This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add Perl_setlocale()
[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
9aac5db8
KW
613 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
614 * this function should be called directly only from this file and from
0d071d52
KW
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
9aac5db8
KW
1211/*
1212
1213=head1 Locale-related functions and macros
1214
1215=for apidoc Perl_setlocale
1216
1217This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
1218taking the same parameters, and returning the same information, except that it
1219returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
1220perl keeps that locale category as C<C>, changing it briefly during the
1221operations where the underlying one is required.
1222
1223The other reason it isn't completely a drop-in replacement is that it is
1224declared to return S<C<const char *>>, whereas the system setlocale omits the
1225C<const>. (If it were being written today, plain setlocale would be declared
1226const, since it is illegal to change the information it returns; doing so leads
1227to segfaults.)
1228
1229C<Perl_setlocale> should not be used to change the locale except on systems
1230where the predefined variable C<${^SAFE_LOCALES}> is 1.
1231
1232The return points to a per-thread static buffer, which is overwritten the next
1233time C<Perl_setlocale> is called from the same thread.
1234
1235=cut
1236
1237*/
1238
1239const char *
1240Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
1241{
1242 /* This wraps POSIX::setlocale() */
1243
9aac5db8
KW
1244 const char * retval;
1245 const char * newlocale;
1246 dSAVEDERRNO;
1247 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
1248 dTHX;
1249
a4f00dcc
KW
1250#ifdef USE_LOCALE_NUMERIC
1251
291a84fb
KW
1252 /* A NULL locale means only query what the current one is. We have the
1253 * LC_NUMERIC name saved, because we are normally switched into the C
1254 * locale for it. For an LC_ALL query, switch back to get the correct
1255 * results. All other categories don't require special handling */
a4f00dcc
KW
1256 if (locale == NULL) {
1257 if (category == LC_NUMERIC) {
9aac5db8
KW
1258
1259 /* We don't have to copy this return value, as it is a per-thread
1260 * variable, and won't change until a future setlocale */
1261 return PL_numeric_name;
a4f00dcc
KW
1262 }
1263
7d4bcc4a 1264# ifdef LC_ALL
a4f00dcc 1265
9aac5db8
KW
1266 else if (category == LC_ALL) {
1267 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
1268 }
1269
7d4bcc4a 1270# endif
a4f00dcc
KW
1271
1272 }
1273
1274#endif
1275
9aac5db8
KW
1276 retval = do_setlocale_r(category, locale);
1277 SAVE_ERRNO;
1278
1279#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
1280
1281 if (locale == NULL && category == LC_ALL) {
1282 RESTORE_LC_NUMERIC();
1283 }
1284
1285#endif
a4f00dcc
KW
1286
1287 DEBUG_L(PerlIO_printf(Perl_debug_log,
1288 "%s:%d: %s\n", __FILE__, __LINE__,
1289 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 1290
9aac5db8
KW
1291 RESTORE_ERRNO;
1292
1293 if (! retval) {
a4f00dcc
KW
1294 return NULL;
1295 }
1296
9aac5db8
KW
1297 save_to_buffer(retval, &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
1298 retval = PL_setlocale_buf;
1299
1300 /* If locale == NULL, we are just querying the state */
a4f00dcc 1301 if (locale == NULL) {
a4f00dcc
KW
1302 return retval;
1303 }
a4f00dcc 1304
1159483a
KW
1305 /* Now that have switched locales, we have to update our records to
1306 * correspond. */
a4f00dcc 1307
1159483a 1308 switch (category) {
a4f00dcc 1309
1159483a 1310#ifdef USE_LOCALE_CTYPE
a4f00dcc 1311
1159483a
KW
1312 case LC_CTYPE:
1313 new_ctype(retval);
1314 break;
a4f00dcc 1315
1159483a 1316#endif
a4f00dcc
KW
1317#ifdef USE_LOCALE_COLLATE
1318
1159483a
KW
1319 case LC_COLLATE:
1320 new_collate(retval);
1321 break;
a4f00dcc 1322
1159483a 1323#endif
a4f00dcc
KW
1324#ifdef USE_LOCALE_NUMERIC
1325
1159483a
KW
1326 case LC_NUMERIC:
1327 new_numeric(retval);
1328 break;
a4f00dcc 1329
1159483a
KW
1330#endif
1331#ifdef LC_ALL
a4f00dcc 1332
1159483a 1333 case LC_ALL:
a4f00dcc 1334
1159483a
KW
1335 /* LC_ALL updates all the things we care about. The values may not
1336 * be the same as 'retval', as the locale "" may have set things
1337 * individually */
a4f00dcc 1338
1159483a 1339# ifdef USE_LOCALE_CTYPE
a4f00dcc 1340
1159483a
KW
1341 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1342 new_ctype(newlocale);
a4f00dcc 1343
1159483a
KW
1344# endif /* USE_LOCALE_CTYPE */
1345# ifdef USE_LOCALE_COLLATE
1346
1347 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1348 new_collate(newlocale);
a4f00dcc 1349
7d4bcc4a 1350# endif
1159483a 1351# ifdef USE_LOCALE_NUMERIC
a4f00dcc 1352
1159483a
KW
1353 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1354 new_numeric(newlocale);
a4f00dcc 1355
1159483a
KW
1356# endif /* USE_LOCALE_NUMERIC */
1357#endif /* LC_ALL */
a4f00dcc 1358
1159483a
KW
1359 default:
1360 break;
a4f00dcc
KW
1361 }
1362
1363 return retval;
1364
f7416781
KW
1365}
1366
1367PERL_STATIC_INLINE const char *
1368S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1369{
1370 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1371 * growing it if necessary */
1372
1373 const Size_t string_size = strlen(string) + offset + 1;
1374
1375 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1376
1377 if (*buf_size == 0) {
1378 Newx(*buf, string_size, char);
1379 *buf_size = string_size;
1380 }
1381 else if (string_size > *buf_size) {
1382 Renew(*buf, string_size, char);
1383 *buf_size = string_size;
1384 }
1385
1386 Copy(string, *buf + offset, string_size - offset, char);
1387 return *buf;
1388}
1389
1390/*
1391
f7416781
KW
1392=for apidoc Perl_langinfo
1393
7d4bcc4a 1394This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
1395taking the same C<item> parameter values, and returning the same information.
1396But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1397of Perl's locale handling from your code, and can be used on systems that lack
1398a native C<nl_langinfo>.
1399
1400Expanding on these:
1401
1402=over
1403
1404=item *
1405
1406It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1407without you having to write extra code. The reason for the extra code would be
1408because these are from the C<LC_NUMERIC> locale category, which is normally
1409kept set to the C locale by Perl, no matter what the underlying locale is
1410supposed to be, and so to get the expected results, you have to temporarily
1411toggle into the underlying locale, and later toggle back. (You could use
1412plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1413but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1414keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1415expecting the radix (decimal point) character to be a dot.)
1416
1417=item *
1418
1419Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1420makes your code more portable. Of the fifty-some possible items specified by
1421the POSIX 2008 standard,
1422L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1423only two are completely unimplemented. It uses various techniques to recover
1424the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1425both of which are specified in C89, so should be always be available. Later
1426C<strftime()> versions have additional capabilities; C<""> is returned for
1427those not available on your system.
1428
9f0bc911
KW
1429It is important to note that on such systems, this calls C<localeconv>, and so
1430overwrites the static buffer returned from previous explicit calls to that
1431function. Thus, if the program doesn't use or save the information from an
1432explicit C<localeconv> call (which good practice suggests should be done
1433anyway), use of this function can break it.
1434
f7416781
KW
1435The details for those items which may differ from what this emulation returns
1436and what a native C<nl_langinfo()> would return are:
1437
1438=over
1439
1440=item C<CODESET>
1441
1442=item C<ERA>
1443
1444Unimplemented, so returns C<"">.
1445
1446=item C<YESEXPR>
1447
c1566110
KW
1448=item C<YESSTR>
1449
f7416781
KW
1450=item C<NOEXPR>
1451
c1566110
KW
1452=item C<NOSTR>
1453
1454Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
1455removed from POSIX 2008, and are retained for backwards compatibility. Your
1456platform's C<nl_langinfo> may not support them.
f7416781
KW
1457
1458=item C<D_FMT>
1459
1460Always evaluates to C<%x>, the locale's appropriate date representation.
1461
1462=item C<T_FMT>
1463
1464Always evaluates to C<%X>, the locale's appropriate time representation.
1465
1466=item C<D_T_FMT>
1467
1468Always evaluates to C<%c>, the locale's appropriate date and time
1469representation.
1470
1471=item C<CRNCYSTR>
1472
1473The return may be incorrect for those rare locales where the currency symbol
1474replaces the radix character.
1475Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1476to work differently.
1477
1478=item C<ALT_DIGITS>
1479
1480Currently this gives the same results as Linux does.
1481Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1482to work differently.
1483
1484=item C<ERA_D_FMT>
1485
1486=item C<ERA_T_FMT>
1487
1488=item C<ERA_D_T_FMT>
1489
1490=item C<T_FMT_AMPM>
1491
1492These are derived by using C<strftime()>, and not all versions of that function
1493know about them. C<""> is returned for these on such systems.
1494
1495=back
1496
1497When using C<Perl_langinfo> on systems that don't have a native
1498C<nl_langinfo()>, you must
1499
1500 #include "perl_langinfo.h"
1501
1502before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1503C<#include> with this one. (Doing it this way keeps out the symbols that plain
1504C<langinfo.h> imports into the namespace for code that doesn't need it.)
1505
1506You also should not use the bare C<langinfo.h> item names, but should preface
1507them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1508The C<PERL_I<foo>> versions will also work for this function on systems that do
1509have a native C<nl_langinfo>.
1510
1511=item *
1512
1513It is thread-friendly, returning its result in a buffer that won't be
1514overwritten by another thread, so you don't have to code for that possibility.
1515The buffer can be overwritten by the next call to C<nl_langinfo> or
1516C<Perl_langinfo> in the same thread.
1517
1518=item *
1519
7d4bcc4a 1520ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
1521*>>, but you are (only by documentation) forbidden to write into the buffer.
1522By declaring this C<const>, the compiler enforces this restriction. The extra
1523C<const> is why this isn't an unequivocal drop-in replacement for
1524C<nl_langinfo>.
1525
1526=back
1527
1528The original impetus for C<Perl_langinfo()> was so that code that needs to
1529find out the current currency symbol, floating point radix character, or digit
1530grouping separator can use, on all systems, the simpler and more
1531thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1532pain to make thread-friendly. For other fields returned by C<localeconv>, it
1533is better to use the methods given in L<perlcall> to call
1534L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1535
1536=cut
1537
1538*/
1539
1540const char *
1541#ifdef HAS_NL_LANGINFO
1542Perl_langinfo(const nl_item item)
1543#else
1544Perl_langinfo(const int item)
1545#endif
1546{
f61748ac
KW
1547 return my_nl_langinfo(item, TRUE);
1548}
1549
1550const char *
1551#ifdef HAS_NL_LANGINFO
1552S_my_nl_langinfo(const nl_item item, bool toggle)
1553#else
1554S_my_nl_langinfo(const int item, bool toggle)
1555#endif
1556{
ae74815b 1557 dTHX;
f7416781 1558
5a854ab3
KW
1559 /* We only need to toggle into the underlying LC_NUMERIC locale for these
1560 * two items, and only if not already there */
1561 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
1562 || PL_numeric_underlying))
1563 {
1564 toggle = FALSE;
1565 }
1566
ab340fff
KW
1567#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1568#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1569
ab340fff 1570 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1571 * for those items dependent on it. This must be copied to a buffer before
1572 * switching back, as some systems destroy the buffer when setlocale() is
1573 * called */
f7416781 1574
038d3702
KW
1575 {
1576 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
1577
b2fee59c 1578 if (toggle) {
038d3702 1579 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 1580 }
f7416781 1581
5acc4454
KW
1582 LOCALE_LOCK; /* Prevent interference from another thread executing
1583 this code section (the only call to nl_langinfo in
1584 the core) */
1585
b2fee59c
KW
1586 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf,
1587 &PL_langinfo_bufsize, 0);
f7416781 1588
5acc4454
KW
1589 LOCALE_UNLOCK;
1590
b2fee59c 1591 if (toggle) {
038d3702 1592 RESTORE_LC_NUMERIC();
b2fee59c 1593 }
f7416781
KW
1594 }
1595
ab340fff
KW
1596# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1597
5a854ab3 1598 {
b2fee59c
KW
1599 bool do_free = FALSE;
1600 locale_t cur = uselocale((locale_t) 0);
ab340fff 1601
b2fee59c
KW
1602 if (cur == LC_GLOBAL_LOCALE) {
1603 cur = duplocale(LC_GLOBAL_LOCALE);
1604 do_free = TRUE;
1605 }
ab340fff 1606
b2fee59c 1607 if (toggle) {
e1aa2579
KW
1608 if (PL_underlying_numeric_obj) {
1609 cur = PL_underlying_numeric_obj;
1610 }
1611 else {
1612 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1613 do_free = TRUE;
1614 }
b2fee59c 1615 }
ab340fff 1616
b2fee59c
KW
1617 save_to_buffer(nl_langinfo_l(item, cur),
1618 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1619 if (do_free) {
1620 freelocale(cur);
1621 }
5a854ab3 1622 }
ab340fff 1623
c1566110
KW
1624# endif
1625
1626 if (strEQ(PL_langinfo_buf, "")) {
1627 if (item == PERL_YESSTR) {
1628 return "yes";
1629 }
1630 if (item == PERL_NOSTR) {
1631 return "no";
1632 }
1633 }
1634
ab340fff
KW
1635 return PL_langinfo_buf;
1636
f7416781 1637#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
1638
1639 {
1640
f7416781
KW
1641# ifdef HAS_LOCALECONV
1642
43dd6b15 1643 const struct lconv* lc;
038d3702 1644 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
1645
1646# endif
1647# ifdef HAS_STRFTIME
1648
43dd6b15
KW
1649 struct tm tm;
1650 bool return_format = FALSE; /* Return the %format, not the value */
1651 const char * format;
f7416781
KW
1652
1653# endif
1654
43dd6b15
KW
1655 /* We copy the results to a per-thread buffer, even if not
1656 * multi-threaded. This is in part to simplify this code, and partly
1657 * because we need a buffer anyway for strftime(), and partly because a
1658 * call of localeconv() could otherwise wipe out the buffer, and the
1659 * programmer would not be expecting this, as this is a nl_langinfo()
1660 * substitute after all, so s/he might be thinking their localeconv()
1661 * is safe until another localeconv() call. */
f7416781 1662
43dd6b15
KW
1663 switch (item) {
1664 Size_t len;
1665 const char * retval;
f7416781 1666
43dd6b15
KW
1667 /* These 2 are unimplemented */
1668 case PERL_CODESET:
1669 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 1670
43dd6b15
KW
1671 default:
1672 return "";
f7416781 1673
43dd6b15
KW
1674 /* We use only an English set, since we don't know any more */
1675 case PERL_YESEXPR: return "^[+1yY]";
1676 case PERL_YESSTR: return "yes";
1677 case PERL_NOEXPR: return "^[-0nN]";
1678 case PERL_NOSTR: return "no";
f7416781
KW
1679
1680# ifdef HAS_LOCALECONV
1681
43dd6b15 1682 case PERL_CRNCYSTR:
f7416781 1683
43dd6b15
KW
1684 /* We don't bother with localeconv_l() because any system that
1685 * has it is likely to also have nl_langinfo() */
291a84fb 1686
5acc4454
KW
1687 LOCALE_LOCK; /* Prevent interference with other threads
1688 using localeconv() */
1689
43dd6b15
KW
1690 lc = localeconv();
1691 if ( ! lc
1692 || ! lc->currency_symbol
1693 || strEQ("", lc->currency_symbol))
1694 {
1695 LOCALE_UNLOCK;
1696 return "";
1697 }
f7416781 1698
43dd6b15
KW
1699 /* Leave the first spot empty to be filled in below */
1700 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1701 &PL_langinfo_bufsize, 1);
1702 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1703 { /* khw couldn't figure out how the localedef specifications
1704 would show that the $ should replace the radix; this is
1705 just a guess as to how it might work.*/
1706 *PL_langinfo_buf = '.';
1707 }
1708 else if (lc->p_cs_precedes) {
1709 *PL_langinfo_buf = '-';
1710 }
1711 else {
1712 *PL_langinfo_buf = '+';
1713 }
f7416781 1714
43dd6b15
KW
1715 LOCALE_UNLOCK;
1716 break;
f7416781 1717
43dd6b15
KW
1718 case PERL_RADIXCHAR:
1719 case PERL_THOUSEP:
f7416781 1720
43dd6b15 1721 if (toggle) {
038d3702 1722 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 1723 }
f7416781 1724
5acc4454
KW
1725 LOCALE_LOCK; /* Prevent interference with other threads
1726 using localeconv() */
1727
43dd6b15
KW
1728 lc = localeconv();
1729 if (! lc) {
33394adc
KW
1730 retval = "";
1731 }
43dd6b15
KW
1732 else {
1733 retval = (item == PERL_RADIXCHAR)
1734 ? lc->decimal_point
1735 : lc->thousands_sep;
1736 if (! retval) {
1737 retval = "";
1738 }
1739 }
f7416781 1740
43dd6b15
KW
1741 save_to_buffer(retval, &PL_langinfo_buf,
1742 &PL_langinfo_bufsize, 0);
f7416781 1743
5acc4454
KW
1744 LOCALE_UNLOCK;
1745
43dd6b15 1746 if (toggle) {
038d3702 1747 RESTORE_LC_NUMERIC();
43dd6b15 1748 }
f7416781 1749
43dd6b15 1750 break;
f7416781
KW
1751
1752# endif
1753# ifdef HAS_STRFTIME
1754
43dd6b15
KW
1755 /* These are defined by C89, so we assume that strftime supports
1756 * them, and so are returned unconditionally; they may not be what
1757 * the locale actually says, but should give good enough results
1758 * for someone using them as formats (as opposed to trying to parse
1759 * them to figure out what the locale says). The other format
1760 * items are actually tested to verify they work on the platform */
1761 case PERL_D_FMT: return "%x";
1762 case PERL_T_FMT: return "%X";
1763 case PERL_D_T_FMT: return "%c";
1764
1765 /* These formats are only available in later strfmtime's */
1766 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1767 case PERL_T_FMT_AMPM:
1768
1769 /* The rest can be gotten from most versions of strftime(). */
1770 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1771 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1772 case PERL_ABDAY_7:
1773 case PERL_ALT_DIGITS:
1774 case PERL_AM_STR: case PERL_PM_STR:
1775 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1776 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1777 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1778 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1779 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1780 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1781 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1782 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1783 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
1784 case PERL_MON_12:
1785
1786 LOCALE_LOCK;
1787
1788 init_tm(&tm); /* Precaution against core dumps */
1789 tm.tm_sec = 30;
1790 tm.tm_min = 30;
1791 tm.tm_hour = 6;
1792 tm.tm_year = 2017 - 1900;
1793 tm.tm_wday = 0;
1794 tm.tm_mon = 0;
1795 switch (item) {
1796 default:
1797 LOCALE_UNLOCK;
1798 Perl_croak(aTHX_
1799 "panic: %s: %d: switch case: %d problem",
1800 __FILE__, __LINE__, item);
1801 NOT_REACHED; /* NOTREACHED */
1802
1803 case PERL_PM_STR: tm.tm_hour = 18;
1804 case PERL_AM_STR:
1805 format = "%p";
1806 break;
1807
1808 case PERL_ABDAY_7: tm.tm_wday++;
1809 case PERL_ABDAY_6: tm.tm_wday++;
1810 case PERL_ABDAY_5: tm.tm_wday++;
1811 case PERL_ABDAY_4: tm.tm_wday++;
1812 case PERL_ABDAY_3: tm.tm_wday++;
1813 case PERL_ABDAY_2: tm.tm_wday++;
1814 case PERL_ABDAY_1:
1815 format = "%a";
1816 break;
1817
1818 case PERL_DAY_7: tm.tm_wday++;
1819 case PERL_DAY_6: tm.tm_wday++;
1820 case PERL_DAY_5: tm.tm_wday++;
1821 case PERL_DAY_4: tm.tm_wday++;
1822 case PERL_DAY_3: tm.tm_wday++;
1823 case PERL_DAY_2: tm.tm_wday++;
1824 case PERL_DAY_1:
1825 format = "%A";
1826 break;
1827
1828 case PERL_ABMON_12: tm.tm_mon++;
1829 case PERL_ABMON_11: tm.tm_mon++;
1830 case PERL_ABMON_10: tm.tm_mon++;
1831 case PERL_ABMON_9: tm.tm_mon++;
1832 case PERL_ABMON_8: tm.tm_mon++;
1833 case PERL_ABMON_7: tm.tm_mon++;
1834 case PERL_ABMON_6: tm.tm_mon++;
1835 case PERL_ABMON_5: tm.tm_mon++;
1836 case PERL_ABMON_4: tm.tm_mon++;
1837 case PERL_ABMON_3: tm.tm_mon++;
1838 case PERL_ABMON_2: tm.tm_mon++;
1839 case PERL_ABMON_1:
1840 format = "%b";
1841 break;
1842
1843 case PERL_MON_12: tm.tm_mon++;
1844 case PERL_MON_11: tm.tm_mon++;
1845 case PERL_MON_10: tm.tm_mon++;
1846 case PERL_MON_9: tm.tm_mon++;
1847 case PERL_MON_8: tm.tm_mon++;
1848 case PERL_MON_7: tm.tm_mon++;
1849 case PERL_MON_6: tm.tm_mon++;
1850 case PERL_MON_5: tm.tm_mon++;
1851 case PERL_MON_4: tm.tm_mon++;
1852 case PERL_MON_3: tm.tm_mon++;
1853 case PERL_MON_2: tm.tm_mon++;
1854 case PERL_MON_1:
1855 format = "%B";
1856 break;
1857
1858 case PERL_T_FMT_AMPM:
1859 format = "%r";
1860 return_format = TRUE;
1861 break;
1862
1863 case PERL_ERA_D_FMT:
1864 format = "%Ex";
1865 return_format = TRUE;
1866 break;
1867
1868 case PERL_ERA_T_FMT:
1869 format = "%EX";
1870 return_format = TRUE;
1871 break;
1872
1873 case PERL_ERA_D_T_FMT:
1874 format = "%Ec";
1875 return_format = TRUE;
1876 break;
1877
1878 case PERL_ALT_DIGITS:
1879 tm.tm_wday = 0;
1880 format = "%Ow"; /* Find the alternate digit for 0 */
1881 break;
1882 }
f7416781 1883
43dd6b15
KW
1884 /* We can't use my_strftime() because it doesn't look at
1885 * tm_wday */
1886 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1887 format, &tm))
1888 {
1889 /* A zero return means one of:
1890 * a) there wasn't enough space in PL_langinfo_buf
1891 * b) the format, like a plain %p, returns empty
1892 * c) it was an illegal format, though some
1893 * implementations of strftime will just return the
1894 * illegal format as a plain character sequence.
1895 *
1896 * To quickly test for case 'b)', try again but precede
1897 * the format with a plain character. If that result is
1898 * still empty, the problem is either 'a)' or 'c)' */
1899
1900 Size_t format_size = strlen(format) + 1;
1901 Size_t mod_size = format_size + 1;
1902 char * mod_format;
1903 char * temp_result;
1904
1905 Newx(mod_format, mod_size, char);
1906 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 1907 *mod_format = ' ';
43dd6b15
KW
1908 my_strlcpy(mod_format + 1, format, mod_size);
1909 len = strftime(temp_result,
1910 PL_langinfo_bufsize,
1911 mod_format, &tm);
1912 Safefree(mod_format);
1913 Safefree(temp_result);
1914
1915 /* If 'len' is non-zero, it means that we had a case like
1916 * %p which means the current locale doesn't use a.m. or
1917 * p.m., and that is valid */
1918 if (len == 0) {
1919
1920 /* Here, still didn't work. If we get well beyond a
1921 * reasonable size, bail out to prevent an infinite
1922 * loop. */
1923
1924 if (PL_langinfo_bufsize > 100 * format_size) {
1925 *PL_langinfo_buf = '\0';
1926 }
1927 else {
1928 /* Double the buffer size to retry; Add 1 in case
1929 * original was 0, so we aren't stuck at 0. */
1930 PL_langinfo_bufsize *= 2;
1931 PL_langinfo_bufsize++;
1932 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1933 continue;
1934 }
1935 }
f7416781 1936
f7416781 1937 break;
43dd6b15 1938 }
f7416781 1939
43dd6b15
KW
1940 /* Here, we got a result.
1941 *
1942 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1943 * alternate format for wday 0. If the value is the same as
1944 * the normal 0, there isn't an alternate, so clear the buffer.
1945 * */
1946 if ( item == PERL_ALT_DIGITS
1947 && strEQ(PL_langinfo_buf, "0"))
1948 {
1949 *PL_langinfo_buf = '\0';
1950 }
f7416781 1951
43dd6b15
KW
1952 /* ALT_DIGITS is problematic. Experiments on it showed that
1953 * strftime() did not always work properly when going from
1954 * alt-9 to alt-10. Only a few locales have this item defined,
1955 * and in all of them on Linux that khw was able to find,
1956 * nl_langinfo() merely returned the alt-0 character, possibly
1957 * doubled. Most Unicode digits are in blocks of 10
1958 * consecutive code points, so that is sufficient information
1959 * for those scripts, as we can infer alt-1, alt-2, .... But
1960 * for a Japanese locale, a CJK ideographic 0 is returned, and
1961 * the CJK digits are not in code point order, so you can't
1962 * really infer anything. The localedef for this locale did
1963 * specify the succeeding digits, so that strftime() works
1964 * properly on them, without needing to infer anything. But
1965 * the nl_langinfo() return did not give sufficient information
1966 * for the caller to understand what's going on. So until
1967 * there is evidence that it should work differently, this
1968 * returns the alt-0 string for ALT_DIGITS.
1969 *
1970 * wday was chosen because its range is all a single digit.
1971 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 1972
43dd6b15 1973 LOCALE_UNLOCK;
f7416781 1974
43dd6b15
KW
1975 /* If to return the format, not the value, overwrite the buffer
1976 * with it. But some strftime()s will keep the original format
1977 * if illegal, so change those to "" */
1978 if (return_format) {
1979 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
1980 *PL_langinfo_buf = '\0';
1981 }
43dd6b15
KW
1982 else {
1983 save_to_buffer(format, &PL_langinfo_buf,
1984 &PL_langinfo_bufsize, 0);
f7416781
KW
1985 }
1986 }
1987
1988 break;
f7416781
KW
1989
1990# endif
1991
43dd6b15 1992 }
f7416781
KW
1993 }
1994
1995 return PL_langinfo_buf;
1996
1997#endif
1998
a4f00dcc 1999}
b385bb4d 2000
98994639
HS
2001/*
2002 * Initialize locale awareness.
2003 */
2004int
2005Perl_init_i18nl10n(pTHX_ int printwarn)
2006{
0e92a118
KW
2007 /* printwarn is
2008 *
2009 * 0 if not to output warning when setup locale is bad
2010 * 1 if to output warning based on value of PERL_BADLANG
2011 * >1 if to output regardless of PERL_BADLANG
2012 *
2013 * returns
98994639 2014 * 1 = set ok or not applicable,
0e92a118
KW
2015 * 0 = fallback to a locale of lower priority
2016 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2017 *
2018 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2019 * set, debugging information is output.
2020 *
2021 * This looks more complicated than it is, mainly due to the #ifdefs.
2022 *
2023 * We try to set LC_ALL to the value determined by the environment. If
2024 * there is no LC_ALL on this platform, we try the individual categories we
2025 * know about. If this works, we are done.
2026 *
2027 * But if it doesn't work, we have to do something else. We search the
2028 * environment variables ourselves instead of relying on the system to do
2029 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2030 * think there is one), and the ultimate fallback "C". This is all done in
2031 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2032 * more complex. The 'trial_locales' array is initialized with just one
2033 * element; it causes the behavior described in the paragraph above this to
2034 * happen. If that fails, we add elements to 'trial_locales', and do extra
2035 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2036 *
2037 * On Ultrix, the locale MUST come from the environment, so there is
2038 * preliminary code to set it. I (khw) am not sure that it is necessary,
2039 * and that this couldn't be folded into the loop, but barring any real
2040 * platforms to test on, it's staying as-is
2041 *
2042 * A slight complication is that in embedded Perls, the locale may already
2043 * be set-up, and we don't want to get it from the normal environment
2044 * variables. This is handled by having a special environment variable
2045 * indicate we're in this situation. We simply set setlocale's 2nd
2046 * parameter to be a NULL instead of "". That indicates to setlocale that
2047 * it is not to change anything, but to return the current value,
2048 * effectively initializing perl's db to what the locale already is.
2049 *
2050 * We play the same trick with NULL if a LC_ALL succeeds. We call
2051 * setlocale() on the individual categores with NULL to get their existing
2052 * values for our db, instead of trying to change them.
2053 * */
98994639 2054
0e92a118
KW
2055 int ok = 1;
2056
7d4bcc4a
KW
2057#ifndef USE_LOCALE
2058
2059 PERL_UNUSED_ARG(printwarn);
2060
2061#else /* USE_LOCALE */
7d4bcc4a
KW
2062# ifdef __GLIBC__
2063
175c4cf9 2064 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2065
2066# endif
65ebb059 2067
ccd65d51
KW
2068 /* NULL uses the existing already set up locale */
2069 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2070 ? NULL
2071 : "";
c3fcd832
KW
2072 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2073 unsigned int trial_locales_count;
175c4cf9
KW
2074 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2075 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2076 bool setlocale_failure = FALSE;
65ebb059 2077 unsigned int i;
175c4cf9
KW
2078
2079 /* A later getenv() could zap this, so only use here */
2080 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2081
2082 const bool locwarn = (printwarn > 1
e5f10d49
KW
2083 || ( printwarn
2084 && ( ! bad_lang_use_once
22ff3130 2085 || (
e5f10d49
KW
2086 /* disallow with "" or "0" */
2087 *bad_lang_use_once
2088 && strNE("0", bad_lang_use_once)))));
ea92aad8 2089
291a84fb 2090 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2091 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2092
2093 /* current locale for given category; should have been copied so aren't
2094 * volatile */
8de4332b 2095 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2096
7d4bcc4a
KW
2097# ifdef WIN32
2098
6bce99ee
JH
2099 /* In some systems you can find out the system default locale
2100 * and use that as the fallback locale. */
7d4bcc4a
KW
2101# define SYSTEM_DEFAULT_LOCALE
2102# endif
2103# ifdef SYSTEM_DEFAULT_LOCALE
2104
65ebb059 2105 const char *system_default_locale = NULL;
98994639 2106
7d4bcc4a 2107# endif
948523db
KW
2108
2109# ifndef DEBUGGING
2110# define DEBUG_LOCALE_INIT(a,b,c)
2111# else
7d4bcc4a 2112
8298454c 2113 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2114
2115# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2116 STMT_START { \
2117 if (debug_initialization) { \
2118 PerlIO_printf(Perl_debug_log, \
2119 "%s:%d: %s\n", \
2120 __FILE__, __LINE__, \
a4f00dcc 2121 setlocale_debug_string(category, \
2fcc0ca9
KW
2122 locale, \
2123 result)); \
2124 } \
2125 } STMT_END
2fcc0ca9 2126
948523db
KW
2127/* Make sure the parallel arrays are properly set up */
2128# ifdef USE_LOCALE_NUMERIC
2129 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2130 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
2131# endif
2132# ifdef USE_LOCALE_CTYPE
2133 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2134 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
2135# endif
2136# ifdef USE_LOCALE_COLLATE
2137 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2138 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
2139# endif
2140# ifdef USE_LOCALE_TIME
2141 assert(categories[LC_TIME_INDEX] == LC_TIME);
2142 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2143# endif
2144# ifdef USE_LOCALE_MESSAGES
2145 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2146 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
2147# endif
2148# ifdef USE_LOCALE_MONETARY
2149 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2150 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
2151# endif
9821811f
KW
2152# ifdef USE_LOCALE_ADDRESS
2153 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
2154 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
2155# endif
2156# ifdef USE_LOCALE_IDENTIFICATION
2157 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
2158 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
2159# endif
2160# ifdef USE_LOCALE_MEASUREMENT
2161 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
2162 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
2163# endif
2164# ifdef USE_LOCALE_PAPER
2165 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
2166 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
2167# endif
2168# ifdef USE_LOCALE_TELEPHONE
2169 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
2170 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
2171# endif
948523db
KW
2172# ifdef LC_ALL
2173 assert(categories[LC_ALL_INDEX] == LC_ALL);
2174 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
2175 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
2176# endif
2177# endif /* DEBUGGING */
47280b20
KW
2178
2179 /* Initialize the cache of the program's UTF-8ness for the always known
2180 * locales C and POSIX */
2181 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
2182 sizeof(PL_locale_utf8ness));
2183
3ca88433
KW
2184 PL_numeric_radix_sv = newSVpvs(".");
2185
b56c4436 2186# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
2187
2188 /*
2189 * Ultrix setlocale(..., "") fails if there are no environment
2190 * variables from which to get a locale name.
2191 */
2192
b56c4436
KW
2193# ifndef LC_ALL
2194# error Ultrix without LC_ALL not implemented
2195# else
7d4bcc4a 2196
b56c4436
KW
2197 {
2198 bool done = FALSE;
ea92aad8
KW
2199 if (lang) {
2200 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
2201 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
2202 if (sl_result[LC_ALL_INDEX])
2203 done = TRUE;
2204 else
e5f10d49 2205 setlocale_failure = TRUE;
ea92aad8
KW
2206 }
2207 if (! setlocale_failure) {
2208 const char * locale_param;
2209 for (i = 0; i < LC_ALL_INDEX; i++) {
2210 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
2211 ? setlocale_init
2212 : NULL;
2213 sl_result[i] = do_setlocale_r(categories[i], locale_param);
2214 if (! sl_result[i]) {
2215 setlocale_failure = TRUE;
2216 }
2217 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 2218 }
c835d6be 2219 }
7d4bcc4a
KW
2220 }
2221
2222# endif /* LC_ALL */
e5f10d49 2223# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 2224
65ebb059 2225 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
2226 * the list. Normally the loop is executed just once. But if setting the
2227 * locale fails, inside the loop we add fallback trials to the array and so
2228 * will execute the loop multiple times */
c3fcd832
KW
2229 trial_locales[0] = setlocale_init;
2230 trial_locales_count = 1;
7d4bcc4a 2231
65ebb059
KW
2232 for (i= 0; i < trial_locales_count; i++) {
2233 const char * trial_locale = trial_locales[i];
2234
2235 if (i > 0) {
2236
2237 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
2238 * when i==0, but I (khw) don't think that behavior makes much
2239 * sense */
2240 setlocale_failure = FALSE;
2241
7d4bcc4a 2242# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 2243# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 2244
65ebb059
KW
2245 /* On Windows machines, an entry of "" after the 0th means to use
2246 * the system default locale, which we now proceed to get. */
2247 if (strEQ(trial_locale, "")) {
2248 unsigned int j;
2249
2250 /* Note that this may change the locale, but we are going to do
2251 * that anyway just below */
837ce802 2252 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 2253 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 2254
7d4bcc4a 2255 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
2256 * try */
2257 if (! system_default_locale) {
2258 goto next_iteration;
2259 }
2260 for (j = 0; j < trial_locales_count; j++) {
2261 if (strEQ(system_default_locale, trial_locales[j])) {
2262 goto next_iteration;
2263 }
2264 }
2265
2266 trial_locale = system_default_locale;
2267 }
ec0202b5
KW
2268# else
2269# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
2270# endif
7d4bcc4a 2271# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
2272
2273 } /* For i > 0 */
65ebb059 2274
7d4bcc4a
KW
2275# ifdef LC_ALL
2276
948523db
KW
2277 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
2278 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
2279 if (! sl_result[LC_ALL_INDEX]) {
49c85077 2280 setlocale_failure = TRUE;
7cd8b568
KW
2281 }
2282 else {
2283 /* Since LC_ALL succeeded, it should have changed all the other
2284 * categories it can to its value; so we massage things so that the
2285 * setlocales below just return their category's current values.
2286 * This adequately handles the case in NetBSD where LC_COLLATE may
2287 * not be defined for a locale, and setting it individually will
7d4bcc4a 2288 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
2289 * the POSIX locale. */
2290 trial_locale = NULL;
2291 }
7d4bcc4a
KW
2292
2293# endif /* LC_ALL */
98994639 2294
e5f10d49
KW
2295 if (! setlocale_failure) {
2296 unsigned int j;
2297 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2298 curlocales[j]
2299 = savepv(do_setlocale_r(categories[j], trial_locale));
2300 if (! curlocales[j]) {
2301 setlocale_failure = TRUE;
2302 }
2303 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2304 }
c835d6be 2305
e5f10d49
KW
2306 if (! setlocale_failure) { /* All succeeded */
2307 break; /* Exit trial_locales loop */
49c85077 2308 }
65ebb059 2309 }
98994639 2310
49c85077
KW
2311 /* Here, something failed; will need to try a fallback. */
2312 ok = 0;
65ebb059 2313
49c85077
KW
2314 if (i == 0) {
2315 unsigned int j;
98994639 2316
65ebb059 2317 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2318
2319# ifdef LC_ALL
98994639 2320
49c85077
KW
2321 PerlIO_printf(Perl_error_log,
2322 "perl: warning: Setting locale failed.\n");
98994639 2323
7d4bcc4a 2324# else /* !LC_ALL */
98994639 2325
49c85077
KW
2326 PerlIO_printf(Perl_error_log,
2327 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2328
e5f10d49
KW
2329 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2330 if (! curlocales[j]) {
2331 PerlIO_printf(Perl_error_log, category_names[j]);
2332 }
2333 else {
2334 Safefree(curlocales[j]);
2335 }
2336 }
7d4bcc4a 2337
7d4bcc4a 2338# endif /* LC_ALL */
98994639 2339
49c85077
KW
2340 PerlIO_printf(Perl_error_log,
2341 "perl: warning: Please check that your locale settings:\n");
98994639 2342
7d4bcc4a
KW
2343# ifdef __GLIBC__
2344
49c85077
KW
2345 PerlIO_printf(Perl_error_log,
2346 "\tLANGUAGE = %c%s%c,\n",
2347 language ? '"' : '(',
2348 language ? language : "unset",
2349 language ? '"' : ')');
7d4bcc4a 2350# endif
98994639 2351
49c85077
KW
2352 PerlIO_printf(Perl_error_log,
2353 "\tLC_ALL = %c%s%c,\n",
2354 lc_all ? '"' : '(',
2355 lc_all ? lc_all : "unset",
2356 lc_all ? '"' : ')');
98994639 2357
7d4bcc4a
KW
2358# if defined(USE_ENVIRON_ARRAY)
2359
49c85077 2360 {
cd999af9 2361 char **e;
d5e32b93
KW
2362
2363 /* Look through the environment for any variables of the
2364 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2365 * already handled above. These are assumed to be locale
2366 * settings. Output them and their values. */
cd999af9 2367 for (e = environ; *e; e++) {
d5e32b93
KW
2368 const STRLEN prefix_len = sizeof("LC_") - 1;
2369 STRLEN uppers_len;
2370
cd999af9 2371 if ( strBEGINs(*e, "LC_")
c8b388b0 2372 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2373 && (uppers_len = strspn(*e + prefix_len,
2374 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2375 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2376 {
2377 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2378 (int) (prefix_len + uppers_len), *e,
2379 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2380 }
2381 }
49c85077 2382 }
7d4bcc4a
KW
2383
2384# else
2385
49c85077
KW
2386 PerlIO_printf(Perl_error_log,
2387 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2388
2389# endif
98994639 2390
49c85077
KW
2391 PerlIO_printf(Perl_error_log,
2392 "\tLANG = %c%s%c\n",
2393 lang ? '"' : '(',
2394 lang ? lang : "unset",
2395 lang ? '"' : ')');
98994639 2396
49c85077
KW
2397 PerlIO_printf(Perl_error_log,
2398 " are supported and installed on your system.\n");
2399 }
98994639 2400
65ebb059 2401 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2402 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2403 * usually change the upper bound of the loop we are in.
2404 *
2405 * Since the system's default way of setting the locale has not
2406 * found one that works, We use Perl's defined ordering: LC_ALL,
2407 * LANG, and the C locale. We don't try the same locale twice, so
2408 * don't add to the list if already there. (On POSIX systems, the
2409 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2410 * but there's no harm done by doing it explicitly.
2411 *
2412 * Note that this tries the LC_ALL environment variable even on
2413 * systems which have no LC_ALL locale setting. This may or may
2414 * not have been originally intentional, but there's no real need
2415 * to change the behavior. */
65ebb059
KW
2416 if (lc_all) {
2417 for (j = 0; j < trial_locales_count; j++) {
2418 if (strEQ(lc_all, trial_locales[j])) {
2419 goto done_lc_all;
2420 }
2421 }
2422 trial_locales[trial_locales_count++] = lc_all;
2423 }
2424 done_lc_all:
98994639 2425
65ebb059
KW
2426 if (lang) {
2427 for (j = 0; j < trial_locales_count; j++) {
2428 if (strEQ(lang, trial_locales[j])) {
2429 goto done_lang;
2430 }
2431 }
2432 trial_locales[trial_locales_count++] = lang;
2433 }
2434 done_lang:
2435
7d4bcc4a
KW
2436# if defined(WIN32) && defined(LC_ALL)
2437
65ebb059
KW
2438 /* For Windows, we also try the system default locale before "C".
2439 * (If there exists a Windows without LC_ALL we skip this because
2440 * it gets too complicated. For those, the "C" is the next
2441 * fallback possibility). The "" is the same as the 0th element of
2442 * the array, but the code at the loop above knows to treat it
2443 * differently when not the 0th */
2444 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2445
2446# endif
65ebb059
KW
2447
2448 for (j = 0; j < trial_locales_count; j++) {
2449 if (strEQ("C", trial_locales[j])) {
2450 goto done_C;
2451 }
2452 }
2453 trial_locales[trial_locales_count++] = "C";
98994639 2454
65ebb059
KW
2455 done_C: ;
2456 } /* end of first time through the loop */
98994639 2457
7d4bcc4a
KW
2458# ifdef WIN32
2459
65ebb059 2460 next_iteration: ;
7d4bcc4a
KW
2461
2462# endif
65ebb059
KW
2463
2464 } /* end of looping through the trial locales */
2465
2466 if (ok < 1) { /* If we tried to fallback */
2467 const char* msg;
2468 if (! setlocale_failure) { /* fallback succeeded */
2469 msg = "Falling back to";
2470 }
2471 else { /* fallback failed */
e5f10d49 2472 unsigned int j;
98994639 2473
65ebb059
KW
2474 /* We dropped off the end of the loop, so have to decrement i to
2475 * get back to the value the last time through */
2476 i--;
98994639 2477
65ebb059
KW
2478 ok = -1;
2479 msg = "Failed to fall back to";
2480
2481 /* To continue, we should use whatever values we've got */
7d4bcc4a 2482
e5f10d49
KW
2483 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2484 Safefree(curlocales[j]);
2485 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2486 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2487 }
65ebb059
KW
2488 }
2489
2490 if (locwarn) {
2491 const char * description;
2492 const char * name = "";
2493 if (strEQ(trial_locales[i], "C")) {
2494 description = "the standard locale";
2495 name = "C";
2496 }
7d4bcc4a
KW
2497
2498# ifdef SYSTEM_DEFAULT_LOCALE
2499
65ebb059
KW
2500 else if (strEQ(trial_locales[i], "")) {
2501 description = "the system default locale";
2502 if (system_default_locale) {
2503 name = system_default_locale;
2504 }
2505 }
7d4bcc4a
KW
2506
2507# endif /* SYSTEM_DEFAULT_LOCALE */
2508
65ebb059
KW
2509 else {
2510 description = "a fallback locale";
2511 name = trial_locales[i];
2512 }
2513 if (name && strNE(name, "")) {
2514 PerlIO_printf(Perl_error_log,
2515 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2516 }
2517 else {
2518 PerlIO_printf(Perl_error_log,
2519 "perl: warning: %s %s.\n", msg, description);
2520 }
2521 }
2522 } /* End of tried to fallback */
98994639 2523
e5f10d49
KW
2524 /* Done with finding the locales; update our records */
2525
7d4bcc4a
KW
2526# ifdef USE_LOCALE_CTYPE
2527
948523db 2528 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2529
e5f10d49 2530# endif
7d4bcc4a
KW
2531# ifdef USE_LOCALE_COLLATE
2532
948523db 2533 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2534
e5f10d49 2535# endif
7d4bcc4a
KW
2536# ifdef USE_LOCALE_NUMERIC
2537
948523db 2538 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2539
2540# endif
2541
948523db 2542 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20
KW
2543
2544# if defined(USE_ITHREADS)
2545
2546 /* This caches whether each category's locale is UTF-8 or not. This
2547 * may involve changing the locale. It is ok to do this at
2548 * initialization time before any threads have started, but not later.
2549 * Caching means that if the program heeds our dictate not to change
2550 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
2551 * it may get queried without having to change locales. If the
2552 * environment is such that all categories have the same locale, this
2553 * isn't needed, as the code will not change the locale; but this
2554 * handles the uncommon case where the environment has disparate
2555 * locales for the categories */
47280b20
KW
2556 (void) _is_cur_LC_category_utf8(categories[i]);
2557
2558# endif
2559
e5f10d49
KW
2560 Safefree(curlocales[i]);
2561 }
b310b053 2562
7d4bcc4a
KW
2563# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2564
49c85077 2565 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
2566 * locale is UTF-8. The call to new_ctype() just above has already
2567 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
2568 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
2569 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
2570 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
2571 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 2572
a05d7ebb 2573 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2574 This is an alternative to using the -C command line switch
2575 (the -C if present will override this). */
2576 {
dd374669 2577 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2578 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2579 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2580 PL_utf8cache = -1;
b310b053
JH
2581 }
2582
7d4bcc4a 2583# endif
7d4bcc4a
KW
2584# ifdef __GLIBC__
2585
175c4cf9 2586 Safefree(language);
7d4bcc4a
KW
2587
2588# endif
175c4cf9
KW
2589
2590 Safefree(lc_all);
2591 Safefree(lang);
2592
e3305790 2593#endif /* USE_LOCALE */
2fcc0ca9 2594#ifdef DEBUGGING
7d4bcc4a 2595
2fcc0ca9 2596 /* So won't continue to output stuff */
27cdc72e 2597 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2598
2fcc0ca9
KW
2599#endif
2600
98994639
HS
2601 return ok;
2602}
2603
98994639
HS
2604#ifdef USE_LOCALE_COLLATE
2605
a4a439fb 2606char *
a4a439fb
KW
2607Perl__mem_collxfrm(pTHX_ const char *input_string,
2608 STRLEN len, /* Length of 'input_string' */
2609 STRLEN *xlen, /* Set to length of returned string
2610 (not including the collation index
2611 prefix) */
2612 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2613 )
98994639 2614{
a4a439fb
KW
2615
2616 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2617 * differences. First, it handles embedded NULs. Second, it allocates a bit
2618 * more memory than needed for the transformed data itself. The real
55e5378d 2619 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2620 * the length of that, and doesn't include the collation index size.
2621 * Please see sv_collxfrm() to see how this is used. */
2622
55e5378d
KW
2623#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2624
6696cfa7
KW
2625 char * s = (char *) input_string;
2626 STRLEN s_strlen = strlen(input_string);
79f120c8 2627 char *xbuf = NULL;
55e5378d 2628 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2629 STRLEN length_in_chars;
c664130f 2630 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2631
a4a439fb
KW
2632 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2633
2634 /* Must be NUL-terminated */
2635 assert(*(input_string + len) == '\0');
7918f24d 2636
79f120c8
KW
2637 /* If this locale has defective collation, skip */
2638 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2639 DEBUG_L(PerlIO_printf(Perl_debug_log,
2640 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2641 goto bad;
2642 }
2643
6696cfa7
KW
2644 /* Replace any embedded NULs with the control that sorts before any others.
2645 * This will give as good as possible results on strings that don't
2646 * otherwise contain that character, but otherwise there may be
2647 * less-than-perfect results with that character and NUL. This is
fdc080f3 2648 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2649 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2650 NUL */
6696cfa7
KW
2651 char * e = s + len;
2652 char * sans_nuls;
fdc080f3 2653 STRLEN sans_nuls_len;
94762aa0 2654 int try_non_controls;
afc4976f
KW
2655 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2656 making sure 2nd byte is NUL.
2657 */
2658 STRLEN this_replacement_len;
2659
1e4c9676
KW
2660 /* If we don't know what non-NUL control character sorts lowest for
2661 * this locale, find it */
f28f4d2a 2662 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2663 int j;
afc4976f 2664 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2665 includes the collation index
2666 prefixed. */
2667
91c0e2e0 2668 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2669
2670 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2671 * NUL, in which case we instead look for any character. Controls
2672 * are preferred because collation order is, in general, context
2673 * sensitive, with adjoining characters affecting the order, and
2674 * controls are less likely to have such interactions, allowing the
2675 * NUL-replacement to stand on its own. (Another way to look at it
2676 * is to imagine what would happen if the NUL were replaced by a
2677 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2678 for (try_non_controls = 0;
2679 try_non_controls < 2;
2680 try_non_controls++)
2681 {
d4ff9586
KW
2682 /* Look through all legal code points (NUL isn't) */
2683 for (j = 1; j < 256; j++) {
2684 char * x; /* j's xfrm plus collation index */
2685 STRLEN x_len; /* length of 'x' */
2686 STRLEN trial_len = 1;
736a4fed 2687 char cur_source[] = { '\0', '\0' };
d4ff9586 2688
736a4fed
KW
2689 /* Skip non-controls the first time through the loop. The
2690 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2691 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2692 ? ! isCNTRL_L1(j)
2693 : ! isCNTRL_LC(j))
2694 {
d4ff9586 2695 continue;
6696cfa7 2696 }
6696cfa7 2697
736a4fed
KW
2698 /* Create a 1-char string of the current code point */
2699 cur_source[0] = (char) j;
2700
d4ff9586
KW
2701 /* Then transform it */
2702 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2703 0 /* The string is not in UTF-8 */);
6696cfa7 2704
1e4c9676 2705 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2706 * */
2707 if (! x) {
2708 continue;
2709 }
6696cfa7 2710
d4ff9586
KW
2711 /* If this character's transformation is lower than
2712 * the current lowest, this one becomes the lowest */
2713 if ( cur_min_x == NULL
2714 || strLT(x + COLLXFRM_HDR_LEN,
2715 cur_min_x + COLLXFRM_HDR_LEN))
2716 {
f28f4d2a 2717 PL_strxfrm_NUL_replacement = j;
d4ff9586 2718 cur_min_x = x;
d4ff9586
KW
2719 }
2720 else {
2721 Safefree(x);
2722 }
1e4c9676 2723 } /* end of loop through all 255 characters */
6696cfa7 2724
1e4c9676 2725 /* Stop looking if found */
94762aa0
KW
2726 if (cur_min_x) {
2727 break;
2728 }
2729
2730 /* Unlikely, but possible, if there aren't any controls that
2731 * work in the locale, repeat the loop, looking for any
2732 * character that works */
2733 DEBUG_L(PerlIO_printf(Perl_debug_log,
2734 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2735 } /* End of loop to try first the controls, then any char */
6696cfa7 2736
94762aa0
KW
2737 if (! cur_min_x) {
2738 DEBUG_L(PerlIO_printf(Perl_debug_log,
2739 "_mem_collxfrm: Couldn't find any character to replace"
2740 " embedded NULs in locale %s with", PL_collation_name));
2741 goto bad;
58eebef2
KW
2742 }
2743
94762aa0
KW
2744 DEBUG_L(PerlIO_printf(Perl_debug_log,
2745 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2746 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2747
6696cfa7 2748 Safefree(cur_min_x);
1e4c9676 2749 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2750
2751 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2752 * UTF8-ness of the original */
f28f4d2a
KW
2753 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2754 this_replacement_char[0] =
2755 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2756 this_replacement_char[1] =
2757 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2758 this_replacement_len = 2;
2759 }
2760 else {
f28f4d2a 2761 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2762 /* this_replacement_char[1] = '\0' was done at initialization */
2763 this_replacement_len = 1;
6696cfa7
KW
2764 }
2765
2766 /* The worst case length for the replaced string would be if every
2767 * character in it is NUL. Multiply that by the length of each
2768 * replacement, and allow for a trailing NUL */
afc4976f 2769 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2770 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2771 *sans_nuls = '\0';
2772
6696cfa7
KW
2773 /* Replace each NUL with the lowest collating control. Loop until have
2774 * exhausted all the NULs */
2775 while (s + s_strlen < e) {
6069d6c5 2776 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2777
2778 /* Do the actual replacement */
6069d6c5 2779 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2780
2781 /* Move past the input NUL */
2782 s += s_strlen + 1;
2783 s_strlen = strlen(s);
2784 }
2785
2786 /* And add anything that trails the final NUL */
6069d6c5 2787 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2788
2789 /* Switch so below we transform this modified string */
2790 s = sans_nuls;
2791 len = strlen(s);
1e4c9676 2792 } /* End of replacing NULs */
6696cfa7 2793
a4a439fb
KW
2794 /* Make sure the UTF8ness of the string and locale match */
2795 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 2796 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
2797 const char * const t = s; /* Temporary so we can later find where the
2798 input was */
2799
2800 /* Here they don't match. Change the string's to be what the locale is
2801 * expecting */
2802
2803 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2804 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2805 utf8 = TRUE;
2806 }
2807 else { /* locale is not UTF-8; but input is; downgrade the input */
2808
2809 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2810
2811 /* If the downgrade was successful we are done, but if the input
2812 * contains things that require UTF-8 to represent, have to do
2813 * damage control ... */
2814 if (UNLIKELY(utf8)) {
2815
2816 /* What we do is construct a non-UTF-8 string with
2817 * 1) the characters representable by a single byte converted
2818 * to be so (if necessary);
2819 * 2) and the rest converted to collate the same as the
2820 * highest collating representable character. That makes
2821 * them collate at the end. This is similar to how we
2822 * handle embedded NULs, but we use the highest collating
2823 * code point instead of the smallest. Like the NUL case,
2824 * this isn't perfect, but is the best we can reasonably
2825 * do. Every above-255 code point will sort the same as
2826 * the highest-sorting 0-255 code point. If that code
2827 * point can combine in a sequence with some other code
2828 * points for weight calculations, us changing something to
2829 * be it can adversely affect the results. But in most
2830 * cases, it should work reasonably. And note that this is
2831 * really an illegal situation: using code points above 255
2832 * on a locale where only 0-255 are valid. If two strings
2833 * sort entirely equal, then the sort order for the
2834 * above-255 code points will be in code point order. */
2835
2836 utf8 = FALSE;
2837
2838 /* If we haven't calculated the code point with the maximum
2839 * collating order for this locale, do so now */
2840 if (! PL_strxfrm_max_cp) {
2841 int j;
2842
2843 /* The current transformed string that collates the
2844 * highest (except it also includes the prefixed collation
2845 * index. */
2846 char * cur_max_x = NULL;
2847
2848 /* Look through all legal code points (NUL isn't) */
2849 for (j = 1; j < 256; j++) {
2850 char * x;
2851 STRLEN x_len;
736a4fed 2852 char cur_source[] = { '\0', '\0' };
a4a439fb 2853
736a4fed
KW
2854 /* Create a 1-char string of the current code point */
2855 cur_source[0] = (char) j;
a4a439fb
KW
2856
2857 /* Then transform it */
2858 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2859
2860 /* If something went wrong (which it shouldn't), just
2861 * ignore this code point */
94762aa0 2862 if (! x) {
a4a439fb
KW
2863 continue;
2864 }
2865
2866 /* If this character's transformation is higher than
2867 * the current highest, this one becomes the highest */
2868 if ( cur_max_x == NULL
55e5378d
KW
2869 || strGT(x + COLLXFRM_HDR_LEN,
2870 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2871 {
2872 PL_strxfrm_max_cp = j;
2873 cur_max_x = x;
2874 }
2875 else {
2876 Safefree(x);
2877 }
2878 }
2879
94762aa0
KW
2880 if (! cur_max_x) {
2881 DEBUG_L(PerlIO_printf(Perl_debug_log,
2882 "_mem_collxfrm: Couldn't find any character to"
2883 " replace above-Latin1 chars in locale %s with",
2884 PL_collation_name));
2885 goto bad;
2886 }
2887
58eebef2
KW
2888 DEBUG_L(PerlIO_printf(Perl_debug_log,
2889 "_mem_collxfrm: highest 1-byte collating character"
2890 " in locale %s is 0x%02X\n",
2891 PL_collation_name,
2892 PL_strxfrm_max_cp));
58eebef2 2893
a4a439fb
KW
2894 Safefree(cur_max_x);
2895 }
2896
2897 /* Here we know which legal code point collates the highest.
2898 * We are ready to construct the non-UTF-8 string. The length
2899 * will be at least 1 byte smaller than the input string
2900 * (because we changed at least one 2-byte character into a
2901 * single byte), but that is eaten up by the trailing NUL */
2902 Newx(s, len, char);
2903
2904 {
2905 STRLEN i;
2906 STRLEN d= 0;
042d9e50 2907 char * e = (char *) t + len;
a4a439fb
KW
2908
2909 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2910 U8 cur_char = t[i];
2911 if (UTF8_IS_INVARIANT(cur_char)) {
2912 s[d++] = cur_char;
2913 }
042d9e50 2914 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2915 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2916 }
2917 else { /* Replace illegal cp with highest collating
2918 one */
2919 s[d++] = PL_strxfrm_max_cp;
2920 }
2921 }
2922 s[d++] = '\0';
2923 Renew(s, d, char); /* Free up unused space */
2924 }
2925 }
2926 }
2927
2928 /* Here, we have constructed a modified version of the input. It could
2929 * be that we already had a modified copy before we did this version.
2930 * If so, that copy is no longer needed */
2931 if (t != input_string) {
2932 Safefree(t);
2933 }
2934 }
2935
17f41037
KW
2936 length_in_chars = (utf8)
2937 ? utf8_length((U8 *) s, (U8 *) s + len)
2938 : len;
2939
59c018b9
KW
2940 /* The first element in the output is the collation id, used by
2941 * sv_collxfrm(); then comes the space for the transformed string. The
2942 * equation should give us a good estimate as to how much is needed */
55e5378d 2943 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2944 + PL_collxfrm_base
17f41037 2945 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2946 Newx(xbuf, xAlloc, char);
c7202dee
KW
2947 if (UNLIKELY(! xbuf)) {
2948 DEBUG_L(PerlIO_printf(Perl_debug_log,
2949 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2950 goto bad;
c7202dee 2951 }
98994639 2952
d35fca5f 2953 /* Store the collation id */
98994639 2954 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2955
2956 /* Then the transformation of the input. We loop until successful, or we
2957 * give up */
4ebeff16 2958 for (;;) {
1adab0a7 2959
55e5378d 2960 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2961
2962 /* If the transformed string occupies less space than we told strxfrm()
2963 * was available, it means it successfully transformed the whole
2964 * string. */
55e5378d 2965 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2966
1adab0a7
KW
2967 /* Some systems include a trailing NUL in the returned length.
2968 * Ignore it, using a loop in case multiple trailing NULs are
2969 * returned. */
2970 while ( (*xlen) > 0
2971 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2972 {
2973 (*xlen)--;
2974 }
2975
17f41037
KW
2976 /* If the first try didn't get it, it means our prediction was low.
2977 * Modify the coefficients so that we predict a larger value in any
2978 * future transformations */
2979 if (! first_time) {
2980 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2981 STRLEN computed_guess = PL_collxfrm_base
2982 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2983
2984 /* On zero-length input, just keep current slope instead of
2985 * dividing by 0 */
2986 const STRLEN new_m = (length_in_chars != 0)
2987 ? needed / length_in_chars
2988 : PL_collxfrm_mult;
17f41037
KW
2989
2990 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2991 "%s: %d: initial size of %zu bytes for a length "
2992 "%zu string was insufficient, %zu needed\n",
17f41037 2993 __FILE__, __LINE__,
b07929e4 2994 computed_guess, length_in_chars, needed));
17f41037
KW
2995
2996 /* If slope increased, use it, but discard this result for
2997 * length 1 strings, as we can't be sure that it's a real slope
2998 * change */
2999 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
3000
3001# ifdef DEBUGGING
3002
17f41037
KW
3003 STRLEN old_m = PL_collxfrm_mult;
3004 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
3005
3006# endif
3007
17f41037
KW
3008 PL_collxfrm_mult = new_m;
3009 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3010 computed_guess = PL_collxfrm_base
3011 + (PL_collxfrm_mult * length_in_chars);
3012 if (computed_guess < needed) {
3013 PL_collxfrm_base += needed - computed_guess;
3014 }
3015
3016 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
3017 "%s: %d: slope is now %zu; was %zu, base "
3018 "is now %zu; was %zu\n",
17f41037 3019 __FILE__, __LINE__,
b07929e4
KW
3020 PL_collxfrm_mult, old_m,
3021 PL_collxfrm_base, old_b));
17f41037
KW
3022 }
3023 else { /* Slope didn't change, but 'b' did */
3024 const STRLEN new_b = needed
3025 - computed_guess
3026 + PL_collxfrm_base;
3027 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 3028 "%s: %d: base is now %zu; was %zu\n",
17f41037 3029 __FILE__, __LINE__,
b07929e4 3030 new_b, PL_collxfrm_base));
17f41037
KW
3031 PL_collxfrm_base = new_b;
3032 }
3033 }
3034
4ebeff16
KW
3035 break;
3036 }
bb0f664e 3037
c7202dee
KW
3038 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3039 DEBUG_L(PerlIO_printf(Perl_debug_log,
3040 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3041 *xlen, PERL_INT_MAX));
4ebeff16 3042 goto bad;
c7202dee 3043 }
d35fca5f 3044
c664130f 3045 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
3046 * (usually not including the trailing NUL) when it fails due to not
3047 * enough space being provided. Assume that this is the case unless
3048 * it's been proven otherwise */
c664130f 3049 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 3050 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
3051 }
3052 else { /* Here, either:
3053 * 1) The strxfrm() has previously shown bad behavior; or
3054 * 2) It isn't the first time through the loop, which means
3055 * that the strxfrm() is now showing bad behavior, because
3056 * we gave it what it said was needed in the previous
3057 * iteration, and it came back saying it needed still more.
3058 * (Many versions of cygwin fit this. When the buffer size
3059 * isn't sufficient, they return the input size instead of
3060 * how much is needed.)
d4ff9586
KW
3061 * Increase the buffer size by a fixed percentage and try again.
3062 * */
6ddd902c 3063 xAlloc += (xAlloc / 4) + 1;
c664130f 3064 PL_strxfrm_is_behaved = FALSE;
c664130f 3065
7d4bcc4a
KW
3066# ifdef DEBUGGING
3067
58eebef2
KW
3068 if (DEBUG_Lv_TEST || debug_initialization) {
3069 PerlIO_printf(Perl_debug_log,
3070 "_mem_collxfrm required more space than previously calculated"
b07929e4 3071 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 3072 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 3073 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 3074 }
7d4bcc4a
KW
3075
3076# endif
3077
58eebef2 3078 }
c664130f 3079
4ebeff16 3080 Renew(xbuf, xAlloc, char);
c7202dee
KW
3081 if (UNLIKELY(! xbuf)) {
3082 DEBUG_L(PerlIO_printf(Perl_debug_log,
3083 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 3084 goto bad;
c7202dee 3085 }
c664130f
KW
3086
3087 first_time = FALSE;
4ebeff16 3088 }
98994639 3089
6696cfa7 3090
7d4bcc4a
KW
3091# ifdef DEBUGGING
3092
58eebef2 3093 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
3094
3095 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
3096 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
3097 PerlIO_printf(Perl_debug_log, "%s\n",
3098 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
3099 *xlen, 1));
58eebef2 3100 }
7d4bcc4a
KW
3101
3102# endif
58eebef2 3103
3c5f993e 3104 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 3105 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 3106
6696cfa7
KW
3107 if (s != input_string) {
3108 Safefree(s);
98994639
HS
3109 }
3110
98994639
HS
3111 return xbuf;
3112
3113 bad:
3114 Safefree(xbuf);
6696cfa7
KW
3115 if (s != input_string) {
3116 Safefree(s);
3117 }
98994639 3118 *xlen = 0;
7d4bcc4a
KW
3119
3120# ifdef DEBUGGING
3121
58eebef2 3122 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 3123 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 3124 }
7d4bcc4a
KW
3125
3126# endif
3127
98994639
HS
3128 return NULL;
3129}
3130
7d4bcc4a 3131# ifdef DEBUGGING
c7202dee 3132
4cbaac56 3133STATIC void
c7202dee
KW
3134S_print_collxfrm_input_and_return(pTHX_
3135 const char * const s,
3136 const char * const e,
3137 const STRLEN * const xlen,
3138 const bool is_utf8)
3139{
c7202dee
KW
3140
3141 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
3142
511e4ff7
DM
3143 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
3144 (UV)PL_collation_ix);
c7202dee 3145 if (xlen) {
08b6dc1d 3146 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
3147 }
3148 else {
3149 PerlIO_printf(Perl_debug_log, "NULL");
3150 }
3151 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
3152 PL_collation_name);
9c8a6dc2
KW
3153 print_bytes_for_locale(s, e, is_utf8);
3154
3155 PerlIO_printf(Perl_debug_log, "'\n");
3156}
3157
3158STATIC void
3159S_print_bytes_for_locale(pTHX_
3160 const char * const s,
3161 const char * const e,
3162 const bool is_utf8)
3163{
3164 const char * t = s;
3165 bool prev_was_printable = TRUE;
3166 bool first_time = TRUE;
3167
3168 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
3169
3170 while (t < e) {
3171 UV cp = (is_utf8)
3172 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
3173 : * (U8 *) t;
3174 if (isPRINT(cp)) {
3175 if (! prev_was_printable) {
3176 PerlIO_printf(Perl_debug_log, " ");
3177 }
3178 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
3179 prev_was_printable = TRUE;
3180 }
3181 else {
3182 if (! first_time) {
3183 PerlIO_printf(Perl_debug_log, " ");
3184 }
147e3846 3185 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
3186 prev_was_printable = FALSE;
3187 }
3188 t += (is_utf8) ? UTF8SKIP(t) : 1;
3189 first_time = FALSE;
3190 }
c7202dee
KW
3191}
3192
7d4bcc4a 3193# endif /* #ifdef DEBUGGING */
98994639 3194#endif /* USE_LOCALE_COLLATE */
58eebef2 3195
8ef6e574
KW
3196#ifdef USE_LOCALE
3197
962aa53f
KW
3198STATIC const char *
3199S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
3200{
3201 /* Changes the locale for LC_'switch_category" to that of
3202 * LC_'template_category', if they aren't already the same. If not NULL,
3203 * 'template_locale' is the locale that 'template_category' is in.
3204 *
9de6fe47
KW
3205 * Returns a copy of the name of the original locale for 'switch_category'
3206 * so can be switched back to with the companion function
3207 * restore_switched_locale(), (NULL if no restoral is necessary.) */
962aa53f
KW
3208
3209 char * restore_to_locale = NULL;
3210
3211 if (switch_category == template_category) { /* No changes needed */
3212 return NULL;
3213 }
3214
3215 /* Find the original locale of the category we may need to change, so that
3216 * it can be restored to later */
d6189047
KW
3217 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
3218 NULL)));
962aa53f
KW
3219 if (! restore_to_locale) {
3220 Perl_croak(aTHX_
3221 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3222 __FILE__, __LINE__, category_name(switch_category), errno);
3223 }
962aa53f
KW
3224
3225 /* If the locale of the template category wasn't passed in, find it now */
3226 if (template_locale == NULL) {
3227 template_locale = do_setlocale_r(template_category, NULL);
3228 if (! template_locale) {
3229 Perl_croak(aTHX_
3230 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3231 __FILE__, __LINE__, category_name(template_category), errno);
3232 }
3233 }
3234
3235 /* It the locales are the same, there's nothing to do */
3236 if (strEQ(restore_to_locale, template_locale)) {
3237 Safefree(restore_to_locale);
3238
3239 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
3240 category_name(switch_category), restore_to_locale));
3241
3242 return NULL;
3243 }
3244
3245 /* Finally, change the locale to the template one */
3246 if (! do_setlocale_r(switch_category, template_locale)) {
3247 Perl_croak(aTHX_
3248 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
3249 __FILE__, __LINE__, category_name(switch_category),
3250 template_locale, errno);
3251 }
3252
3253 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
3254 category_name(switch_category), template_locale));
3255
3256 return restore_to_locale;
3257}
3258
3259STATIC void
3260S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
3261{
9de6fe47
KW
3262 /* Restores the locale for LC_'category' to 'original_locale' (which is a
3263 * copy that will be freed by this function), or do nothing if the latter
3264 * parameter is NULL */
962aa53f
KW
3265
3266 if (original_locale == NULL) {
3267 return;
3268 }
3269
3270 if (! do_setlocale_r(category, original_locale)) {
3271 Perl_croak(aTHX_
3272 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
3273 __FILE__, __LINE__,
3274 category_name(category), original_locale, errno);
3275 }
3276
3277 Safefree(original_locale);
3278}
3279
c1284011
KW
3280bool
3281Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
3282{
3283 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
3284 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 3285 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
3286 * could give the wrong result. The result will very likely be correct for
3287 * languages that have commonly used non-ASCII characters, but for notably
3288 * English, it comes down to if the locale's name ends in something like
9de6fe47
KW
3289 * "UTF-8". It errs on the side of not being a UTF-8 locale.
3290 *
3291 * If the platform is early C89, not containing mbtowc(), or we are
3292 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
3293 * These work very well for non-Latin locales or those whose currency
3294 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
3295 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
3296 * on the name containing UTF-8 or not. */
7d74bb61 3297
47280b20 3298 /* Name of current locale corresponding to the input category */
8de4332b 3299 const char *save_input_locale = NULL;
47280b20
KW
3300
3301 bool is_utf8 = FALSE; /* The return value */
7d74bb61 3302
47280b20
KW
3303 /* The variables below are for the cache of previous lookups using this
3304 * function. The cache is a C string, described at the definition for
3305 * 'C_and_POSIX_utf8ness'.
3306 *
3307 * The first part of the cache is fixed, for the C and POSIX locales. The
3308 * varying part starts just after them. */
3309 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
3310
3311 Size_t utf8ness_cache_size; /* Size of the varying portion */
3312 Size_t input_name_len; /* Length in bytes of save_input_locale */
3313 Size_t input_name_len_with_overhead; /* plus extra chars used to store
3314 the name in the cache */
3315 char * delimited; /* The name plus the delimiters used to store
3316 it in the cache */
3317 char * name_pos; /* position of 'delimited' in the cache, or 0
3318 if not there */
3319
3320
7d4bcc4a
KW
3321# ifdef LC_ALL
3322
7d74bb61 3323 assert(category != LC_ALL);
7d4bcc4a
KW
3324
3325# endif
7d74bb61 3326
47280b20 3327 /* Get the desired category's locale */
d6189047 3328 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
7d74bb61 3329 if (! save_input_locale) {
47280b20 3330 Perl_croak(aTHX_
d707d779
KW
3331 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3332 __FILE__, __LINE__, category_name(category), errno);
7d74bb61 3333 }
47280b20 3334
47280b20
KW
3335 DEBUG_L(PerlIO_printf(Perl_debug_log,
3336 "Current locale for %s is %s\n",
3337 category_name(category), save_input_locale));
3338
3339 input_name_len = strlen(save_input_locale);
3340
3341 /* In our cache, each name is accompanied by two delimiters and a single
3342 * utf8ness digit */
3343 input_name_len_with_overhead = input_name_len + 3;
3344
3345 /* Allocate and populate space for a copy of the name surrounded by the
3346 * delimiters */
3347 Newx(delimited, input_name_len_with_overhead, char);
3348 delimited[0] = UTF8NESS_SEP[0];
3349 Copy(save_input_locale, delimited + 1, input_name_len, char);
3350 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
3351 delimited[input_name_len+2] = '\0';
3352
3353 /* And see if that is in the cache */
3354 name_pos = instr(PL_locale_utf8ness, delimited);
3355 if (name_pos) {
3356 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
3357
3358# ifdef DEBUGGING
3359
3360 if (DEBUG_Lv_TEST || debug_initialization) {
3361 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
3362 save_input_locale, is_utf8);
3363 }
3364
3365# endif
3366
3367 /* And, if not already in that position, move it to the beginning of
3368 * the non-constant portion of the list, since it is the most recently
3369 * used. (We don't have to worry about overflow, since just moving
3370 * existing names around) */
3371 if (name_pos > utf8ness_cache) {
3372 Move(utf8ness_cache,
3373 utf8ness_cache + input_name_len_with_overhead,
3374 name_pos - utf8ness_cache, char);
3375 Copy(delimited,
3376 utf8ness_cache,
3377 input_name_len_with_overhead - 1, char);
3378 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3379 }
3380
3381 Safefree(delimited);
b07fffd1 3382 Safefree(save_input_locale);
47280b20 3383 return is_utf8;
7d74bb61
KW
3384 }
3385
47280b20
KW
3386 /* Here we don't have stored the utf8ness for the input locale. We have to
3387 * calculate it */
3388
94646a69
KW
3389# if defined(USE_LOCALE_CTYPE) \
3390 && ( (defined(HAS_NL_LANGINFO) && defined(CODESET)) \
3391 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
7d74bb61 3392
0dec74cd 3393 {
962aa53f
KW
3394 const char *original_ctype_locale
3395 = switch_category_locale_to_template(LC_CTYPE,
3396 category,
3397 save_input_locale);
69014004 3398
7d74bb61 3399 /* Here the current LC_CTYPE is set to the locale of the category whose
0dec74cd 3400 * information is desired. This means that nl_langinfo() and mbtowc()
1d958db2 3401 * should give the correct results */
119ee68b 3402
0dec74cd
KW
3403# ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
3404 calling the functions if we have this */
3405
3406 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3407 * Unicode code point. */
3408
de3a3072
KW
3409 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
3410 __FILE__, __LINE__, (int) MB_CUR_MAX));
0dec74cd
KW
3411 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
3412 is_utf8 = FALSE;
b5b2847c
KW
3413 restore_switched_locale(LC_CTYPE, original_ctype_locale);
3414 goto finish_and_return;
0dec74cd
KW
3415 }
3416
3417# endif
7d4bcc4a
KW
3418# if defined(HAS_NL_LANGINFO) && defined(CODESET)
3419
0dec74cd
KW
3420 { /* The task is easiest if the platform has this POSIX 2001 function.
3421 Except on some platforms it can wrongly return "", so have to have
3422 a fallback. And it can return that it's UTF-8, even if there are
3423 variances from that. For example, Turkish locales may use the
3424 alternate dotted I rules, and sometimes it appears to be a
3425 defective locale definition. XXX We should probably check for
9de6fe47
KW
3426 these in the Latin1 range and warn (but on glibc, requires
3427 iswalnum() etc. due to their not handling 80-FF correctly */
c70a3e68
KW
3428 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
3429 /* FALSE => already in dest locale */
119ee68b 3430
af84e886 3431 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
c70a3e68
KW
3432 "\tnllanginfo returned CODESET '%s'\n", codeset));
3433
3434 if (codeset && strNE(codeset, "")) {
1d958db2 3435
1d075173
KW
3436 /* If the implementation of foldEQ() somehow were
3437 * to change to not go byte-by-byte, this could
3438 * read past end of string, as only one length is
3439 * checked. But currently, a premature NUL will
3440 * compare false, and it will stop there */
3441 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
3442 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
1d958db2 3443
69014004
KW
3444 DEBUG_L(PerlIO_printf(Perl_debug_log,
3445 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
3446 codeset, is_utf8));
b5b2847c
KW
3447 restore_switched_locale(LC_CTYPE, original_ctype_locale);
3448 goto finish_and_return;
1d958db2 3449 }
119ee68b
KW
3450 }
3451
7d4bcc4a 3452# endif
94646a69 3453# if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
0dec74cd
KW
3454 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
3455 * late adder to C89, so very likely to have it. However, testing has
3456 * shown that, like nl_langinfo() above, there are locales that are not
3457 * strictly UTF-8 that this will return that they are */
69014004 3458
0dec74cd 3459 {
119ee68b 3460 wchar_t wc;
51fc4b19 3461 int len;
48015184 3462 dSAVEDERRNO;
51fc4b19 3463
94646a69
KW
3464# if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
3465
3466 mbstate_t ps;
3467
3468# endif
3469
3470 /* mbrtowc() and mbtowc() convert a byte string to a wide
3471 * character. Feed a byte string to one of them and check that the
3472 * result is the expected Unicode code point */
3473
3474# if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
3475 /* Prefer this function if available, as it's reentrant */
3476
3477 memset(&ps, 0, sizeof(ps));;
3478 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
3479 state */
48015184 3480 SETERRNO(0, 0);
94646a69 3481 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
48015184 3482 SAVE_ERRNO;
94646a69
KW
3483
3484# else
0dec74cd 3485
5a91a58b 3486 LOCALE_LOCK;
856b881c 3487 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
48015184 3488 SETERRNO(0, 0);
b1d4925c 3489 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
48015184 3490 SAVE_ERRNO;
5a91a58b 3491 LOCALE_UNLOCK;
51fc4b19 3492
94646a69
KW
3493# endif
3494
48015184 3495 RESTORE_ERRNO;
af84e886 3496 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
0dec74cd 3497 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
48015184 3498 len, (unsigned int) wc, GET_ERRNO));
b1d4925c 3499
0dec74cd
KW
3500 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
3501 && wc == (wchar_t) UNICODE_REPLACEMENT);
119ee68b 3502 }
7d4bcc4a 3503
962aa53f 3504 restore_switched_locale(LC_CTYPE, original_ctype_locale);
47280b20 3505 goto finish_and_return;
7d74bb61 3506 }
119ee68b 3507
0dec74cd
KW
3508# endif
3509# else
7d74bb61 3510
0dec74cd
KW
3511 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
3512 * try looking at the currency symbol to see if it disambiguates
3513 * things. Often that will be in the native script, and if the symbol
3514 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
3515 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
3516 * string being valid UTF-8 are quite small */
fa9b773e 3517
660ee27b
KW
3518# ifdef USE_LOCALE_MONETARY
3519
3520 /* If have LC_MONETARY, we can look at the currency symbol. Often that
3521 * will be in the native script. We do this one first because there is
3522 * just one string to examine, so potentially avoids work */
7d4bcc4a 3523
9db16864
KW
3524 {
3525 const char *original_monetary_locale
660ee27b
KW
3526 = switch_category_locale_to_template(LC_MONETARY,
3527 category,
3528 save_input_locale);
9db16864 3529 bool only_ascii = FALSE;
660ee27b
KW
3530 const U8 * currency_string
3531 = (const U8 *) my_nl_langinfo(PERL_CRNCYSTR, FALSE);
3532 /* 2nd param not relevant for this item */
3533 const U8 * first_variant;
fa9b773e 3534
660ee27b
KW
3535 assert( *currency_string == '-'
3536 || *currency_string == '+'
3537 || *currency_string == '.');
97f4de96 3538
660ee27b 3539 currency_string++;
fa9b773e 3540
660ee27b 3541 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
9db16864
KW
3542 {
3543 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));
3544 only_ascii = TRUE;
3545 }
3546 else {
660ee27b 3547 is_utf8 = is_strict_utf8_string(first_variant, 0);
9db16864 3548 }
fa9b773e 3549
9db16864 3550 restore_switched_locale(LC_MONETARY, original_monetary_locale);
fa9b773e 3551
9db16864 3552 if (! only_ascii) {
fa9b773e 3553
9db16864
KW
3554 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3555 * otherwise assume the locale is UTF-8 if and only if the symbol
3556 * is non-ascii UTF-8. */
af84e886 3557 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
9db16864
KW
3558 save_input_locale, is_utf8));
3559 goto finish_and_return;
3560 }
13542a67 3561 }
fa9b773e 3562
660ee27b 3563# endif /* USE_LOCALE_MONETARY */
7d4bcc4a 3564# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e 3565
9db16864
KW
3566 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3567 * the names of the months and weekdays, timezone, and am/pm indicator */
3568 {
3569 const char *original_time_locale
3570 = switch_category_locale_to_template(LC_TIME,
3571 category,
3572 save_input_locale);
3573 int hour = 10;
3574 bool is_dst = FALSE;
3575 int dom = 1;
3576 int month = 0;
3577 int i;
3578 char * formatted_time;
3579
3580 /* Here the current LC_TIME is set to the locale of the category
3581 * whose information is desired. Look at all the days of the week and
3582 * month names, and the timezone and am/pm indicator for UTF-8 variant
3583 * characters. The first such a one found will tell us if the locale
3584 * is UTF-8 or not */
3585
3586 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3587 formatted_time = my_strftime("%A %B %Z %p",
3588 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
3589 if ( ! formatted_time
3590 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3591 {
15f7e74e 3592
9db16864
KW
3593 /* Here, we didn't find a non-ASCII. Try the next time through
3594 * with the complemented dst and am/pm, and try with the next
3595 * weekday. After we have gotten all weekdays, try the next
3596 * month */
3597 is_dst = ! is_dst;
3598 hour = (hour + 12) % 24;
3599 dom++;
3600 if (i > 6) {
3601 month++;
3602 }
3603 continue;
15f7e74e 3604 }
9db16864
KW
3605
3606 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3607 * false otherwise. But first, restore LC_TIME to its original
3608 * locale if we changed it */
3609 restore_switched_locale(LC_TIME, original_time_locale);
3610
af84e886 3611 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
9db16864
KW
3612 save_input_locale,
3613 is_utf8_string((U8 *) formatted_time, 0)));
3614 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
3615 goto finish_and_return;
15f7e74e
KW
3616 }
3617
9db16864
KW
3618 /* Falling off the end of the loop indicates all the names were just
3619 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3620 * to its original locale */
962aa53f 3621 restore_switched_locale(LC_TIME, original_time_locale);
af84e886 3622 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
3623 }
3624
7d4bcc4a 3625# endif
15f7e74e 3626
7d4bcc4a 3627# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93 3628
9db16864
KW
3629 /* This code is ifdefd out because it was found to not be necessary in testing
3630 * on our dromedary test machine, which has over 700 locales. There, this
3631 * added no value to looking at the currency symbol and the time strings. I
3632 * left it in so as to avoid rewriting it if real-world experience indicates
3633 * that dromedary is an outlier. Essentially, instead of returning abpve if we
3634 * haven't found illegal utf8, we continue on and examine all the strerror()
3635 * messages on the platform for utf8ness. If all are ASCII, we still don't
3636 * know the answer; but otherwise we have a pretty good indication of the
3637 * utf8ness. The reason this doesn't help much is that the messages may not
3638 * have been translated into the locale. The currency symbol and time strings
3639 * are much more likely to have been translated. */
3640 {
3641 int e;
3642 bool non_ascii = FALSE;
3643 const char *original_messages_locale
3644 = switch_category_locale_to_template(LC_MESSAGES,
3645 category,
3646 save_input_locale);
3647 const char * errmsg = NULL;
3648
3649 /* Here the current LC_MESSAGES is set to the locale of the category
3650 * whose information is desired. Look through all the messages. We
3651 * can't use Strerror() here because it may expand to code that
3652 * segfaults in miniperl */
3653
3654 for (e = 0; e <= sys_nerr; e++) {
3655 errno = 0;
3656 errmsg = sys_errlist[e];
3657 if (errno || !errmsg) {
3658 break;
3659 }
3660 errmsg = savepv(errmsg);
3661 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
3662 non_ascii = TRUE;
3663 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3664 break;
3665 }
855aeb93 3666 }
9db16864 3667 Safefree(errmsg);
855aeb93 3668
9db16864 3669 restore_switched_locale(LC_MESSAGES, original_messages_locale);
855aeb93 3670
9db16864 3671 if (non_ascii) {
5857e934 3672
9db16864
KW
3673 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3674 * any non-ascii means it is one; otherwise we assume it isn't */
af84e886 3675 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
9db16864
KW
3676 save_input_locale,
3677 is_utf8));
3678 goto finish_and_return;
3679 }
855aeb93 3680
9db16864
KW
3681 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));
3682 }
855aeb93 3683
7d4bcc4a 3684# endif
0dec74cd 3685# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3686 UTF-8 locale */
7d4bcc4a 3687
97f4de96
KW
3688 /* As a last resort, look at the locale name to see if it matches
3689 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3690 * return of setlocale(), is actually defined to be opaque, so we can't
3691 * really rely on the absence of various substrings in the name to indicate
3692 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3693 * be a UTF-8 locale. Similarly for the other common names */
3694
0dec74cd 3695 {
c36d8df8 3696 const Size_t final_pos = strlen(save_input_locale) - 1;
0dec74cd 3697
c36d8df8
KW
3698 if (final_pos >= 3) {
3699 const char *name = save_input_locale;
97f4de96 3700
c36d8df8
KW
3701 /* Find next 'U' or 'u' and look from there */
3702 while ((name += strcspn(name, "Uu") + 1)
3703 <= save_input_locale + final_pos - 2)
97f4de96 3704 {
c36d8df8
KW
3705 if ( isALPHA_FOLD_NE(*name, 't')
3706 || isALPHA_FOLD_NE(*(name + 1), 'f'))
3707 {
3708 continue;
3709 }
3710 name += 2;
3711 if (*(name) == '-') {
3712 if ((name > save_input_locale + final_pos - 1)) {
3713 break;
3714 }
3715 name++;
3716 }
3717 if (*(name) == '8') {
3718 DEBUG_L(PerlIO_printf(Perl_debug_log,
3719 "Locale %s ends with UTF-8 in name\n",
3720 save_input_locale));
3721 is_utf8 = TRUE;
3722 goto finish_and_return;
97f4de96 3723 }
97f4de96 3724 }
c36d8df8
KW
3725 DEBUG_L(PerlIO_printf(Perl_debug_log,
3726 "Locale %s doesn't end with UTF-8 in name\n",
3727 save_input_locale));
97f4de96 3728 }
97f4de96 3729
4d8d465a 3730# ifdef WIN32
7d4bcc4a 3731
c36d8df8
KW
3732 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
3733 if (memENDs(save_input_locale, final_pos, "65001")) {
3734 DEBUG_L(PerlIO_printf(Perl_debug_log,
8a0832a1 3735 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
97f4de96 3736 save_input_locale));
c36d8df8
KW
3737 is_utf8 = TRUE;
3738 goto finish_and_return;
3739 }
0dec74cd 3740 }
7d4bcc4a 3741