
2131 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
2132# endif
2133# ifdef USE_LOCALE_TELEPHONE
2134 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
2135 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
2136# endif
948523db
KW
2137# ifdef LC_ALL
2138 assert(categories[LC_ALL_INDEX] == LC_ALL);
2139 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
2140 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
2141# endif
2142# endif /* DEBUGGING */
47280b20
KW
2143
2144 /* Initialize the cache of the program's UTF-8ness for the always known
2145 * locales C and POSIX */
2146 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
2147 sizeof(PL_locale_utf8ness));
2148
3ca88433
KW
2149 PL_numeric_radix_sv = newSVpvs(".");
2150
b56c4436 2151# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
2152
2153 /*
2154 * Ultrix setlocale(..., "") fails if there are no environment
2155 * variables from which to get a locale name.
2156 */
2157
b56c4436
KW
2158# ifndef LC_ALL
2159# error Ultrix without LC_ALL not implemented
2160# else
7d4bcc4a 2161
b56c4436
KW
2162 {
2163 bool done = FALSE;
ea92aad8
KW
2164 if (lang) {
2165 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
2166 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
2167 if (sl_result[LC_ALL_INDEX])
2168 done = TRUE;
2169 else
e5f10d49 2170 setlocale_failure = TRUE;
ea92aad8
KW
2171 }
2172 if (! setlocale_failure) {
2173 const char * locale_param;
2174 for (i = 0; i < LC_ALL_INDEX; i++) {
2175 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
2176 ? setlocale_init
2177 : NULL;
2178 sl_result[i] = do_setlocale_r(categories[i], locale_param);
2179 if (! sl_result[i]) {
2180 setlocale_failure = TRUE;
2181 }
2182 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 2183 }
c835d6be 2184 }
7d4bcc4a
KW
2185 }
2186
2187# endif /* LC_ALL */
e5f10d49 2188# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 2189
65ebb059 2190 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
2191 * the list. Normally the loop is executed just once. But if setting the
2192 * locale fails, inside the loop we add fallback trials to the array and so
2193 * will execute the loop multiple times */
c3fcd832
KW
2194 trial_locales[0] = setlocale_init;
2195 trial_locales_count = 1;
7d4bcc4a 2196
65ebb059
KW
2197 for (i= 0; i < trial_locales_count; i++) {
2198 const char * trial_locale = trial_locales[i];
2199
2200 if (i > 0) {
2201
2202 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
2203 * when i==0, but I (khw) don't think that behavior makes much
2204 * sense */
2205 setlocale_failure = FALSE;
2206
7d4bcc4a 2207# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 2208# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 2209
65ebb059
KW
2210 /* On Windows machines, an entry of "" after the 0th means to use
2211 * the system default locale, which we now proceed to get. */
2212 if (strEQ(trial_locale, "")) {
2213 unsigned int j;
2214
2215 /* Note that this may change the locale, but we are going to do
2216 * that anyway just below */
837ce802 2217 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 2218 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 2219
7d4bcc4a 2220 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
2221 * try */
2222 if (! system_default_locale) {
2223 goto next_iteration;
2224 }
2225 for (j = 0; j < trial_locales_count; j++) {
2226 if (strEQ(system_default_locale, trial_locales[j])) {
2227 goto next_iteration;
2228 }
2229 }
2230
2231 trial_locale = system_default_locale;
2232 }
ec0202b5
KW
2233# else
2234# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
2235# endif
7d4bcc4a 2236# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
2237
2238 } /* For i > 0 */
65ebb059 2239
7d4bcc4a
KW
2240# ifdef LC_ALL
2241
948523db
KW
2242 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
2243 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
2244 if (! sl_result[LC_ALL_INDEX]) {
49c85077 2245 setlocale_failure = TRUE;
7cd8b568
KW
2246 }
2247 else {
2248 /* Since LC_ALL succeeded, it should have changed all the other
2249 * categories it can to its value; so we massage things so that the
2250 * setlocales below just return their category's current values.
2251 * This adequately handles the case in NetBSD where LC_COLLATE may
2252 * not be defined for a locale, and setting it individually will
7d4bcc4a 2253 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
2254 * the POSIX locale. */
2255 trial_locale = NULL;
2256 }
7d4bcc4a
KW
2257
2258# endif /* LC_ALL */
98994639 2259
e5f10d49
KW
2260 if (! setlocale_failure) {
2261 unsigned int j;
2262 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2263 curlocales[j]
2264 = savepv(do_setlocale_r(categories[j], trial_locale));
2265 if (! curlocales[j]) {
2266 setlocale_failure = TRUE;
2267 }
2268 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2269 }
c835d6be 2270
e5f10d49
KW
2271 if (! setlocale_failure) { /* All succeeded */
2272 break; /* Exit trial_locales loop */
49c85077 2273 }
65ebb059 2274 }
98994639 2275
49c85077
KW
2276 /* Here, something failed; will need to try a fallback. */
2277 ok = 0;
65ebb059 2278
49c85077
KW
2279 if (i == 0) {
2280 unsigned int j;
98994639 2281
65ebb059 2282 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2283
2284# ifdef LC_ALL
98994639 2285
49c85077
KW
2286 PerlIO_printf(Perl_error_log,
2287 "perl: warning: Setting locale failed.\n");
98994639 2288
7d4bcc4a 2289# else /* !LC_ALL */
98994639 2290
49c85077
KW
2291 PerlIO_printf(Perl_error_log,
2292 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2293
e5f10d49
KW
2294 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2295 if (! curlocales[j]) {
2296 PerlIO_printf(Perl_error_log, category_names[j]);
2297 }
2298 else {
2299 Safefree(curlocales[j]);
2300 }
2301 }
7d4bcc4a 2302
7d4bcc4a 2303# endif /* LC_ALL */
98994639 2304
49c85077
KW
2305 PerlIO_printf(Perl_error_log,
2306 "perl: warning: Please check that your locale settings:\n");
98994639 2307
7d4bcc4a
KW
2308# ifdef __GLIBC__
2309
49c85077
KW
2310 PerlIO_printf(Perl_error_log,
2311 "\tLANGUAGE = %c%s%c,\n",
2312 language ? '"' : '(',
2313 language ? language : "unset",
2314 language ? '"' : ')');
7d4bcc4a 2315# endif
98994639 2316
49c85077
KW
2317 PerlIO_printf(Perl_error_log,
2318 "\tLC_ALL = %c%s%c,\n",
2319 lc_all ? '"' : '(',
2320 lc_all ? lc_all : "unset",
2321 lc_all ? '"' : ')');
98994639 2322
7d4bcc4a
KW
2323# if defined(USE_ENVIRON_ARRAY)
2324
49c85077 2325 {
cd999af9 2326 char **e;
d5e32b93
KW
2327
2328 /* Look through the environment for any variables of the
2329 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2330 * already handled above. These are assumed to be locale
2331 * settings. Output them and their values. */
cd999af9 2332 for (e = environ; *e; e++) {
d5e32b93
KW
2333 const STRLEN prefix_len = sizeof("LC_") - 1;
2334 STRLEN uppers_len;
2335
cd999af9 2336 if ( strBEGINs(*e, "LC_")
c8b388b0 2337 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2338 && (uppers_len = strspn(*e + prefix_len,
2339 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2340 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2341 {
2342 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2343 (int) (prefix_len + uppers_len), *e,
2344 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2345 }
2346 }
49c85077 2347 }
7d4bcc4a
KW
2348
2349# else
2350
49c85077
KW
2351 PerlIO_printf(Perl_error_log,
2352 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2353
2354# endif
98994639 2355
49c85077
KW
2356 PerlIO_printf(Perl_error_log,
2357 "\tLANG = %c%s%c\n",
2358 lang ? '"' : '(',
2359 lang ? lang : "unset",
2360 lang ? '"' : ')');
98994639 2361
49c85077
KW
2362 PerlIO_printf(Perl_error_log,
2363 " are supported and installed on your system.\n");
2364 }
98994639 2365
65ebb059 2366 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2367 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2368 * usually change the upper bound of the loop we are in.
2369 *
2370 * Since the system's default way of setting the locale has not
2371 * found one that works, We use Perl's defined ordering: LC_ALL,
2372 * LANG, and the C locale. We don't try the same locale twice, so
2373 * don't add to the list if already there. (On POSIX systems, the
2374 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2375 * but there's no harm done by doing it explicitly.
2376 *
2377 * Note that this tries the LC_ALL environment variable even on
2378 * systems which have no LC_ALL locale setting. This may or may
2379 * not have been originally intentional, but there's no real need
2380 * to change the behavior. */
65ebb059
KW
2381 if (lc_all) {
2382 for (j = 0; j < trial_locales_count; j++) {
2383 if (strEQ(lc_all, trial_locales[j])) {
2384 goto done_lc_all;
2385 }
2386 }
2387 trial_locales[trial_locales_count++] = lc_all;
2388 }
2389 done_lc_all:
98994639 2390
65ebb059
KW
2391 if (lang) {
2392 for (j = 0; j < trial_locales_count; j++) {
2393 if (strEQ(lang, trial_locales[j])) {
2394 goto done_lang;
2395 }
2396 }
2397 trial_locales[trial_locales_count++] = lang;
2398 }
2399 done_lang:
2400
7d4bcc4a
KW
2401# if defined(WIN32) && defined(LC_ALL)
2402
65ebb059
KW
2403 /* For Windows, we also try the system default locale before "C".
2404 * (If there exists a Windows without LC_ALL we skip this because
2405 * it gets too complicated. For those, the "C" is the next
2406 * fallback possibility). The "" is the same as the 0th element of
2407 * the array, but the code at the loop above knows to treat it
2408 * differently when not the 0th */
2409 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2410
2411# endif
65ebb059
KW
2412
2413 for (j = 0; j < trial_locales_count; j++) {
2414 if (strEQ("C", trial_locales[j])) {
2415 goto done_C;
2416 }
2417 }
2418 trial_locales[trial_locales_count++] = "C";
98994639 2419
65ebb059
KW
2420 done_C: ;
2421 } /* end of first time through the loop */
98994639 2422
7d4bcc4a
KW
2423# ifdef WIN32
2424
65ebb059 2425 next_iteration: ;
7d4bcc4a
KW
2426
2427# endif
65ebb059
KW
2428
2429 } /* end of looping through the trial locales */
2430
2431 if (ok < 1) { /* If we tried to fallback */
2432 const char* msg;
2433 if (! setlocale_failure) { /* fallback succeeded */
2434 msg = "Falling back to";
2435 }
2436 else { /* fallback failed */
e5f10d49 2437 unsigned int j;
98994639 2438
65ebb059
KW
2439 /* We dropped off the end of the loop, so have to decrement i to
2440 * get back to the value the last time through */
2441 i--;
98994639 2442
65ebb059
KW
2443 ok = -1;
2444 msg = "Failed to fall back to";
2445
2446 /* To continue, we should use whatever values we've got */
7d4bcc4a 2447
e5f10d49
KW
2448 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2449 Safefree(curlocales[j]);
2450 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2451 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2452 }
65ebb059
KW
2453 }
2454
2455 if (locwarn) {
2456 const char * description;
2457 const char * name = "";
2458 if (strEQ(trial_locales[i], "C")) {
2459 description = "the standard locale";
2460 name = "C";
2461 }
7d4bcc4a
KW
2462
2463# ifdef SYSTEM_DEFAULT_LOCALE
2464
65ebb059
KW
2465 else if (strEQ(trial_locales[i], "")) {
2466 description = "the system default locale";
2467 if (system_default_locale) {
2468 name = system_default_locale;
2469 }
2470 }
7d4bcc4a
KW
2471
2472# endif /* SYSTEM_DEFAULT_LOCALE */
2473
65ebb059
KW
2474 else {
2475 description = "a fallback locale";
2476 name = trial_locales[i];
2477 }
2478 if (name && strNE(name, "")) {
2479 PerlIO_printf(Perl_error_log,
2480 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2481 }
2482 else {
2483 PerlIO_printf(Perl_error_log,
2484 "perl: warning: %s %s.\n", msg, description);
2485 }
2486 }
2487 } /* End of tried to fallback */
98994639 2488
e5f10d49
KW
2489 /* Done with finding the locales; update our records */
2490
7d4bcc4a
KW
2491# ifdef USE_LOCALE_CTYPE
2492
948523db 2493 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2494
e5f10d49 2495# endif
7d4bcc4a
KW
2496# ifdef USE_LOCALE_COLLATE
2497
948523db 2498 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2499
e5f10d49 2500# endif
7d4bcc4a
KW
2501# ifdef USE_LOCALE_NUMERIC
2502
948523db 2503 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2504
2505# endif
2506
948523db 2507 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20
KW
2508
2509# if defined(USE_ITHREADS)
2510
2511 /* This caches whether each category's locale is UTF-8 or not. This
2512 * may involve changing the locale. It is ok to do this at
2513 * initialization time before any threads have started, but not later.
2514 * Caching means that if the program heeds our dictate not to change
2515 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
2516 * it may get queried without having to change locales. If the
2517 * environment is such that all categories have the same locale, this
2518 * isn't needed, as the code will not change the locale; but this
2519 * handles the uncommon case where the environment has disparate
2520 * locales for the categories */
47280b20
KW
2521 (void) _is_cur_LC_category_utf8(categories[i]);
2522
2523# endif
2524
e5f10d49
KW
2525 Safefree(curlocales[i]);
2526 }
b310b053 2527
7d4bcc4a
KW
2528# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2529
49c85077 2530 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
2531 * locale is UTF-8. The call to new_ctype() just above has already
2532 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
2533 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
2534 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
2535 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
2536 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 2537
a05d7ebb 2538 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2539 This is an alternative to using the -C command line switch
2540 (the -C if present will override this). */
2541 {
dd374669 2542 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2543 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2544 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2545 PL_utf8cache = -1;
b310b053
JH
2546 }
2547
7d4bcc4a 2548# endif
7d4bcc4a
KW
2549# ifdef __GLIBC__
2550
175c4cf9 2551 Safefree(language);
7d4bcc4a
KW
2552
2553# endif
175c4cf9
KW
2554
2555 Safefree(lc_all);
2556 Safefree(lang);
2557
e3305790 2558#endif /* USE_LOCALE */
2fcc0ca9 2559#ifdef DEBUGGING
7d4bcc4a 2560
2fcc0ca9 2561 /* So won't continue to output stuff */
27cdc72e 2562 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2563
2fcc0ca9
KW
2564#endif
2565
98994639
HS
2566 return ok;
2567}
2568
98994639
HS
2569#ifdef USE_LOCALE_COLLATE
2570
a4a439fb 2571char *
a4a439fb
KW
2572Perl__mem_collxfrm(pTHX_ const char *input_string,
2573 STRLEN len, /* Length of 'input_string' */
2574 STRLEN *xlen, /* Set to length of returned string
2575 (not including the collation index
2576 prefix) */
2577 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2578 )
98994639 2579{
a4a439fb
KW
2580
2581 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2582 * differences. First, it handles embedded NULs. Second, it allocates a bit
2583 * more memory than needed for the transformed data itself. The real
55e5378d 2584 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2585 * the length of that, and doesn't include the collation index size.
2586 * Please see sv_collxfrm() to see how this is used. */
2587
55e5378d
KW
2588#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2589
6696cfa7
KW
2590 char * s = (char *) input_string;
2591 STRLEN s_strlen = strlen(input_string);
79f120c8 2592 char *xbuf = NULL;
55e5378d 2593 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2594 STRLEN length_in_chars;
c664130f 2595 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2596
a4a439fb
KW
2597 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2598
2599 /* Must be NUL-terminated */
2600 assert(*(input_string + len) == '\0');
7918f24d 2601
79f120c8
KW
2602 /* If this locale has defective collation, skip */
2603 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2604 DEBUG_L(PerlIO_printf(Perl_debug_log,
2605 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2606 goto bad;
2607 }
2608
6696cfa7
KW
2609 /* Replace any embedded NULs with the control that sorts before any others.
2610 * This will give as good as possible results on strings that don't
2611 * otherwise contain that character, but otherwise there may be
2612 * less-than-perfect results with that character and NUL. This is
fdc080f3 2613 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2614 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2615 NUL */
6696cfa7
KW
2616 char * e = s + len;
2617 char * sans_nuls;
fdc080f3 2618 STRLEN sans_nuls_len;
94762aa0 2619 int try_non_controls;
afc4976f
KW
2620 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2621 making sure 2nd byte is NUL.
2622 */
2623 STRLEN this_replacement_len;
2624
1e4c9676
KW
2625 /* If we don't know what non-NUL control character sorts lowest for
2626 * this locale, find it */
f28f4d2a 2627 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2628 int j;
afc4976f 2629 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2630 includes the collation index
2631 prefixed. */
2632
91c0e2e0 2633 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2634
2635 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2636 * NUL, in which case we instead look for any character. Controls
2637 * are preferred because collation order is, in general, context
2638 * sensitive, with adjoining characters affecting the order, and
2639 * controls are less likely to have such interactions, allowing the
2640 * NUL-replacement to stand on its own. (Another way to look at it
2641 * is to imagine what would happen if the NUL were replaced by a
2642 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2643 for (try_non_controls = 0;
2644 try_non_controls < 2;
2645 try_non_controls++)
2646 {
d4ff9586
KW
2647 /* Look through all legal code points (NUL isn't) */
2648 for (j = 1; j < 256; j++) {
2649 char * x; /* j's xfrm plus collation index */
2650 STRLEN x_len; /* length of 'x' */
2651 STRLEN trial_len = 1;
736a4fed 2652 char cur_source[] = { '\0', '\0' };
d4ff9586 2653
736a4fed
KW
2654 /* Skip non-controls the first time through the loop. The
2655 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2656 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2657 ? ! isCNTRL_L1(j)
2658 : ! isCNTRL_LC(j))
2659 {
d4ff9586 2660 continue;
6696cfa7 2661 }
6696cfa7 2662
736a4fed
KW
2663 /* Create a 1-char string of the current code point */
2664 cur_source[0] = (char) j;
2665
d4ff9586
KW
2666 /* Then transform it */
2667 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2668 0 /* The string is not in UTF-8 */);
6696cfa7 2669
1e4c9676 2670 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2671 * */
2672 if (! x) {
2673 continue;
2674 }
6696cfa7 2675
d4ff9586
KW
2676 /* If this character's transformation is lower than
2677 * the current lowest, this one becomes the lowest */
2678 if ( cur_min_x == NULL
2679 || strLT(x + COLLXFRM_HDR_LEN,
2680 cur_min_x + COLLXFRM_HDR_LEN))
2681 {
f28f4d2a 2682 PL_strxfrm_NUL_replacement = j;
d4ff9586 2683 cur_min_x = x;
d4ff9586
KW
2684 }
2685 else {
2686 Safefree(x);
2687 }
1e4c9676 2688 } /* end of loop through all 255 characters */
6696cfa7 2689
1e4c9676 2690 /* Stop looking if found */
94762aa0
KW
2691 if (cur_min_x) {
2692 break;
2693 }
2694
2695 /* Unlikely, but possible, if there aren't any controls that
2696 * work in the locale, repeat the loop, looking for any
2697 * character that works */
2698 DEBUG_L(PerlIO_printf(Perl_debug_log,
2699 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2700 } /* End of loop to try first the controls, then any char */
6696cfa7 2701
94762aa0
KW
2702 if (! cur_min_x) {
2703 DEBUG_L(PerlIO_printf(Perl_debug_log,
2704 "_mem_collxfrm: Couldn't find any character to replace"
2705 " embedded NULs in locale %s with", PL_collation_name));
2706 goto bad;
58eebef2
KW
2707 }
2708
94762aa0
KW
2709 DEBUG_L(PerlIO_printf(Perl_debug_log,
2710 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2711 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2712
6696cfa7 2713 Safefree(cur_min_x);
1e4c9676 2714 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2715
2716 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2717 * UTF8-ness of the original */
f28f4d2a
KW
2718 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2719 this_replacement_char[0] =
2720 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2721 this_replacement_char[1] =
2722 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2723 this_replacement_len = 2;
2724 }
2725 else {
f28f4d2a 2726 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2727 /* this_replacement_char[1] = '\0' was done at initialization */
2728 this_replacement_len = 1;
6696cfa7
KW
2729 }
2730
2731 /* The worst case length for the replaced string would be if every
2732 * character in it is NUL. Multiply that by the length of each
2733 * replacement, and allow for a trailing NUL */
afc4976f 2734 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2735 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2736 *sans_nuls = '\0';
2737
6696cfa7
KW
2738 /* Replace each NUL with the lowest collating control. Loop until have
2739 * exhausted all the NULs */
2740 while (s + s_strlen < e) {
6069d6c5 2741 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2742
2743 /* Do the actual replacement */
6069d6c5 2744 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2745
2746 /* Move past the input NUL */
2747 s += s_strlen + 1;
2748 s_strlen = strlen(s);
2749 }
2750
2751 /* And add anything that trails the final NUL */
6069d6c5 2752 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2753
2754 /* Switch so below we transform this modified string */
2755 s = sans_nuls;
2756 len = strlen(s);
1e4c9676 2757 } /* End of replacing NULs */
6696cfa7 2758
a4a439fb
KW
2759 /* Make sure the UTF8ness of the string and locale match */
2760 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 2761 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
2762 const char * const t = s; /* Temporary so we can later find where the
2763 input was */
2764
2765 /* Here they don't match. Change the string's to be what the locale is
2766 * expecting */
2767
2768 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2769 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2770 utf8 = TRUE;
2771 }
2772 else { /* locale is not UTF-8; but input is; downgrade the input */
2773
2774 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2775
2776 /* If the downgrade was successful we are done, but if the input
2777 * contains things that require UTF-8 to represent, have to do
2778 * damage control ... */
2779 if (UNLIKELY(utf8)) {
2780
2781 /* What we do is construct a non-UTF-8 string with
2782 * 1) the characters representable by a single byte converted
2783 * to be so (if necessary);
2784 * 2) and the rest converted to collate the same as the
2785 * highest collating representable character. That makes
2786 * them collate at the end. This is similar to how we
2787 * handle embedded NULs, but we use the highest collating
2788 * code point instead of the smallest. Like the NUL case,
2789 * this isn't perfect, but is the best we can reasonably
2790 * do. Every above-255 code point will sort the same as
2791 * the highest-sorting 0-255 code point. If that code
2792 * point can combine in a sequence with some other code
2793 * points for weight calculations, us changing something to
2794 * be it can adversely affect the results. But in most
2795 * cases, it should work reasonably. And note that this is
2796 * really an illegal situation: using code points above 255
2797 * on a locale where only 0-255 are valid. If two strings
2798 * sort entirely equal, then the sort order for the
2799 * above-255 code points will be in code point order. */
2800
2801 utf8 = FALSE;
2802
2803 /* If we haven't calculated the code point with the maximum
2804 * collating order for this locale, do so now */
2805 if (! PL_strxfrm_max_cp) {
2806 int j;
2807
2808 /* The current transformed string that collates the
2809 * highest (except it also includes the prefixed collation
2810 * index. */
2811 char * cur_max_x = NULL;
2812
2813 /* Look through all legal code points (NUL isn't) */
2814 for (j = 1; j < 256; j++) {
2815 char * x;
2816 STRLEN x_len;
736a4fed 2817 char cur_source[] = { '\0', '\0' };
a4a439fb 2818
736a4fed
KW
2819 /* Create a 1-char string of the current code point */
2820 cur_source[0] = (char) j;
a4a439fb
KW
2821
2822 /* Then transform it */
2823 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2824
2825 /* If something went wrong (which it shouldn't), just
2826 * ignore this code point */
94762aa0 2827 if (! x) {
a4a439fb
KW
2828 continue;
2829 }
2830
2831 /* If this character's transformation is higher than
2832 * the current highest, this one becomes the highest */
2833 if ( cur_max_x == NULL
55e5378d
KW
2834 || strGT(x + COLLXFRM_HDR_LEN,
2835 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2836 {
2837 PL_strxfrm_max_cp = j;
2838 cur_max_x = x;
2839 }
2840 else {
2841 Safefree(x);
2842 }
2843 }
2844
94762aa0
KW
2845 if (! cur_max_x) {
2846 DEBUG_L(PerlIO_printf(Perl_debug_log,
2847 "_mem_collxfrm: Couldn't find any character to"
2848 " replace above-Latin1 chars in locale %s with",
2849 PL_collation_name));
2850 goto bad;
2851 }
2852
58eebef2
KW
2853 DEBUG_L(PerlIO_printf(Perl_debug_log,
2854 "_mem_collxfrm: highest 1-byte collating character"
2855 " in locale %s is 0x%02X\n",
2856 PL_collation_name,
2857 PL_strxfrm_max_cp));
58eebef2 2858
a4a439fb
KW
2859 Safefree(cur_max_x);
2860 }
2861
2862 /* Here we know which legal code point collates the highest.
2863 * We are ready to construct the non-UTF-8 string. The length
2864 * will be at least 1 byte smaller than the input string
2865 * (because we changed at least one 2-byte character into a
2866 * single byte), but that is eaten up by the trailing NUL */
2867 Newx(s, len, char);
2868
2869 {
2870 STRLEN i;
2871 STRLEN d= 0;
042d9e50 2872 char * e = (char *) t + len;
a4a439fb
KW
2873
2874 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2875 U8 cur_char = t[i];
2876 if (UTF8_IS_INVARIANT(cur_char)) {
2877 s[d++] = cur_char;
2878 }
042d9e50 2879 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2880 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2881 }
2882 else { /* Replace illegal cp with highest collating
2883 one */
2884 s[d++] = PL_strxfrm_max_cp;
2885 }
2886 }
2887 s[d++] = '\0';
2888 Renew(s, d, char); /* Free up unused space */
2889 }
2890 }
2891 }
2892
2893 /* Here, we have constructed a modified version of the input. It could
2894 * be that we already had a modified copy before we did this version.
2895 * If so, that copy is no longer needed */
2896 if (t != input_string) {
2897 Safefree(t);
2898 }
2899 }
2900
17f41037
KW
2901 length_in_chars = (utf8)
2902 ? utf8_length((U8 *) s, (U8 *) s + len)
2903 : len;
2904
59c018b9
KW
2905 /* The first element in the output is the collation id, used by
2906 * sv_collxfrm(); then comes the space for the transformed string. The
2907 * equation should give us a good estimate as to how much is needed */
55e5378d 2908 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2909 + PL_collxfrm_base
17f41037 2910 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2911 Newx(xbuf, xAlloc, char);
c7202dee
KW
2912 if (UNLIKELY(! xbuf)) {
2913 DEBUG_L(PerlIO_printf(Perl_debug_log,
2914 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2915 goto bad;
c7202dee 2916 }
98994639 2917
d35fca5f 2918 /* Store the collation id */
98994639 2919 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2920
2921 /* Then the transformation of the input. We loop until successful, or we
2922 * give up */
4ebeff16 2923 for (;;) {
1adab0a7 2924
55e5378d 2925 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2926
2927 /* If the transformed string occupies less space than we told strxfrm()
2928 * was available, it means it successfully transformed the whole
2929 * string. */
55e5378d 2930 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2931
1adab0a7
KW
2932 /* Some systems include a trailing NUL in the returned length.
2933 * Ignore it, using a loop in case multiple trailing NULs are
2934 * returned. */
2935 while ( (*xlen) > 0
2936 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2937 {
2938 (*xlen)--;
2939 }
2940
17f41037
KW
2941 /* If the first try didn't get it, it means our prediction was low.
2942 * Modify the coefficients so that we predict a larger value in any
2943 * future transformations */
2944 if (! first_time) {
2945 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2946 STRLEN computed_guess = PL_collxfrm_base
2947 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2948
2949 /* On zero-length input, just keep current slope instead of
2950 * dividing by 0 */
2951 const STRLEN new_m = (length_in_chars != 0)
2952 ? needed / length_in_chars
2953 : PL_collxfrm_mult;
17f41037
KW
2954
2955 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2956 "%s: %d: initial size of %zu bytes for a length "
2957 "%zu string was insufficient, %zu needed\n",
17f41037 2958 __FILE__, __LINE__,
b07929e4 2959 computed_guess, length_in_chars, needed));
17f41037
KW
2960
2961 /* If slope increased, use it, but discard this result for
2962 * length 1 strings, as we can't be sure that it's a real slope
2963 * change */
2964 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
2965
2966# ifdef DEBUGGING
2967
17f41037
KW
2968 STRLEN old_m = PL_collxfrm_mult;
2969 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
2970
2971# endif
2972
17f41037
KW
2973 PL_collxfrm_mult = new_m;
2974 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2975 computed_guess = PL_collxfrm_base
2976 + (PL_collxfrm_mult * length_in_chars);
2977 if (computed_guess < needed) {
2978 PL_collxfrm_base += needed - computed_guess;
2979 }
2980
2981 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2982 "%s: %d: slope is now %zu; was %zu, base "
2983 "is now %zu; was %zu\n",
17f41037 2984 __FILE__, __LINE__,
b07929e4
KW
2985 PL_collxfrm_mult, old_m,
2986 PL_collxfrm_base, old_b));
17f41037
KW
2987 }
2988 else { /* Slope didn't change, but 'b' did */
2989 const STRLEN new_b = needed
2990 - computed_guess
2991 + PL_collxfrm_base;
2992 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2993 "%s: %d: base is now %zu; was %zu\n",
17f41037 2994 __FILE__, __LINE__,
b07929e4 2995 new_b, PL_collxfrm_base));
17f41037
KW
2996 PL_collxfrm_base = new_b;
2997 }
2998 }
2999
4ebeff16
KW
3000 break;
3001 }
bb0f664e 3002
c7202dee
KW
3003 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3004 DEBUG_L(PerlIO_printf(Perl_debug_log,
3005 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3006 *xlen, PERL_INT_MAX));
4ebeff16 3007 goto bad;
c7202dee 3008 }
d35fca5f 3009
c664130f 3010 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
3011 * (usually not including the trailing NUL) when it fails due to not
3012 * enough space being provided. Assume that this is the case unless
3013 * it's been proven otherwise */
c664130f 3014 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 3015 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
3016 }
3017 else { /* Here, either:
3018 * 1) The strxfrm() has previously shown bad behavior; or
3019 * 2) It isn't the first time through the loop, which means
3020 * that the strxfrm() is now showing bad behavior, because
3021 * we gave it what it said was needed in the previous
3022 * iteration, and it came back saying it needed still more.
3023 * (Many versions of cygwin fit this. When the buffer size
3024 * isn't sufficient, they return the input size instead of
3025 * how much is needed.)
d4ff9586
KW
3026 * Increase the buffer size by a fixed percentage and try again.
3027 * */
6ddd902c 3028 xAlloc += (xAlloc / 4) + 1;
c664130f 3029 PL_strxfrm_is_behaved = FALSE;
c664130f 3030
7d4bcc4a
KW
3031# ifdef DEBUGGING
3032
58eebef2
KW
3033 if (DEBUG_Lv_TEST || debug_initialization) {
3034 PerlIO_printf(Perl_debug_log,
3035 "_mem_collxfrm required more space than previously calculated"
b07929e4 3036 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 3037 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 3038 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 3039 }
7d4bcc4a
KW
3040
3041# endif
3042
58eebef2 3043 }
c664130f 3044
4ebeff16 3045 Renew(xbuf, xAlloc, char);
c7202dee
KW
3046 if (UNLIKELY(! xbuf)) {
3047 DEBUG_L(PerlIO_printf(Perl_debug_log,
3048 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 3049 goto bad;
c7202dee 3050 }
c664130f
KW
3051
3052 first_time = FALSE;
4ebeff16 3053 }
98994639 3054
6696cfa7 3055
7d4bcc4a
KW
3056# ifdef DEBUGGING
3057
58eebef2 3058 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
3059
3060 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
3061 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
3062 PerlIO_printf(Perl_debug_log, "%s\n",
3063 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
3064 *xlen, 1));
58eebef2 3065 }
7d4bcc4a
KW
3066
3067# endif
58eebef2 3068
3c5f993e 3069 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 3070 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 3071
6696cfa7
KW
3072 if (s != input_string) {
3073 Safefree(s);
98994639
HS
3074 }
3075
98994639
HS
3076 return xbuf;
3077
3078 bad:
3079 Safefree(xbuf);
6696cfa7
KW
3080 if (s != input_string) {
3081 Safefree(s);
3082 }
98994639 3083 *xlen = 0;
7d4bcc4a
KW
3084
3085# ifdef DEBUGGING
3086
58eebef2 3087 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 3088 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 3089 }
7d4bcc4a
KW
3090
3091# endif
3092
98994639
HS
3093 return NULL;
3094}
3095
7d4bcc4a 3096# ifdef DEBUGGING
c7202dee 3097
4cbaac56 3098STATIC void
c7202dee
KW
3099S_print_collxfrm_input_and_return(pTHX_
3100 const char * const s,
3101 const char * const e,
3102 const STRLEN * const xlen,
3103 const bool is_utf8)
3104{
c7202dee
KW
3105
3106 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
3107
511e4ff7
DM
3108 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
3109 (UV)PL_collation_ix);
c7202dee 3110 if (xlen) {
08b6dc1d 3111 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
3112 }
3113 else {
3114 PerlIO_printf(Perl_debug_log, "NULL");
3115 }
3116 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
3117 PL_collation_name);
9c8a6dc2
KW
3118 print_bytes_for_locale(s, e, is_utf8);
3119
3120 PerlIO_printf(Perl_debug_log, "'\n");
3121}
3122
3123STATIC void
3124S_print_bytes_for_locale(pTHX_
3125 const char * const s,
3126 const char * const e,
3127 const bool is_utf8)
3128{
3129 const char * t = s;
3130 bool prev_was_printable = TRUE;
3131 bool first_time = TRUE;
3132
3133 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
3134
3135 while (t < e) {
3136 UV cp = (is_utf8)
3137 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
3138 : * (U8 *) t;
3139 if (isPRINT(cp)) {
3140 if (! prev_was_printable) {
3141 PerlIO_printf(Perl_debug_log, " ");
3142 }
3143 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
3144 prev_was_printable = TRUE;
3145 }
3146 else {
3147 if (! first_time) {
3148 PerlIO_printf(Perl_debug_log, " ");
3149 }
147e3846 3150 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
3151 prev_was_printable = FALSE;
3152 }
3153 t += (is_utf8) ? UTF8SKIP(t) : 1;
3154 first_time = FALSE;
3155 }
c7202dee
KW
3156}
3157
7d4bcc4a 3158# endif /* #ifdef DEBUGGING */
98994639 3159#endif /* USE_LOCALE_COLLATE */
58eebef2 3160
8ef6e574
KW
3161#ifdef USE_LOCALE
3162
962aa53f
KW
3163STATIC const char *
3164S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
3165{
3166 /* Changes the locale for LC_'switch_category" to that of
3167 * LC_'template_category', if they aren't already the same. If not NULL,
3168 * 'template_locale' is the locale that 'template_category' is in.
3169 *
9de6fe47
KW
3170 * Returns a copy of the name of the original locale for 'switch_category'
3171 * so can be switched back to with the companion function
3172 * restore_switched_locale(), (NULL if no restoral is necessary.) */
962aa53f
KW
3173
3174 char * restore_to_locale = NULL;
3175
3176 if (switch_category == template_category) { /* No changes needed */
3177 return NULL;
3178 }
3179
3180 /* Find the original locale of the category we may need to change, so that
3181 * it can be restored to later */
d6189047
KW
3182 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
3183 NULL)));
962aa53f
KW
3184 if (! restore_to_locale) {
3185 Perl_croak(aTHX_
3186 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3187 __FILE__, __LINE__, category_name(switch_category), errno);
3188 }
962aa53f
KW
3189
3190 /* If the locale of the template category wasn't passed in, find it now */
3191 if (template_locale == NULL) {
3192 template_locale = do_setlocale_r(template_category, NULL);
3193 if (! template_locale) {
3194 Perl_croak(aTHX_
3195 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3196 __FILE__, __LINE__, category_name(template_category), errno);
3197 }
3198 }
3199
3200 /* It the locales are the same, there's nothing to do */
3201 if (strEQ(restore_to_locale, template_locale)) {
3202 Safefree(restore_to_locale);
3203
3204 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
3205 category_name(switch_category), restore_to_locale));
3206
3207 return NULL;
3208 }
3209
3210 /* Finally, change the locale to the template one */
3211 if (! do_setlocale_r(switch_category, template_locale)) {
3212 Perl_croak(aTHX_
3213 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
3214 __FILE__, __LINE__, category_name(switch_category),
3215 template_locale, errno);
3216 }
3217
3218 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
3219 category_name(switch_category), template_locale));
3220
3221 return restore_to_locale;
3222}
3223
3224STATIC void
3225S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
3226{
9de6fe47
KW
3227 /* Restores the locale for LC_'category' to 'original_locale' (which is a
3228 * copy that will be freed by this function), or do nothing if the latter
3229 * parameter is NULL */
962aa53f
KW
3230
3231 if (original_locale == NULL) {
3232 return;
3233 }
3234
3235 if (! do_setlocale_r(category, original_locale)) {
3236 Perl_croak(aTHX_
3237 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
3238 __FILE__, __LINE__,
3239 category_name(category), original_locale, errno);
3240 }
3241
3242 Safefree(original_locale);
3243}
3244
c1284011
KW
3245bool
3246Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
3247{
3248 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
3249 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 3250 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
3251 * could give the wrong result. The result will very likely be correct for
3252 * languages that have commonly used non-ASCII characters, but for notably
3253 * English, it comes down to if the locale's name ends in something like
9de6fe47
KW
3254 * "UTF-8". It errs on the side of not being a UTF-8 locale.
3255 *
3256 * If the platform is early C89, not containing mbtowc(), or we are
3257 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
3258 * These work very well for non-Latin locales or those whose currency
3259 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
3260 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
3261 * on the name containing UTF-8 or not. */
7d74bb61 3262
47280b20 3263 /* Name of current locale corresponding to the input category */
8de4332b 3264 const char *save_input_locale = NULL;
47280b20
KW
3265
3266 bool is_utf8 = FALSE; /* The return value */
7d74bb61 3267
47280b20
KW
3268 /* The variables below are for the cache of previous lookups using this
3269 * function. The cache is a C string, described at the definition for
3270 * 'C_and_POSIX_utf8ness'.
3271 *
3272 * The first part of the cache is fixed, for the C and POSIX locales. The
3273 * varying part starts just after them. */
3274 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
3275
3276 Size_t utf8ness_cache_size; /* Size of the varying portion */
3277 Size_t input_name_len; /* Length in bytes of save_input_locale */
3278 Size_t input_name_len_with_overhead; /* plus extra chars used to store
3279 the name in the cache */
3280 char * delimited; /* The name plus the delimiters used to store
3281 it in the cache */
3282 char * name_pos; /* position of 'delimited' in the cache, or 0
3283 if not there */
3284
3285
7d4bcc4a
KW
3286# ifdef LC_ALL
3287
7d74bb61 3288 assert(category != LC_ALL);
7d4bcc4a
KW
3289
3290# endif
7d74bb61 3291
47280b20 3292 /* Get the desired category's locale */
d6189047 3293 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
7d74bb61 3294 if (! save_input_locale) {
47280b20 3295 Perl_croak(aTHX_
d707d779
KW
3296 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3297 __FILE__, __LINE__, category_name(category), errno);
7d74bb61 3298 }
47280b20 3299
47280b20
KW
3300 DEBUG_L(PerlIO_printf(Perl_debug_log,
3301 "Current locale for %s is %s\n",
3302 category_name(category), save_input_locale));
3303
3304 input_name_len = strlen(save_input_locale);
3305
3306 /* In our cache, each name is accompanied by two delimiters and a single
3307 * utf8ness digit */
3308 input_name_len_with_overhead = input_name_len + 3;
3309
3310 /* Allocate and populate space for a copy of the name surrounded by the
3311 * delimiters */
3312 Newx(delimited, input_name_len_with_overhead, char);
3313 delimited[0] = UTF8NESS_SEP[0];
3314 Copy(save_input_locale, delimited + 1, input_name_len, char);
3315 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
3316 delimited[input_name_len+2] = '\0';
3317
3318 /* And see if that is in the cache */
3319 name_pos = instr(PL_locale_utf8ness, delimited);
3320 if (name_pos) {
3321 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
3322
3323# ifdef DEBUGGING
3324
3325 if (DEBUG_Lv_TEST || debug_initialization) {
3326 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
3327 save_input_locale, is_utf8);
3328 }
3329
3330# endif
3331
3332 /* And, if not already in that position, move it to the beginning of
3333 * the non-constant portion of the list, since it is the most recently
3334 * used. (We don't have to worry about overflow, since just moving
3335 * existing names around) */
3336 if (name_pos > utf8ness_cache) {
3337 Move(utf8ness_cache,
3338 utf8ness_cache + input_name_len_with_overhead,
3339 name_pos - utf8ness_cache, char);
3340 Copy(delimited,
3341 utf8ness_cache,
3342 input_name_len_with_overhead - 1, char);
3343 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3344 }
3345
3346 Safefree(delimited);
b07fffd1 3347 Safefree(save_input_locale);
47280b20 3348 return is_utf8;
7d74bb61
KW
3349 }
3350
47280b20
KW
3351 /* Here we don't have stored the utf8ness for the input locale. We have to
3352 * calculate it */
3353
94646a69
KW
3354# if defined(USE_LOCALE_CTYPE) \
3355 && ( (defined(HAS_NL_LANGINFO) && defined(CODESET)) \
3356 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
7d74bb61 3357
0dec74cd 3358 {
962aa53f
KW
3359 const char *original_ctype_locale
3360 = switch_category_locale_to_template(LC_CTYPE,
3361 category,
3362 save_input_locale);
69014004 3363
7d74bb61 3364 /* Here the current LC_CTYPE is set to the locale of the category whose
0dec74cd 3365 * information is desired. This means that nl_langinfo() and mbtowc()
1d958db2 3366 * should give the correct results */
119ee68b 3367
0dec74cd
KW
3368# ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
3369 calling the functions if we have this */
3370
3371 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3372 * Unicode code point. */
3373
de3a3072
KW
3374 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
3375 __FILE__, __LINE__, (int) MB_CUR_MAX));
0dec74cd
KW
3376 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
3377 is_utf8 = FALSE;
b5b2847c
KW
3378 restore_switched_locale(LC_CTYPE, original_ctype_locale);
3379 goto finish_and_return;
0dec74cd
KW
3380 }
3381
3382# endif
7d4bcc4a
KW
3383# if defined(HAS_NL_LANGINFO) && defined(CODESET)
3384
0dec74cd
KW
3385 { /* The task is easiest if the platform has this POSIX 2001 function.
3386 Except on some platforms it can wrongly return "", so have to have
3387 a fallback. And it can return that it's UTF-8, even if there are
3388 variances from that. For example, Turkish locales may use the
3389 alternate dotted I rules, and sometimes it appears to be a
3390 defective locale definition. XXX We should probably check for
9de6fe47
KW
3391 these in the Latin1 range and warn (but on glibc, requires
3392 iswalnum() etc. due to their not handling 80-FF correctly */
c70a3e68
KW
3393 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
3394 /* FALSE => already in dest locale */
119ee68b 3395
af84e886 3396 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
c70a3e68
KW
3397 "\tnllanginfo returned CODESET '%s'\n", codeset));
3398
3399 if (codeset && strNE(codeset, "")) {
1d958db2 3400
1d075173
KW
3401 /* If the implementation of foldEQ() somehow were
3402 * to change to not go byte-by-byte, this could
3403 * read past end of string, as only one length is
3404 * checked. But currently, a premature NUL will
3405 * compare false, and it will stop there */
3406 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
3407 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
1d958db2 3408
69014004
KW
3409 DEBUG_L(PerlIO_printf(Perl_debug_log,
3410 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
3411 codeset, is_utf8));
b5b2847c
KW
3412 restore_switched_locale(LC_CTYPE, original_ctype_locale);
3413 goto finish_and_return;
1d958db2 3414 }
119ee68b
KW
3415 }
3416
7d4bcc4a 3417# endif
94646a69 3418# if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
0dec74cd
KW
3419 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
3420 * late adder to C89, so very likely to have it. However, testing has
3421 * shown that, like nl_langinfo() above, there are locales that are not
3422 * strictly UTF-8 that this will return that they are */
69014004 3423
0dec74cd 3424 {
119ee68b 3425 wchar_t wc;
51fc4b19 3426 int len;
48015184 3427 dSAVEDERRNO;
51fc4b19 3428
94646a69
KW
3429# if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
3430
3431 mbstate_t ps;
3432
3433# endif
3434
3435 /* mbrtowc() and mbtowc() convert a byte string to a wide
3436 * character. Feed a byte string to one of them and check that the
3437 * result is the expected Unicode code point */
3438
3439# if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
3440 /* Prefer this function if available, as it's reentrant */
3441
3442 memset(&ps, 0, sizeof(ps));;
3443 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
3444 state */
48015184 3445 SETERRNO(0, 0);
94646a69 3446 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
48015184 3447 SAVE_ERRNO;
94646a69
KW
3448
3449# else
0dec74cd 3450
5a91a58b 3451 LOCALE_LOCK;
856b881c 3452 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
48015184 3453 SETERRNO(0, 0);
b1d4925c 3454 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
48015184 3455 SAVE_ERRNO;
5a91a58b 3456 LOCALE_UNLOCK;
51fc4b19 3457
94646a69
KW
3458# endif
3459
48015184 3460 RESTORE_ERRNO;
af84e886 3461 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
0dec74cd 3462 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
48015184 3463 len, (unsigned int) wc, GET_ERRNO));
b1d4925c 3464
0dec74cd
KW
3465 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
3466 && wc == (wchar_t) UNICODE_REPLACEMENT);
119ee68b 3467 }
7d4bcc4a 3468
962aa53f 3469 restore_switched_locale(LC_CTYPE, original_ctype_locale);
47280b20 3470 goto finish_and_return;
7d74bb61 3471 }
119ee68b 3472
0dec74cd
KW
3473# endif
3474# else
7d74bb61 3475
0dec74cd
KW
3476 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
3477 * try looking at the currency symbol to see if it disambiguates
3478 * things. Often that will be in the native script, and if the symbol
3479 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
3480 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
3481 * string being valid UTF-8 are quite small */
fa9b773e 3482
660ee27b
KW
3483# ifdef USE_LOCALE_MONETARY
3484
3485 /* If have LC_MONETARY, we can look at the currency symbol. Often that
3486 * will be in the native script. We do this one first because there is
3487 * just one string to examine, so potentially avoids work */
7d4bcc4a 3488
9db16864
KW
3489 {
3490 const char *original_monetary_locale
660ee27b
KW
3491 = switch_category_locale_to_template(LC_MONETARY,
3492 category,
3493 save_input_locale);
9db16864 3494 bool only_ascii = FALSE;
660ee27b
KW
3495 const U8 * currency_string
3496 = (const U8 *) my_nl_langinfo(PERL_CRNCYSTR, FALSE);
3497 /* 2nd param not relevant for this item */
3498 const U8 * first_variant;
fa9b773e 3499
660ee27b
KW
3500 assert( *currency_string == '-'
3501 || *currency_string == '+'
3502 || *currency_string == '.');
97f4de96 3503
660ee27b 3504 currency_string++;
fa9b773e 3505
660ee27b 3506 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
9db16864
KW
3507 {
3508 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));
3509 only_ascii = TRUE;
3510 }
3511 else {
660ee27b 3512 is_utf8 = is_strict_utf8_string(first_variant, 0);
9db16864 3513 }
fa9b773e 3514
9db16864 3515 restore_switched_locale(LC_MONETARY, original_monetary_locale);
fa9b773e 3516
9db16864 3517 if (! only_ascii) {
fa9b773e 3518
9db16864
KW
3519 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3520 * otherwise assume the locale is UTF-8 if and only if the symbol
3521 * is non-ascii UTF-8. */
af84e886 3522 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
9db16864
KW
3523 save_input_locale, is_utf8));
3524 goto finish_and_return;
3525 }
13542a67 3526 }
fa9b773e 3527
660ee27b 3528# endif /* USE_LOCALE_MONETARY */
7d4bcc4a 3529# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e 3530
9db16864
KW
3531 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3532 * the names of the months and weekdays, timezone, and am/pm indicator */
3533 {
3534 const char *original_time_locale
3535 = switch_category_locale_to_template(LC_TIME,
3536 category,
3537 save_input_locale);
3538 int hour = 10;
3539 bool is_dst = FALSE;
3540 int dom = 1;
3541 int month = 0;
3542 int i;
3543 char * formatted_time;
3544
3545 /* Here the current LC_TIME is set to the locale of the category
3546 * whose information is desired. Look at all the days of the week and
3547 * month names, and the timezone and am/pm indicator for UTF-8 variant
3548 * characters. The first such a one found will tell us if the locale
3549 * is UTF-8 or not */
3550
3551 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3552 formatted_time = my_strftime("%A %B %Z %p",
3553 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
3554 if ( ! formatted_time
3555 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3556 {
15f7e74e 3557
9db16864
KW
3558 /* Here, we didn't find a non-ASCII. Try the next time through
3559 * with the complemented dst and am/pm, and try with the next
3560 * weekday. After we have gotten all weekdays, try the next
3561 * month */
3562 is_dst = ! is_dst;
3563 hour = (hour + 12) % 24;
3564 dom++;
3565 if (i > 6) {
3566 month++;
3567 }
3568 continue;
15f7e74e 3569 }
9db16864
KW
3570
3571 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3572 * false otherwise. But first, restore LC_TIME to its original
3573 * locale if we changed it */
3574 restore_switched_locale(LC_TIME, original_time_locale);
3575
af84e886 3576 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
9db16864
KW
3577 save_input_locale,
3578 is_utf8_string((U8 *) formatted_time, 0)));
3579 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
3580 goto finish_and_return;
15f7e74e
KW
3581 }
3582
9db16864
KW
3583 /* Falling off the end of the loop indicates all the names were just
3584 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3585 * to its original locale */
962aa53f 3586 restore_switched_locale(LC_TIME, original_time_locale);
af84e886 3587 DEBUG_Lv(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
3588 }
3589
7d4bcc4a 3590# endif
15f7e74e 3591
7d4bcc4a 3592# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93 3593
9db16864
KW
3594 /* This code is ifdefd out because it was found to not be necessary in testing
3595 * on our dromedary test machine, which has over 700 locales. There, this
3596 * added no value to looking at the currency symbol and the time strings. I
3597 * left it in so as to avoid rewriting it if real-world experience indicates
3598 * that dromedary is an outlier. Essentially, instead of returning abpve if we
3599 * haven't found illegal utf8, we continue on and examine all the strerror()
3600 * messages on the platform for utf8ness. If all are ASCII, we still don't
3601 * know the answer; but otherwise we have a pretty good indication of the
3602 * utf8ness. The reason this doesn't help much is that the messages may not
3603 * have been translated into the locale. The currency symbol and time strings
3604 * are much more likely to have been translated. */
3605 {
3606 int e;
3607 bool non_ascii = FALSE;
3608 const char *original_messages_locale
3609 = switch_category_locale_to_template(LC_MESSAGES,
3610 category,
3611 save_input_locale);
3612 const char * errmsg = NULL;
3613
3614 /* Here the current LC_MESSAGES is set to the locale of the category
3615 * whose information is desired. Look through all the messages. We
3616 * can't use Strerror() here because it may expand to code that
3617 * segfaults in miniperl */
3618
3619 for (e = 0; e <= sys_nerr; e++) {
3620 errno = 0;
3621 errmsg = sys_errlist[e];
3622 if (errno || !errmsg) {
3623 break;
3624 }
3625 errmsg = savepv(errmsg);
3626 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
3627 non_ascii = TRUE;
3628 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3629 break;
3630 }
855aeb93 3631 }
9db16864 3632 Safefree(errmsg);
855aeb93 3633
9db16864 3634 restore_switched_locale(LC_MESSAGES, original_messages_locale);
855aeb93 3635
9db16864 3636 if (non_ascii) {
5857e934 3637
9db16864
KW
3638 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3639 * any non-ascii means it is one; otherwise we assume it isn't */
af84e886 3640 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
9db16864
KW
3641 save_input_locale,
3642 is_utf8));
3643 goto finish_and_return;
3644 }
855aeb93 3645
9db16864
KW
3646 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));
3647 }
855aeb93 3648
7d4bcc4a 3649# endif
0dec74cd 3650# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3651 UTF-8 locale */
7d4bcc4a 3652
97f4de96
KW
3653 /* As a last resort, look at the locale name to see if it matches
3654 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3655 * return of setlocale(), is actually defined to be opaque, so we can't
3656 * really rely on the absence of various substrings in the name to indicate
3657 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3658 * be a UTF-8 locale. Similarly for the other common names */
3659
0dec74cd 3660 {
c36d8df8 3661 const Size_t final_pos = strlen(save_input_locale) - 1;
0dec74cd 3662
c36d8df8
KW
3663 if (final_pos >= 3) {
3664 const char *name = save_input_locale;
97f4de96 3665
c36d8df8
KW
3666 /* Find next 'U' or 'u' and look from there */
3667 while ((name += strcspn(name, "Uu") + 1)
3668 <= save_input_locale + final_pos - 2)
97f4de96 3669 {
c36d8df8
KW
3670 if ( isALPHA_FOLD_NE(*name, 't')
3671 || isALPHA_FOLD_NE(*(name + 1), 'f'))
3672 {
3673 continue;
3674 }
3675 name += 2;
3676 if (*(name) == '-') {
3677 if ((name > save_input_locale + final_pos - 1)) {
3678 break;
3679 }
3680 name++;
3681 }
3682 if (*(name) == '8') {
3683 DEBUG_L(PerlIO_printf(Perl_debug_log,
3684 "Locale %s ends with UTF-8 in name\n",
3685 save_input_locale));
3686 is_utf8 = TRUE;
3687 goto finish_and_return;
97f4de96 3688 }
97f4de96 3689 }
c36d8df8
KW
3690 DEBUG_L(PerlIO_printf(Perl_debug_log,
3691 "Locale %s doesn't end with UTF-8 in name\n",
3692 save_input_locale));
97f4de96 3693 }
97f4de96 3694
4d8d465a 3695# ifdef WIN32
7d4bcc4a 3696
c36d8df8
KW
3697 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
3698 if (memENDs(save_input_locale, final_pos, "65001")) {
3699 DEBUG_L(PerlIO_printf(Perl_debug_log,
8a0832a1 3700 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
97f4de96 3701 save_input_locale));
c36d8df8
KW
3702 is_utf8 = TRUE;
3703 goto finish_and_return;
3704 }
0dec74cd 3705 }
7d4bcc4a 3706
4d8d465a
KW
3707# endif
3708# endif
97f4de96
KW
3709
3710 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3711 * since we are about to return FALSE anyway, there is no point in doing
3712 * this extra work */
7d4bcc4a 3713
0dec74cd 3714# if 0
97f4de96
KW
3715 if (instr(save_input_locale, "8859")) {
3716 DEBUG_L(PerlIO_printf(Perl_debug_log,
3717 "Locale %s has 8859 in name, not UTF-8 locale\n",
3718 save_input_locale));
47280b20
KW
3719 is_utf8 = FALSE;
3720 goto finish_and_return;
97f4de96 3721 }
0dec74cd 3722# endif
97f4de96 3723
69014004
KW
3724 DEBUG_L(PerlIO_printf(Perl_debug_log,
3725 "Assuming locale %s is not a UTF-8 locale\n",
3726 save_input_locale));
47280b20
KW
3727 is_utf8 = FALSE;
3728
0dec74cd
KW
3729# endif /* the code that is compiled when no modern LC_CTYPE */
3730
47280b20
KW
3731 finish_and_return:
3732
3733 /* Cache this result so we don't have to go through all this next time. */
3734 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
3735 - (utf8ness_cache - PL_locale_utf8ness);
3736
3737 /* But we can't save it if it is too large for the total space available */