This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Add, move, clarify comments
[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
KW
25 *
26 * All C programs have an underlying locale. Perl generally doesn't pay any
27 * attention to it except within the scope of a 'use locale'. For most
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
39#include "perl.h"
40
b310b053
JH
41#ifdef I_LANGINFO
42# include <langinfo.h>
43#endif
44
a4af207c
JH
45#include "reentr.h"
46
8ef6e574
KW
47#ifdef USE_LOCALE
48
98994639 49/*
0d071d52
KW
50 * Standardize the locale name from a string returned by 'setlocale', possibly
51 * modifying that string.
98994639 52 *
0ef2a2b2 53 * The typical return value of setlocale() is either
98994639
HS
54 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
55 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
56 * (the space-separated values represent the various sublocales,
0ef2a2b2 57 * in some unspecified order). This is not handled by this function.
98994639
HS
58 *
59 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
0ef2a2b2
KW
60 * which is harmful for further use of the string in setlocale(). This
61 * function removes the trailing new line and everything up through the '='
98994639
HS
62 *
63 */
64STATIC char *
65S_stdize_locale(pTHX_ char *locs)
66{
7452cf6a 67 const char * const s = strchr(locs, '=');
98994639
HS
68 bool okay = TRUE;
69
7918f24d
NC
70 PERL_ARGS_ASSERT_STDIZE_LOCALE;
71
8772537c
AL
72 if (s) {
73 const char * const t = strchr(s, '.');
98994639 74 okay = FALSE;
8772537c
AL
75 if (t) {
76 const char * const u = strchr(t, '\n');
77 if (u && (u[1] == 0)) {
78 const STRLEN len = u - s;
79 Move(s + 1, locs, len, char);
80 locs[len] = 0;
81 okay = TRUE;
98994639
HS
82 }
83 }
84 }
85
86 if (!okay)
87 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
88
89 return locs;
90}
91
8ef6e574
KW
92#endif
93
98994639
HS
94void
95Perl_set_numeric_radix(pTHX)
96{
97#ifdef USE_LOCALE_NUMERIC
98# ifdef HAS_LOCALECONV
7452cf6a 99 const struct lconv* const lc = localeconv();
98994639 100
98994639
HS
101 if (lc && lc->decimal_point) {
102 if (lc->decimal_point[0] == '.' && lc->decimal_point[1] == 0) {
103 SvREFCNT_dec(PL_numeric_radix_sv);
a0714e2c 104 PL_numeric_radix_sv = NULL;
98994639
HS
105 }
106 else {
107 if (PL_numeric_radix_sv)
108 sv_setpv(PL_numeric_radix_sv, lc->decimal_point);
109 else
110 PL_numeric_radix_sv = newSVpv(lc->decimal_point, 0);
9f10db87 111 if (! is_invariant_string((U8 *) lc->decimal_point, 0)
28acfe03 112 && is_utf8_string((U8 *) lc->decimal_point, 0)
c1284011 113 && _is_cur_LC_category_utf8(LC_NUMERIC))
28acfe03
KW
114 {
115 SvUTF8_on(PL_numeric_radix_sv);
116 }
98994639
HS
117 }
118 }
119 else
a0714e2c 120 PL_numeric_radix_sv = NULL;
69014004 121
ba99daed 122 DEBUG_L(PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
69014004 123 (PL_numeric_radix_sv)
37b7e435
KW
124 ? SvPVX(PL_numeric_radix_sv)
125 : "NULL",
126 (PL_numeric_radix_sv)
39eb7305 127 ? cBOOL(SvUTF8(PL_numeric_radix_sv))
37b7e435 128 : 0));
69014004 129
98994639
HS
130# endif /* HAS_LOCALECONV */
131#endif /* USE_LOCALE_NUMERIC */
132}
133
a39edc4c
KW
134/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
135 * return of setlocale(), then this is extremely likely to be the C or POSIX
136 * locale. However, the output of setlocale() is documented to be opaque, but
137 * the odds are extremely small that it would return these two strings for some
138 * other locale. Note that VMS in these two locales includes many non-ASCII
139 * characters as controls and punctuation (below are hex bytes):
140 * cntrl: 00-1F 7F 84-97 9B-9F
141 * punct: 21-2F 3A-40 5B-60 7B-7E A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
142 * Oddly, none there are listed as alphas, though some represent alphabetics
143 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
98b630b3
KW
144#define isNAME_C_OR_POSIX(name) ((name) != NULL \
145 && ((*(name) == 'C' && (*(name + 1)) == '\0') \
a39edc4c
KW
146 || strEQ((name), "POSIX")))
147
98994639 148void
8772537c 149Perl_new_numeric(pTHX_ const char *newnum)
98994639
HS
150{
151#ifdef USE_LOCALE_NUMERIC
0d071d52
KW
152
153 /* Called after all libc setlocale() calls affecting LC_NUMERIC, to tell
154 * core Perl this and that 'newnum' is the name of the new locale.
155 * It installs this locale as the current underlying default.
156 *
157 * The default locale and the C locale can be toggled between by use of the
158 * set_numeric_local() and set_numeric_standard() functions, which should
159 * probably not be called directly, but only via macros like
160 * SET_NUMERIC_STANDARD() in perl.h.
161 *
162 * The toggling is necessary mainly so that a non-dot radix decimal point
163 * character can be output, while allowing internal calculations to use a
164 * dot.
165 *
166 * This sets several interpreter-level variables:
bb304765 167 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
0d071d52 168 * PL_numeric_local A boolean indicating if the toggled state is such
7738054c
KW
169 * that the current locale is the program's underlying
170 * locale
171 * PL_numeric_standard An int indicating if the toggled state is such
172 * that the current locale is the C locale. If non-zero,
173 * it is in C; if > 1, it means it may not be toggled away
174 * from C.
0d071d52
KW
175 * Note that both of the last two variables can be true at the same time,
176 * if the underlying locale is C. (Toggling is a no-op under these
177 * circumstances.)
178 *
179 * Any code changing the locale (outside this file) should use
180 * POSIX::setlocale, which calls this function. Therefore this function
181 * should be called directly only from this file and from
182 * POSIX::setlocale() */
183
b03f34cf 184 char *save_newnum;
98994639
HS
185
186 if (! newnum) {
43c5f42d
NC
187 Safefree(PL_numeric_name);
188 PL_numeric_name = NULL;
98994639
HS
189 PL_numeric_standard = TRUE;
190 PL_numeric_local = TRUE;
191 return;
192 }
193
b03f34cf 194 save_newnum = stdize_locale(savepv(newnum));
abe1abcf
KW
195
196 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
197 PL_numeric_local = TRUE;
198
b03f34cf 199 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 200 Safefree(PL_numeric_name);
b03f34cf 201 PL_numeric_name = save_newnum;
b03f34cf 202 }
abe1abcf
KW
203 else {
204 Safefree(save_newnum);
205 }
4c28b29c
KW
206
207 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
208 * have to worry about the radix being a non-dot. (Core operations that
209 * need the underlying locale change to it temporarily). */
210 set_numeric_standard();
211
e19f01cb 212 set_numeric_radix();
6959d69d 213
f2ce9e1c
JH
214#else
215 PERL_UNUSED_ARG(newnum);
98994639
HS
216#endif /* USE_LOCALE_NUMERIC */
217}
218
219void
220Perl_set_numeric_standard(pTHX)
221{
222#ifdef USE_LOCALE_NUMERIC
28c1bf33
KW
223 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
224 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
225 * macro avoids calling this routine if toggling isn't necessary according
226 * to our records (which could be wrong if some XS code has changed the
227 * locale behind our back) */
0d071d52 228
a9b8c0d8
KW
229 setlocale(LC_NUMERIC, "C");
230 PL_numeric_standard = TRUE;
231 PL_numeric_local = isNAME_C_OR_POSIX(PL_numeric_name);
232 set_numeric_radix();
69014004
KW
233 DEBUG_L(PerlIO_printf(Perl_debug_log,
234 "Underlying LC_NUMERIC locale now is C\n"));
98994639
HS
235
236#endif /* USE_LOCALE_NUMERIC */
237}
238
239void
240Perl_set_numeric_local(pTHX)
241{
242#ifdef USE_LOCALE_NUMERIC
28c1bf33
KW
243 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
244 * code should use the macros like SET_NUMERIC_LOCAL() in perl.h instead of
245 * calling this directly. The macro avoids calling this routine if
246 * toggling isn't necessary according to our records (which could be wrong
247 * if some XS code has changed the locale behind our back) */
a9b8c0d8
KW
248
249 setlocale(LC_NUMERIC, PL_numeric_name);
250 PL_numeric_standard = isNAME_C_OR_POSIX(PL_numeric_name);
251 PL_numeric_local = TRUE;
252 set_numeric_radix();
69014004
KW
253 DEBUG_L(PerlIO_printf(Perl_debug_log,
254 "Underlying LC_NUMERIC locale now is %s\n",
255 PL_numeric_name));
98994639
HS
256
257#endif /* USE_LOCALE_NUMERIC */
258}
259
260/*
261 * Set up for a new ctype locale.
262 */
263void
8772537c 264Perl_new_ctype(pTHX_ const char *newctype)
98994639
HS
265{
266#ifdef USE_LOCALE_CTYPE
0d071d52
KW
267
268 /* Called after all libc setlocale() calls affecting LC_CTYPE, to tell
269 * core Perl this and that 'newctype' is the name of the new locale.
270 *
271 * This function sets up the folding arrays for all 256 bytes, assuming
272 * that tofold() is tolc() since fold case is not a concept in POSIX,
273 *
274 * Any code changing the locale (outside this file) should use
275 * POSIX::setlocale, which calls this function. Therefore this function
276 * should be called directly only from this file and from
277 * POSIX::setlocale() */
278
27da23d5 279 dVAR;
68067e4e 280 UV i;
98994639 281
7918f24d
NC
282 PERL_ARGS_ASSERT_NEW_CTYPE;
283
215c5139
KW
284 /* We will replace any bad locale warning with 1) nothing if the new one is
285 * ok; or 2) a new warning for the bad new locale */
286 if (PL_warn_locale) {
287 SvREFCNT_dec_NN(PL_warn_locale);
288 PL_warn_locale = NULL;
289 }
290
c1284011 291 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
292
293 /* A UTF-8 locale gets standard rules. But note that code still has to
294 * handle this specially because of the three problematic code points */
295 if (PL_in_utf8_CTYPE_locale) {
296 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
297 }
298 else {
8c6180a9
KW
299 /* Assume enough space for every character being bad. 4 spaces each
300 * for the 94 printable characters that are output like "'x' "; and 5
301 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
302 * NUL */
303 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
304
305 bool check_for_problems = ckWARN_d(WARN_LOCALE); /* No warnings means
306 no check */
307 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
308 to start */
309 unsigned int bad_count = 0; /* Count of bad characters */
310
baa60164
KW
311 for (i = 0; i < 256; i++) {
312 if (isUPPER_LC((U8) i))
313 PL_fold_locale[i] = (U8) toLOWER_LC((U8) i);
314 else if (isLOWER_LC((U8) i))
315 PL_fold_locale[i] = (U8) toUPPER_LC((U8) i);
316 else
317 PL_fold_locale[i] = (U8) i;
8c6180a9
KW
318
319 /* If checking for locale problems, see if the native ASCII-range
320 * printables plus \n and \t are in their expected categories in
321 * the new locale. If not, this could mean big trouble, upending
322 * Perl's and most programs' assumptions, like having a
323 * metacharacter with special meaning become a \w. Fortunately,
324 * it's very rare to find locales that aren't supersets of ASCII
325 * nowadays. It isn't a problem for most controls to be changed
326 * into something else; we check only \n and \t, though perhaps \r
327 * could be an issue as well. */
328 if (check_for_problems
329 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
330 {
331 if ((isALPHANUMERIC_A(i) && ! isALPHANUMERIC_LC(i))
332 || (isPUNCT_A(i) && ! isPUNCT_LC(i))
333 || (isBLANK_A(i) && ! isBLANK_LC(i))
334 || (i == '\n' && ! isCNTRL_LC(i)))
335 {
336 if (bad_count) { /* Separate multiple entries with a
337 blank */
338 bad_chars_list[bad_count++] = ' ';
339 }
340 bad_chars_list[bad_count++] = '\'';
341 if (isPRINT_A(i)) {
342 bad_chars_list[bad_count++] = (char) i;
343 }
344 else {
345 bad_chars_list[bad_count++] = '\\';
346 if (i == '\n') {
347 bad_chars_list[bad_count++] = 'n';
348 }
349 else {
350 assert(i == '\t');
351 bad_chars_list[bad_count++] = 't';
352 }
353 }
354 bad_chars_list[bad_count++] = '\'';
355 bad_chars_list[bad_count] = '\0';
356 }
357 }
358 }
359
360#ifdef MB_CUR_MAX
361 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 362 * this locale requires more than one byte, there are going to be
8c6180a9 363 * problems. */
ba1a4362
KW
364 if (check_for_problems && MB_CUR_MAX > 1
365
366 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
367 * locale. Just assume that the implementation for them (plus
368 * for POSIX) is correct and the > 1 value is spurious. (Since
369 * these are specially handled to never be considered UTF-8
370 * locales, as long as this is the only problem, everything
371 * should work fine */
372 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
373 {
8c6180a9
KW
374 multi_byte_locale = TRUE;
375 }
376#endif
377
378 if (bad_count || multi_byte_locale) {
780fcc9f 379 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 380 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 381 newctype,
8c6180a9
KW
382 (multi_byte_locale)
383 ? " Some characters in it are not recognized by"
384 " Perl."
385 : "",
386 (bad_count)
387 ? "\nThe following characters (and maybe others)"
388 " may not have the same meaning as the Perl"
389 " program expects:\n"
390 : "",
391 (bad_count)
392 ? bad_chars_list
393 : ""
394 );
780fcc9f
KW
395 /* If we are actually in the scope of the locale, output the
396 * message now. Otherwise we save it to be output at the first
397 * operation using this locale, if that actually happens. Most
398 * programs don't use locales, so they are immune to bad ones */
399 if (IN_LC(LC_CTYPE)) {
400
401 /* We have to save 'newctype' because the setlocale() just
402 * below may destroy it. The next setlocale() further down
403 * should restore it properly so that the intermediate change
404 * here is transparent to this function's caller */
405 const char * const badlocale = savepv(newctype);
406
407 setlocale(LC_CTYPE, "C");
408
409 /* The '0' below suppresses a bogus gcc compiler warning */
410 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
411 setlocale(LC_CTYPE, badlocale);
c0f3a893 412 Safefree(badlocale);
780fcc9f
KW
413 SvREFCNT_dec_NN(PL_warn_locale);
414 PL_warn_locale = NULL;
415 }
baa60164 416 }
31f05a37 417 }
98994639
HS
418
419#endif /* USE_LOCALE_CTYPE */
7918f24d 420 PERL_ARGS_ASSERT_NEW_CTYPE;
8772537c 421 PERL_UNUSED_ARG(newctype);
96a5add6 422 PERL_UNUSED_CONTEXT;
98994639
HS
423}
424
98994639 425void
2726666d
KW
426Perl__warn_problematic_locale()
427{
2726666d
KW
428
429#ifdef USE_LOCALE_CTYPE
430
5f04a188
KW
431 dTHX;
432
433 /* Internal-to-core function that outputs the message in PL_warn_locale,
434 * and then NULLS it. Should be called only through the macro
435 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
436
2726666d
KW
437 if (PL_warn_locale) {
438 /*GCC_DIAG_IGNORE(-Wformat-security); Didn't work */
439 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
440 SvPVX(PL_warn_locale),
441 0 /* dummy to avoid compiler warning */ );
442 /* GCC_DIAG_RESTORE; */
443 SvREFCNT_dec_NN(PL_warn_locale);
444 PL_warn_locale = NULL;
445 }
446
447#endif
448
449}
450
451void
8772537c 452Perl_new_collate(pTHX_ const char *newcoll)
98994639
HS
453{
454#ifdef USE_LOCALE_COLLATE
0d071d52
KW
455
456 /* Called after all libc setlocale() calls affecting LC_COLLATE, to tell
457 * core Perl this and that 'newcoll' is the name of the new locale.
458 *
459 * Any code changing the locale (outside this file) should use
460 * POSIX::setlocale, which calls this function. Therefore this function
461 * should be called directly only from this file and from
d35fca5f
KW
462 * POSIX::setlocale().
463 *
464 * The design of locale collation is that every locale change is given an
465 * index 'PL_collation_ix'. The first time a string particpates in an
466 * operation that requires collation while locale collation is active, it
467 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
468 * magic includes the collation index, and the transformation of the string
469 * by strxfrm(), q.v. That transformation is used when doing comparisons,
470 * instead of the string itself. If a string changes, the magic is
471 * cleared. The next time the locale changes, the index is incremented,
472 * and so we know during a comparison that the transformation is not
473 * necessarily still valid, and so is recomputed. Note that if the locale
474 * changes enough times, the index could wrap (a U32), and it is possible
475 * that a transformation would improperly be considered valid, leading to
476 * an unlikely bug */
0d071d52 477
98994639
HS
478 if (! newcoll) {
479 if (PL_collation_name) {
480 ++PL_collation_ix;
481 Safefree(PL_collation_name);
482 PL_collation_name = NULL;
483 }
484 PL_collation_standard = TRUE;
00bf60ca 485 is_standard_collation:
98994639
HS
486 PL_collxfrm_base = 0;
487 PL_collxfrm_mult = 2;
165a1c52 488 PL_in_utf8_COLLATE_locale = FALSE;
98994639
HS
489 return;
490 }
491
d35fca5f 492 /* If this is not the same locale as currently, set the new one up */
98994639
HS
493 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
494 ++PL_collation_ix;
495 Safefree(PL_collation_name);
496 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 497 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
498 if (PL_collation_standard) {
499 goto is_standard_collation;
500 }
98994639 501
165a1c52
KW
502 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
503
59c018b9
KW
504 /* A locale collation definition includes primary, secondary, tertiary,
505 * etc. weights for each character. To sort, the primary weights are
506 * used, and only if they compare equal, then the secondary weights are
507 * used, and only if they compare equal, then the tertiary, etc.
508 *
509 * strxfrm() works by taking the input string, say ABC, and creating an
510 * output transformed string consisting of first the primary weights,
511 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
512 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
513 * may not have weights at every level. In our example, let's say B
514 * doesn't have a tertiary weight, and A doesn't have a secondary
515 * weight. The constructed string is then going to be
516 * A¹B¹C¹ B²C² A³C³ ....
517 * This has the desired effect that strcmp() will look at the secondary
518 * or tertiary weights only if the strings compare equal at all higher
519 * priority weights. The spaces shown here, like in
520 * "A¹B¹C¹ * A²B²C² "
521 * are not just for readability. In the general case, these must
522 * actually be bytes, which we will call here 'separator weights'; and
523 * they must be smaller than any other weight value, but since these
524 * are C strings, only the terminating one can be a NUL (some
525 * implementations may include a non-NUL separator weight just before
526 * the NUL). Implementations tend to reserve 01 for the separator
527 * weights. They are needed so that a shorter string's secondary
528 * weights won't be misconstrued as primary weights of a longer string,
529 * etc. By making them smaller than any other weight, the shorter
530 * string will sort first. (Actually, if all secondary weights are
531 * smaller than all primary ones, there is no need for a separator
532 * weight between those two levels, etc.)
533 *
534 * The length of the transformed string is roughly a linear function of
535 * the input string. It's not exactly linear because some characters
536 * don't have weights at all levels. When we call strxfrm() we have to
537 * allocate some memory to hold the transformed string. The
538 * calculations below try to find coefficients 'm' and 'b' for this
539 * locale so that m*x + b equals how much space we need, given the size
540 * of the input string in 'x'. If we calculate too small, we increase
541 * the size as needed, and call strxfrm() again, but it is better to
542 * get it right the first time to avoid wasted expensive string
543 * transformations. */
544
98994639
HS
545 {
546 /* 2: at most so many chars ('a', 'b'). */
547 /* 50: surely no system expands a char more. */
548#define XFRMBUFSIZE (2 * 50)
549 char xbuf[XFRMBUFSIZE];
8772537c
AL
550 const Size_t fa = strxfrm(xbuf, "a", XFRMBUFSIZE);
551 const Size_t fb = strxfrm(xbuf, "ab", XFRMBUFSIZE);
552 const SSize_t mult = fb - fa;
e0601d3c 553 if (mult < 1 && !(fa == 0 && fb == 0))
ad49ad39
NC
554 Perl_croak(aTHX_ "panic: strxfrm() gets absurd - a => %"UVuf", ab => %"UVuf,
555 (UV) fa, (UV) fb);
eb160463 556 PL_collxfrm_base = (fa > (Size_t)mult) ? (fa - mult) : 0;
98994639
HS
557 PL_collxfrm_mult = mult;
558 }
559 }
560
f2ce9e1c
JH
561#else
562 PERL_UNUSED_ARG(newcoll);
98994639
HS
563#endif /* USE_LOCALE_COLLATE */
564}
565
b385bb4d
KW
566#ifdef WIN32
567
568char *
569Perl_my_setlocale(pTHX_ int category, const char* locale)
570{
571 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
572 * difference unless the input locale is "", which means on Windows to get
573 * the machine default, which is set via the computer's "Regional and
574 * Language Options" (or its current equivalent). In POSIX, it instead
575 * means to find the locale from the user's environment. This routine
576 * looks in the environment, and, if anything is found, uses that instead
577 * of going to the machine default. If there is no environment override,
578 * the machine default is used, as normal, by calling the real setlocale()
579 * with "". The POSIX behavior is to use the LC_ALL variable if set;
580 * otherwise to use the particular category's variable if set; otherwise to
581 * use the LANG variable. */
582
175c4cf9 583 bool override_LC_ALL = FALSE;
89f7b9aa
KW
584 char * result;
585
b385bb4d
KW
586 if (locale && strEQ(locale, "")) {
587# ifdef LC_ALL
588 locale = PerlEnv_getenv("LC_ALL");
589 if (! locale) {
590#endif
591 switch (category) {
592# ifdef LC_ALL
593 case LC_ALL:
481465ea 594 override_LC_ALL = TRUE;
b385bb4d
KW
595 break; /* We already know its variable isn't set */
596# endif
597# ifdef USE_LOCALE_TIME
598 case LC_TIME:
599 locale = PerlEnv_getenv("LC_TIME");
600 break;
601# endif
602# ifdef USE_LOCALE_CTYPE
603 case LC_CTYPE:
604 locale = PerlEnv_getenv("LC_CTYPE");
605 break;
606# endif
607# ifdef USE_LOCALE_COLLATE
608 case LC_COLLATE:
609 locale = PerlEnv_getenv("LC_COLLATE");
610 break;
611# endif
612# ifdef USE_LOCALE_MONETARY
613 case LC_MONETARY:
614 locale = PerlEnv_getenv("LC_MONETARY");
615 break;
616# endif
617# ifdef USE_LOCALE_NUMERIC
618 case LC_NUMERIC:
619 locale = PerlEnv_getenv("LC_NUMERIC");
620 break;
621# endif
622# ifdef USE_LOCALE_MESSAGES
623 case LC_MESSAGES:
624 locale = PerlEnv_getenv("LC_MESSAGES");
625 break;
626# endif
627 default:
628 /* This is a category, like PAPER_SIZE that we don't
629 * know about; and so can't provide a wrapper. */
630 break;
631 }
632 if (! locale) {
633 locale = PerlEnv_getenv("LANG");
481465ea 634 if (! locale) {
b385bb4d
KW
635 locale = "";
636 }
637 }
638# ifdef LC_ALL
639 }
640# endif
641 }
642
89f7b9aa 643 result = setlocale(category, locale);
bbc98134
KW
644 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
645 _setlocale_debug_string(category, locale, result)));
89f7b9aa 646
481465ea 647 if (! override_LC_ALL) {
89f7b9aa
KW
648 return result;
649 }
650
dfd77d7a 651 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
652 * LANG variable or the system default if there is no LANG. But these have
653 * lower priority than the other LC_foo variables, so override it for each
654 * one that is set. (If they are set to "", it means to use the same thing
655 * we just set LC_ALL to, so can skip) */
89f7b9aa
KW
656# ifdef USE_LOCALE_TIME
657 result = PerlEnv_getenv("LC_TIME");
730252b2 658 if (result && strNE(result, "")) {
89f7b9aa 659 setlocale(LC_TIME, result);
bbc98134
KW
660 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
661 __FILE__, __LINE__,
662 _setlocale_debug_string(LC_TIME, result, "not captured")));
89f7b9aa
KW
663 }
664# endif
665# ifdef USE_LOCALE_CTYPE
666 result = PerlEnv_getenv("LC_CTYPE");
730252b2 667 if (result && strNE(result, "")) {
89f7b9aa 668 setlocale(LC_CTYPE, result);
bbc98134
KW
669 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
670 __FILE__, __LINE__,
671 _setlocale_debug_string(LC_CTYPE, result, "not captured")));
89f7b9aa
KW
672 }
673# endif
674# ifdef USE_LOCALE_COLLATE
675 result = PerlEnv_getenv("LC_COLLATE");
730252b2 676 if (result && strNE(result, "")) {
89f7b9aa 677 setlocale(LC_COLLATE, result);
bbc98134
KW
678 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
679 __FILE__, __LINE__,
680 _setlocale_debug_string(LC_COLLATE, result, "not captured")));
89f7b9aa
KW
681 }
682# endif
683# ifdef USE_LOCALE_MONETARY
684 result = PerlEnv_getenv("LC_MONETARY");
730252b2 685 if (result && strNE(result, "")) {
89f7b9aa 686 setlocale(LC_MONETARY, result);
bbc98134
KW
687 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
688 __FILE__, __LINE__,
689 _setlocale_debug_string(LC_MONETARY, result, "not captured")));
89f7b9aa
KW
690 }
691# endif
692# ifdef USE_LOCALE_NUMERIC
693 result = PerlEnv_getenv("LC_NUMERIC");
730252b2 694 if (result && strNE(result, "")) {
89f7b9aa 695 setlocale(LC_NUMERIC, result);
bbc98134
KW
696 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
697 __FILE__, __LINE__,
698 _setlocale_debug_string(LC_NUMERIC, result, "not captured")));
89f7b9aa
KW
699 }
700# endif
701# ifdef USE_LOCALE_MESSAGES
702 result = PerlEnv_getenv("LC_MESSAGES");
730252b2 703 if (result && strNE(result, "")) {
89f7b9aa 704 setlocale(LC_MESSAGES, result);
bbc98134
KW
705 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
706 __FILE__, __LINE__,
707 _setlocale_debug_string(LC_MESSAGES, result, "not captured")));
89f7b9aa
KW
708 }
709# endif
710
bbc98134
KW
711 result = setlocale(LC_ALL, NULL);
712 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
713 __FILE__, __LINE__,
714 _setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 715
bbc98134 716 return result;
b385bb4d
KW
717}
718
719#endif
720
721
98994639
HS
722/*
723 * Initialize locale awareness.
724 */
725int
726Perl_init_i18nl10n(pTHX_ int printwarn)
727{
0e92a118
KW
728 /* printwarn is
729 *
730 * 0 if not to output warning when setup locale is bad
731 * 1 if to output warning based on value of PERL_BADLANG
732 * >1 if to output regardless of PERL_BADLANG
733 *
734 * returns
98994639 735 * 1 = set ok or not applicable,
0e92a118
KW
736 * 0 = fallback to a locale of lower priority
737 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
738 *
739 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
740 * set, debugging information is output.
741 *
742 * This looks more complicated than it is, mainly due to the #ifdefs.
743 *
744 * We try to set LC_ALL to the value determined by the environment. If
745 * there is no LC_ALL on this platform, we try the individual categories we
746 * know about. If this works, we are done.
747 *
748 * But if it doesn't work, we have to do something else. We search the
749 * environment variables ourselves instead of relying on the system to do
750 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
751 * think there is one), and the ultimate fallback "C". This is all done in
752 * the same loop as above to avoid duplicating code, but it makes things
753 * more complex. After the original failure, we add the fallback
754 * possibilities to the list of locales to try, and iterate the loop
755 * through them all until one succeeds.
756 *
757 * On Ultrix, the locale MUST come from the environment, so there is
758 * preliminary code to set it. I (khw) am not sure that it is necessary,
759 * and that this couldn't be folded into the loop, but barring any real
760 * platforms to test on, it's staying as-is
761 *
762 * A slight complication is that in embedded Perls, the locale may already
763 * be set-up, and we don't want to get it from the normal environment
764 * variables. This is handled by having a special environment variable
765 * indicate we're in this situation. We simply set setlocale's 2nd
766 * parameter to be a NULL instead of "". That indicates to setlocale that
767 * it is not to change anything, but to return the current value,
768 * effectively initializing perl's db to what the locale already is.
769 *
770 * We play the same trick with NULL if a LC_ALL succeeds. We call
771 * setlocale() on the individual categores with NULL to get their existing
772 * values for our db, instead of trying to change them.
773 * */
98994639 774
0e92a118
KW
775 int ok = 1;
776
98994639 777#if defined(USE_LOCALE)
98994639
HS
778#ifdef USE_LOCALE_CTYPE
779 char *curctype = NULL;
780#endif /* USE_LOCALE_CTYPE */
781#ifdef USE_LOCALE_COLLATE
782 char *curcoll = NULL;
783#endif /* USE_LOCALE_COLLATE */
784#ifdef USE_LOCALE_NUMERIC
785 char *curnum = NULL;
786#endif /* USE_LOCALE_NUMERIC */
787#ifdef __GLIBC__
175c4cf9 788 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
98994639 789#endif
65ebb059 790
ccd65d51
KW
791 /* NULL uses the existing already set up locale */
792 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
793 ? NULL
794 : "";
5d1187d1
KW
795#ifdef DEBUGGING
796 const bool debug = (PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT"))
797 ? TRUE
798 : FALSE;
799# define DEBUG_LOCALE_INIT(category, locale, result) \
800 STMT_START { \
801 if (debug) { \
802 PerlIO_printf(Perl_debug_log, \
803 "%s:%d: %s\n", \
804 __FILE__, __LINE__, \
805 _setlocale_debug_string(category, \
806 locale, \
807 result)); \
808 } \
809 } STMT_END
810#else
811# define DEBUG_LOCALE_INIT(a,b,c)
812#endif
c3fcd832
KW
813 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
814 unsigned int trial_locales_count;
175c4cf9
KW
815 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
816 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 817 bool setlocale_failure = FALSE;
65ebb059
KW
818 unsigned int i;
819 char *p;
175c4cf9
KW
820
821 /* A later getenv() could zap this, so only use here */
822 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
823
824 const bool locwarn = (printwarn > 1
825 || (printwarn
826 && (! bad_lang_use_once
22ff3130
HS
827 || (
828 /* disallow with "" or "0" */
829 *bad_lang_use_once
830 && strNE("0", bad_lang_use_once)))));
0e92a118 831 bool done = FALSE;
5d1187d1
KW
832 char * sl_result; /* return from setlocale() */
833 char * locale_param;
6bce99ee
JH
834#ifdef WIN32
835 /* In some systems you can find out the system default locale
836 * and use that as the fallback locale. */
837# define SYSTEM_DEFAULT_LOCALE
838#endif
839#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059 840 const char *system_default_locale = NULL;
6bce99ee 841#endif
98994639 842
0e92a118
KW
843#ifndef LOCALE_ENVIRON_REQUIRED
844 PERL_UNUSED_VAR(done);
5d1187d1 845 PERL_UNUSED_VAR(locale_param);
0e92a118 846#else
98994639
HS
847
848 /*
849 * Ultrix setlocale(..., "") fails if there are no environment
850 * variables from which to get a locale name.
851 */
852
b3e384bf 853# ifdef LC_ALL
98994639 854 if (lang) {
5d1187d1
KW
855 sl_result = my_setlocale(LC_ALL, setlocale_init);
856 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result);
857 if (sl_result)
98994639
HS
858 done = TRUE;
859 else
860 setlocale_failure = TRUE;
861 }
5d1187d1 862 if (! setlocale_failure) {
b3e384bf 863# ifdef USE_LOCALE_CTYPE
5d1187d1
KW
864 locale_param = (! done && (lang || PerlEnv_getenv("LC_CTYPE")))
865 ? setlocale_init
866 : NULL;
867 curctype = my_setlocale(LC_CTYPE, locale_param);
868 DEBUG_LOCALE_INIT(LC_CTYPE, locale_param, sl_result);
869 if (! curctype)
98994639
HS
870 setlocale_failure = TRUE;
871 else
872 curctype = savepv(curctype);
b3e384bf
KW
873# endif /* USE_LOCALE_CTYPE */
874# ifdef USE_LOCALE_COLLATE
5d1187d1
KW
875 locale_param = (! done && (lang || PerlEnv_getenv("LC_COLLATE")))
876 ? setlocale_init
877 : NULL;
878 curcoll = my_setlocale(LC_COLLATE, locale_param);
879 DEBUG_LOCALE_INIT(LC_COLLATE, locale_param, sl_result);
880 if (! curcoll)
98994639
HS
881 setlocale_failure = TRUE;
882 else
883 curcoll = savepv(curcoll);
b3e384bf
KW
884# endif /* USE_LOCALE_COLLATE */
885# ifdef USE_LOCALE_NUMERIC
5d1187d1
KW
886 locale_param = (! done && (lang || PerlEnv_getenv("LC_NUMERIC")))
887 ? setlocale_init
888 : NULL;
889 curnum = my_setlocale(LC_NUMERIC, locale_param);
890 DEBUG_LOCALE_INIT(LC_NUMERIC, locale_param, sl_result);
891 if (! curnum)
98994639
HS
892 setlocale_failure = TRUE;
893 else
894 curnum = savepv(curnum);
b3e384bf 895# endif /* USE_LOCALE_NUMERIC */
a782673d 896# ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
897 locale_param = (! done && (lang || PerlEnv_getenv("LC_MESSAGES")))
898 ? setlocale_init
899 : NULL;
900 sl_result = my_setlocale(LC_MESSAGES, locale_param);
901 DEBUG_LOCALE_INIT(LC_MESSAGES, locale_param, sl_result);
902 if (! sl_result)
a782673d
KW
903 setlocale_failure = TRUE;
904 }
905# endif /* USE_LOCALE_MESSAGES */
c835d6be 906# ifdef USE_LOCALE_MONETARY
5d1187d1
KW
907 locale_param = (! done && (lang || PerlEnv_getenv("LC_MONETARY")))
908 ? setlocale_init
909 : NULL;
910 sl_result = my_setlocale(LC_MONETARY, locale_param);
911 DEBUG_LOCALE_INIT(LC_MONETARY, locale_param, sl_result);
912 if (! sl_result) {
c835d6be
KW
913 setlocale_failure = TRUE;
914 }
915# endif /* USE_LOCALE_MONETARY */
98994639
HS
916 }
917
b3e384bf 918# endif /* LC_ALL */
98994639
HS
919
920#endif /* !LOCALE_ENVIRON_REQUIRED */
921
65ebb059 922 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
923 * the list. Normally the loop is executed just once. But if setting the
924 * locale fails, inside the loop we add fallback trials to the array and so
925 * will execute the loop multiple times */
c3fcd832
KW
926 trial_locales[0] = setlocale_init;
927 trial_locales_count = 1;
65ebb059
KW
928 for (i= 0; i < trial_locales_count; i++) {
929 const char * trial_locale = trial_locales[i];
930
931 if (i > 0) {
932
933 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
934 * when i==0, but I (khw) don't think that behavior makes much
935 * sense */
936 setlocale_failure = FALSE;
937
6bce99ee
JH
938#ifdef SYSTEM_DEFAULT_LOCALE
939# ifdef WIN32
65ebb059
KW
940 /* On Windows machines, an entry of "" after the 0th means to use
941 * the system default locale, which we now proceed to get. */
942 if (strEQ(trial_locale, "")) {
943 unsigned int j;
944
945 /* Note that this may change the locale, but we are going to do
946 * that anyway just below */
947 system_default_locale = setlocale(LC_ALL, "");
5d1187d1 948 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059
KW
949
950 /* Skip if invalid or it's already on the list of locales to
951 * try */
952 if (! system_default_locale) {
953 goto next_iteration;
954 }
955 for (j = 0; j < trial_locales_count; j++) {
956 if (strEQ(system_default_locale, trial_locales[j])) {
957 goto next_iteration;
958 }
959 }
960
961 trial_locale = system_default_locale;
962 }
6bce99ee
JH
963# endif /* WIN32 */
964#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
965 }
966
98994639 967#ifdef LC_ALL
5d1187d1
KW
968 sl_result = my_setlocale(LC_ALL, trial_locale);
969 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result);
970 if (! sl_result) {
49c85077 971 setlocale_failure = TRUE;
7cd8b568
KW
972 }
973 else {
974 /* Since LC_ALL succeeded, it should have changed all the other
975 * categories it can to its value; so we massage things so that the
976 * setlocales below just return their category's current values.
977 * This adequately handles the case in NetBSD where LC_COLLATE may
978 * not be defined for a locale, and setting it individually will
979 * fail, whereas setting LC_ALL suceeds, leaving LC_COLLATE set to
980 * the POSIX locale. */
981 trial_locale = NULL;
982 }
98994639
HS
983#endif /* LC_ALL */
984
49c85077 985 if (!setlocale_failure) {
98994639 986#ifdef USE_LOCALE_CTYPE
49c85077 987 Safefree(curctype);
5d1187d1
KW
988 curctype = my_setlocale(LC_CTYPE, trial_locale);
989 DEBUG_LOCALE_INIT(LC_CTYPE, trial_locale, curctype);
990 if (! curctype)
49c85077
KW
991 setlocale_failure = TRUE;
992 else
993 curctype = savepv(curctype);
98994639
HS
994#endif /* USE_LOCALE_CTYPE */
995#ifdef USE_LOCALE_COLLATE
49c85077 996 Safefree(curcoll);
5d1187d1
KW
997 curcoll = my_setlocale(LC_COLLATE, trial_locale);
998 DEBUG_LOCALE_INIT(LC_COLLATE, trial_locale, curcoll);
999 if (! curcoll)
49c85077
KW
1000 setlocale_failure = TRUE;
1001 else
1002 curcoll = savepv(curcoll);
98994639
HS
1003#endif /* USE_LOCALE_COLLATE */
1004#ifdef USE_LOCALE_NUMERIC
49c85077 1005 Safefree(curnum);
5d1187d1
KW
1006 curnum = my_setlocale(LC_NUMERIC, trial_locale);
1007 DEBUG_LOCALE_INIT(LC_NUMERIC, trial_locale, curnum);
1008 if (! curnum)
49c85077
KW
1009 setlocale_failure = TRUE;
1010 else
1011 curnum = savepv(curnum);
98994639 1012#endif /* USE_LOCALE_NUMERIC */
a782673d 1013#ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
1014 sl_result = my_setlocale(LC_MESSAGES, trial_locale);
1015 DEBUG_LOCALE_INIT(LC_MESSAGES, trial_locale, sl_result);
1016 if (! (sl_result))
a782673d
KW
1017 setlocale_failure = TRUE;
1018#endif /* USE_LOCALE_MESSAGES */
c835d6be 1019#ifdef USE_LOCALE_MONETARY
5d1187d1
KW
1020 sl_result = my_setlocale(LC_MONETARY, trial_locale);
1021 DEBUG_LOCALE_INIT(LC_MONETARY, trial_locale, sl_result);
1022 if (! (sl_result))
c835d6be
KW
1023 setlocale_failure = TRUE;
1024#endif /* USE_LOCALE_MONETARY */
1025
49c85077
KW
1026 if (! setlocale_failure) { /* Success */
1027 break;
1028 }
65ebb059 1029 }
98994639 1030
49c85077
KW
1031 /* Here, something failed; will need to try a fallback. */
1032 ok = 0;
65ebb059 1033
49c85077
KW
1034 if (i == 0) {
1035 unsigned int j;
98994639 1036
65ebb059 1037 if (locwarn) { /* Output failure info only on the first one */
98994639
HS
1038#ifdef LC_ALL
1039
49c85077
KW
1040 PerlIO_printf(Perl_error_log,
1041 "perl: warning: Setting locale failed.\n");
98994639
HS
1042
1043#else /* !LC_ALL */
1044
49c85077
KW
1045 PerlIO_printf(Perl_error_log,
1046 "perl: warning: Setting locale failed for the categories:\n\t");
20a240df 1047# ifdef USE_LOCALE_CTYPE
49c85077
KW
1048 if (! curctype)
1049 PerlIO_printf(Perl_error_log, "LC_CTYPE ");
20a240df
KW
1050# endif /* USE_LOCALE_CTYPE */
1051# ifdef USE_LOCALE_COLLATE
49c85077
KW
1052 if (! curcoll)
1053 PerlIO_printf(Perl_error_log, "LC_COLLATE ");
20a240df
KW
1054# endif /* USE_LOCALE_COLLATE */
1055# ifdef USE_LOCALE_NUMERIC
49c85077
KW
1056 if (! curnum)
1057 PerlIO_printf(Perl_error_log, "LC_NUMERIC ");
20a240df 1058# endif /* USE_LOCALE_NUMERIC */
a782673d 1059 PerlIO_printf(Perl_error_log, "and possibly others\n");
98994639
HS
1060
1061#endif /* LC_ALL */
1062
49c85077
KW
1063 PerlIO_printf(Perl_error_log,
1064 "perl: warning: Please check that your locale settings:\n");
98994639
HS
1065
1066#ifdef __GLIBC__
49c85077
KW
1067 PerlIO_printf(Perl_error_log,
1068 "\tLANGUAGE = %c%s%c,\n",
1069 language ? '"' : '(',
1070 language ? language : "unset",
1071 language ? '"' : ')');
98994639
HS
1072#endif
1073
49c85077
KW
1074 PerlIO_printf(Perl_error_log,
1075 "\tLC_ALL = %c%s%c,\n",
1076 lc_all ? '"' : '(',
1077 lc_all ? lc_all : "unset",
1078 lc_all ? '"' : ')');
98994639
HS
1079
1080#if defined(USE_ENVIRON_ARRAY)
49c85077
KW
1081 {
1082 char **e;
1083 for (e = environ; *e; e++) {
1084 if (strnEQ(*e, "LC_", 3)
1085 && strnNE(*e, "LC_ALL=", 7)
1086 && (p = strchr(*e, '=')))
1087 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
1088 (int)(p - *e), *e, p + 1);
1089 }
1090 }
98994639 1091#else
49c85077
KW
1092 PerlIO_printf(Perl_error_log,
1093 "\t(possibly more locale environment variables)\n");
98994639
HS
1094#endif
1095
49c85077
KW
1096 PerlIO_printf(Perl_error_log,
1097 "\tLANG = %c%s%c\n",
1098 lang ? '"' : '(',
1099 lang ? lang : "unset",
1100 lang ? '"' : ')');
98994639 1101
49c85077
KW
1102 PerlIO_printf(Perl_error_log,
1103 " are supported and installed on your system.\n");
1104 }
98994639 1105
65ebb059 1106 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 1107 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
1108 * usually change the upper bound of the loop we are in.
1109 *
1110 * Since the system's default way of setting the locale has not
1111 * found one that works, We use Perl's defined ordering: LC_ALL,
1112 * LANG, and the C locale. We don't try the same locale twice, so
1113 * don't add to the list if already there. (On POSIX systems, the
1114 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
1115 * but there's no harm done by doing it explicitly.
1116 *
1117 * Note that this tries the LC_ALL environment variable even on
1118 * systems which have no LC_ALL locale setting. This may or may
1119 * not have been originally intentional, but there's no real need
1120 * to change the behavior. */
65ebb059
KW
1121 if (lc_all) {
1122 for (j = 0; j < trial_locales_count; j++) {
1123 if (strEQ(lc_all, trial_locales[j])) {
1124 goto done_lc_all;
1125 }
1126 }
1127 trial_locales[trial_locales_count++] = lc_all;
1128 }
1129 done_lc_all:
98994639 1130
65ebb059
KW
1131 if (lang) {
1132 for (j = 0; j < trial_locales_count; j++) {
1133 if (strEQ(lang, trial_locales[j])) {
1134 goto done_lang;
1135 }
1136 }
1137 trial_locales[trial_locales_count++] = lang;
1138 }
1139 done_lang:
1140
1141#if defined(WIN32) && defined(LC_ALL)
1142 /* For Windows, we also try the system default locale before "C".
1143 * (If there exists a Windows without LC_ALL we skip this because
1144 * it gets too complicated. For those, the "C" is the next
1145 * fallback possibility). The "" is the same as the 0th element of
1146 * the array, but the code at the loop above knows to treat it
1147 * differently when not the 0th */
1148 trial_locales[trial_locales_count++] = "";
1149#endif
1150
1151 for (j = 0; j < trial_locales_count; j++) {
1152 if (strEQ("C", trial_locales[j])) {
1153 goto done_C;
1154 }
1155 }
1156 trial_locales[trial_locales_count++] = "C";
98994639 1157
65ebb059
KW
1158 done_C: ;
1159 } /* end of first time through the loop */
98994639 1160
65ebb059
KW
1161#ifdef WIN32
1162 next_iteration: ;
1163#endif
1164
1165 } /* end of looping through the trial locales */
1166
1167 if (ok < 1) { /* If we tried to fallback */
1168 const char* msg;
1169 if (! setlocale_failure) { /* fallback succeeded */
1170 msg = "Falling back to";
1171 }
1172 else { /* fallback failed */
98994639 1173
65ebb059
KW
1174 /* We dropped off the end of the loop, so have to decrement i to
1175 * get back to the value the last time through */
1176 i--;
98994639 1177
65ebb059
KW
1178 ok = -1;
1179 msg = "Failed to fall back to";
1180
1181 /* To continue, we should use whatever values we've got */
98994639 1182#ifdef USE_LOCALE_CTYPE
49c85077
KW
1183 Safefree(curctype);
1184 curctype = savepv(setlocale(LC_CTYPE, NULL));
5d1187d1 1185 DEBUG_LOCALE_INIT(LC_CTYPE, NULL, curctype);
98994639
HS
1186#endif /* USE_LOCALE_CTYPE */
1187#ifdef USE_LOCALE_COLLATE
49c85077
KW
1188 Safefree(curcoll);
1189 curcoll = savepv(setlocale(LC_COLLATE, NULL));
5d1187d1 1190 DEBUG_LOCALE_INIT(LC_COLLATE, NULL, curcoll);
98994639
HS
1191#endif /* USE_LOCALE_COLLATE */
1192#ifdef USE_LOCALE_NUMERIC
49c85077
KW
1193 Safefree(curnum);
1194 curnum = savepv(setlocale(LC_NUMERIC, NULL));
5d1187d1 1195 DEBUG_LOCALE_INIT(LC_NUMERIC, NULL, curnum);
98994639 1196#endif /* USE_LOCALE_NUMERIC */
65ebb059
KW
1197 }
1198
1199 if (locwarn) {
1200 const char * description;
1201 const char * name = "";
1202 if (strEQ(trial_locales[i], "C")) {
1203 description = "the standard locale";
1204 name = "C";
1205 }
6bce99ee 1206#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059
KW
1207 else if (strEQ(trial_locales[i], "")) {
1208 description = "the system default locale";
1209 if (system_default_locale) {
1210 name = system_default_locale;
1211 }
1212 }
6bce99ee 1213#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
1214 else {
1215 description = "a fallback locale";
1216 name = trial_locales[i];
1217 }
1218 if (name && strNE(name, "")) {
1219 PerlIO_printf(Perl_error_log,
1220 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
1221 }
1222 else {
1223 PerlIO_printf(Perl_error_log,
1224 "perl: warning: %s %s.\n", msg, description);
1225 }
1226 }
1227 } /* End of tried to fallback */
98994639
HS
1228
1229#ifdef USE_LOCALE_CTYPE
1230 new_ctype(curctype);
1231#endif /* USE_LOCALE_CTYPE */
1232
1233#ifdef USE_LOCALE_COLLATE
1234 new_collate(curcoll);
1235#endif /* USE_LOCALE_COLLATE */
1236
1237#ifdef USE_LOCALE_NUMERIC
1238 new_numeric(curnum);
1239#endif /* USE_LOCALE_NUMERIC */
b310b053 1240
8ef6e574 1241#if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
49c85077
KW
1242 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
1243 * locale is UTF-8. If PL_utf8locale and PL_unicode (set by -C or by
1244 * $ENV{PERL_UNICODE}) are true, perl.c:S_parse_body() will turn on the
1245 * PerlIO :utf8 layer on STDIN, STDOUT, STDERR, _and_ the default open
1246 * discipline. */
c1284011 1247 PL_utf8locale = _is_cur_LC_category_utf8(LC_CTYPE);
49c85077 1248
a05d7ebb 1249 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
1250 This is an alternative to using the -C command line switch
1251 (the -C if present will override this). */
1252 {
dd374669 1253 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 1254 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
1255 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
1256 PL_utf8cache = -1;
b310b053 1257 }
ec71e770 1258#endif
b310b053 1259
98994639 1260#ifdef USE_LOCALE_CTYPE
43c5f42d 1261 Safefree(curctype);
98994639
HS
1262#endif /* USE_LOCALE_CTYPE */
1263#ifdef USE_LOCALE_COLLATE
43c5f42d 1264 Safefree(curcoll);
98994639
HS
1265#endif /* USE_LOCALE_COLLATE */
1266#ifdef USE_LOCALE_NUMERIC
43c5f42d 1267 Safefree(curnum);
98994639 1268#endif /* USE_LOCALE_NUMERIC */
8ef6e574 1269
175c4cf9
KW
1270#ifdef __GLIBC__
1271 Safefree(language);
1272#endif
1273
1274 Safefree(lc_all);
1275 Safefree(lang);
1276
e3305790
KW
1277#else /* !USE_LOCALE */
1278 PERL_UNUSED_ARG(printwarn);
1279#endif /* USE_LOCALE */
1280
98994639
HS
1281 return ok;
1282}
1283
49c85077 1284
98994639
HS
1285#ifdef USE_LOCALE_COLLATE
1286
1287/*
1288 * mem_collxfrm() is a bit like strxfrm() but with two important
1289 * differences. First, it handles embedded NULs. Second, it allocates
1290 * a bit more memory than needed for the transformed data itself.
1291 * The real transformed data begins at offset sizeof(collationix).
d35fca5f
KW
1292 * *xlen is set to the length of that, and doesn't include the collation index
1293 * size.
98994639
HS
1294 * Please see sv_collxfrm() to see how this is used.
1295 */
166f8a29 1296
98994639
HS
1297char *
1298Perl_mem_collxfrm(pTHX_ const char *s, STRLEN len, STRLEN *xlen)
1299{
1300 char *xbuf;
1301 STRLEN xAlloc, xin, xout; /* xalloc is a reserved word in VC */
1302
7918f24d
NC
1303 PERL_ARGS_ASSERT_MEM_COLLXFRM;
1304
59c018b9
KW
1305 /* The first element in the output is the collation id, used by
1306 * sv_collxfrm(); then comes the space for the transformed string. The
1307 * equation should give us a good estimate as to how much is needed */
98994639 1308 xAlloc = sizeof(PL_collation_ix) + PL_collxfrm_base + (PL_collxfrm_mult * len) + 1;
a02a5408 1309 Newx(xbuf, xAlloc, char);
bb0f664e 1310 if (UNLIKELY(! xbuf))
98994639
HS
1311 goto bad;
1312
d35fca5f 1313 /* Store the collation id */
98994639
HS
1314 *(U32*)xbuf = PL_collation_ix;
1315 xout = sizeof(PL_collation_ix);
d35fca5f
KW
1316
1317 /* Then the transformation of the input. We loop until successful, or we
1318 * give up */
98994639 1319 for (xin = 0; xin < len; ) {
224e8ef5 1320 Size_t xused;
98994639
HS
1321
1322 for (;;) {
1323 xused = strxfrm(xbuf + xout, s + xin, xAlloc - xout);
d35fca5f
KW
1324
1325 /* If the transformed string occupies less space than we told
1326 * strxfrm() was available, it means it successfully transformed
1327 * the whole string. */
eb160463 1328 if ((STRLEN)xused < xAlloc - xout)
98994639 1329 break;
bb0f664e
KW
1330
1331 if (UNLIKELY(xused >= PERL_INT_MAX))
1332 goto bad;
d35fca5f
KW
1333
1334 /* Otherwise it should be that the transformation stopped in the
1335 * middle because it ran out of space. Malloc more, and try again.
1336 * */
98994639
HS
1337 xAlloc = (2 * xAlloc) + 1;
1338 Renew(xbuf, xAlloc, char);
bb0f664e 1339 if (UNLIKELY(! xbuf))
98994639
HS
1340 goto bad;
1341 }
1342
1343 xin += strlen(s + xin) + 1;
1344 xout += xused;
1345
1346 /* Embedded NULs are understood but silently skipped
1347 * because they make no sense in locale collation. */
1348 }
1349
1350 xbuf[xout] = '\0';
1351 *xlen = xout - sizeof(PL_collation_ix);
1352 return xbuf;
1353
1354 bad:
1355 Safefree(xbuf);
1356 *xlen = 0;
1357 return NULL;
1358}
1359
1360#endif /* USE_LOCALE_COLLATE */
1361
8ef6e574
KW
1362#ifdef USE_LOCALE
1363
c1284011
KW
1364bool
1365Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
1366{
1367 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
1368 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 1369 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
1370 * could give the wrong result. The result will very likely be correct for
1371 * languages that have commonly used non-ASCII characters, but for notably
1372 * English, it comes down to if the locale's name ends in something like
1373 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61
KW
1374
1375 char *save_input_locale = NULL;
7d74bb61
KW
1376 STRLEN final_pos;
1377
8ef6e574 1378#ifdef LC_ALL
7d74bb61 1379 assert(category != LC_ALL);
8ef6e574 1380#endif
7d74bb61
KW
1381
1382 /* First dispose of the trivial cases */
b07fffd1 1383 save_input_locale = setlocale(category, NULL);
7d74bb61 1384 if (! save_input_locale) {
69014004
KW
1385 DEBUG_L(PerlIO_printf(Perl_debug_log,
1386 "Could not find current locale for category %d\n",
1387 category));
7d74bb61
KW
1388 return FALSE; /* XXX maybe should croak */
1389 }
b07fffd1 1390 save_input_locale = stdize_locale(savepv(save_input_locale));
a39edc4c 1391 if (isNAME_C_OR_POSIX(save_input_locale)) {
69014004
KW
1392 DEBUG_L(PerlIO_printf(Perl_debug_log,
1393 "Current locale for category %d is %s\n",
1394 category, save_input_locale));
b07fffd1 1395 Safefree(save_input_locale);
7d74bb61
KW
1396 return FALSE;
1397 }
1398
1d958db2
KW
1399#if defined(USE_LOCALE_CTYPE) \
1400 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 1401
1d958db2 1402 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
1403
1404 char *save_ctype_locale = NULL;
119ee68b 1405 bool is_utf8;
7d74bb61 1406
119ee68b 1407 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
1408
1409 /* Get the current LC_CTYPE locale */
4f72bb37 1410 save_ctype_locale = setlocale(LC_CTYPE, NULL);
7d74bb61 1411 if (! save_ctype_locale) {
69014004
KW
1412 DEBUG_L(PerlIO_printf(Perl_debug_log,
1413 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
1414 goto cant_use_nllanginfo;
1415 }
4f72bb37 1416 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
1417
1418 /* If LC_CTYPE and the desired category use the same locale, this
1419 * means that finding the value for LC_CTYPE is the same as finding
1420 * the value for the desired category. Otherwise, switch LC_CTYPE
1421 * to the desired category's locale */
1422 if (strEQ(save_ctype_locale, save_input_locale)) {
1423 Safefree(save_ctype_locale);
1424 save_ctype_locale = NULL;
1425 }
1426 else if (! setlocale(LC_CTYPE, save_input_locale)) {
69014004
KW
1427 DEBUG_L(PerlIO_printf(Perl_debug_log,
1428 "Could not change LC_CTYPE locale to %s\n",
1429 save_input_locale));
7d74bb61
KW
1430 Safefree(save_ctype_locale);
1431 goto cant_use_nllanginfo;
1432 }
1433 }
1434
69014004
KW
1435 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
1436 save_input_locale));
1437
7d74bb61 1438 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
1439 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
1440 * should give the correct results */
119ee68b 1441
1d958db2
KW
1442# if defined(HAS_NL_LANGINFO) && defined(CODESET)
1443 {
4f72bb37 1444 char *codeset = nl_langinfo(CODESET);
1d958db2 1445 if (codeset && strNE(codeset, "")) {
4f72bb37 1446 codeset = savepv(codeset);
119ee68b 1447
1d958db2
KW
1448 /* If we switched LC_CTYPE, switch back */
1449 if (save_ctype_locale) {
1450 setlocale(LC_CTYPE, save_ctype_locale);
1451 Safefree(save_ctype_locale);
1452 }
1453
1454 is_utf8 = foldEQ(codeset, STR_WITH_LEN("UTF-8"))
1455 || foldEQ(codeset, STR_WITH_LEN("UTF8"));
1456
69014004
KW
1457 DEBUG_L(PerlIO_printf(Perl_debug_log,
1458 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
1459 codeset, is_utf8));
1d958db2
KW
1460 Safefree(codeset);
1461 Safefree(save_input_locale);
1462 return is_utf8;
1463 }
119ee68b
KW
1464 }
1465
1d958db2
KW
1466# endif
1467# ifdef MB_CUR_MAX
1468
1469 /* Here, either we don't have nl_langinfo, or it didn't return a
1470 * codeset. Try MB_CUR_MAX */
1471
119ee68b
KW
1472 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
1473 * Unicode code point. Since UTF-8 is the only non-single byte
1474 * encoding we handle, we just say any such encoding is UTF-8, and if
1475 * turns out to be wrong, other things will fail */
1476 is_utf8 = MB_CUR_MAX >= 4;
1477
69014004
KW
1478 DEBUG_L(PerlIO_printf(Perl_debug_log,
1479 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
1480 (int) MB_CUR_MAX, is_utf8));
1481
119ee68b
KW
1482 Safefree(save_input_locale);
1483
1484# ifdef HAS_MBTOWC
1485
1486 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
1487 * since they are both in the C99 standard. We can feed a known byte
1488 * string to the latter function, and check that it gives the expected
1489 * result */
1490 if (is_utf8) {
1491 wchar_t wc;
856b881c 1492 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 1493 errno = 0;
f019f68f 1494 if ((size_t)mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8))
119ee68b
KW
1495 != strlen(HYPHEN_UTF8)
1496 || wc != (wchar_t) 0x2010)
1497 {
1498 is_utf8 = FALSE;
abdcbdb8 1499 DEBUG_L(PerlIO_printf(Perl_debug_log, "\thyphen=U+%x\n", (unsigned int)wc));
69014004
KW
1500 DEBUG_L(PerlIO_printf(Perl_debug_log,
1501 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
1502 mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8)), errno));
119ee68b
KW
1503 }
1504 }
119ee68b
KW
1505# endif
1506
1d958db2
KW
1507 /* If we switched LC_CTYPE, switch back */
1508 if (save_ctype_locale) {
1509 setlocale(LC_CTYPE, save_ctype_locale);
1510 Safefree(save_ctype_locale);
119ee68b 1511 }
7d74bb61 1512
1d958db2 1513 return is_utf8;
119ee68b 1514# endif
7d74bb61 1515 }
119ee68b 1516
7d74bb61
KW
1517 cant_use_nllanginfo:
1518
0080c90a
KW
1519#else /* nl_langinfo should work if available, so don't bother compiling this
1520 fallback code. The final fallback of looking at the name is
1521 compiled, and will be executed if nl_langinfo fails */
7d74bb61 1522
97f4de96
KW
1523 /* nl_langinfo not available or failed somehow. Next try looking at the
1524 * currency symbol to see if it disambiguates things. Often that will be
1525 * in the native script, and if the symbol isn't in UTF-8, we know that the
1526 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
1527 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
1528 * */
fa9b773e
KW
1529
1530#ifdef HAS_LOCALECONV
fa9b773e 1531# ifdef USE_LOCALE_MONETARY
fa9b773e
KW
1532 {
1533 char *save_monetary_locale = NULL;
fa9b773e 1534 bool only_ascii = FALSE;
13542a67
KW
1535 bool is_utf8 = FALSE;
1536 struct lconv* lc;
fa9b773e 1537
97f4de96
KW
1538 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
1539 * the desired category, if it isn't that locale already */
1540
fa9b773e
KW
1541 if (category != LC_MONETARY) {
1542
4f72bb37 1543 save_monetary_locale = setlocale(LC_MONETARY, NULL);
fa9b773e 1544 if (! save_monetary_locale) {
69014004
KW
1545 DEBUG_L(PerlIO_printf(Perl_debug_log,
1546 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
1547 goto cant_use_monetary;
1548 }
4f72bb37 1549 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 1550
13542a67
KW
1551 if (strEQ(save_monetary_locale, save_input_locale)) {
1552 Safefree(save_monetary_locale);
1553 save_monetary_locale = NULL;
1554 }
1555 else if (! setlocale(LC_MONETARY, save_input_locale)) {
59c234b4
KW
1556 DEBUG_L(PerlIO_printf(Perl_debug_log,
1557 "Could not change LC_MONETARY locale to %s\n",
1558 save_input_locale));
1559 Safefree(save_monetary_locale);
1560 goto cant_use_monetary;
fa9b773e
KW
1561 }
1562 }
1563
1564 /* Here the current LC_MONETARY is set to the locale of the category
1565 * whose information is desired. */
1566
13542a67
KW
1567 lc = localeconv();
1568 if (! lc
1569 || ! lc->currency_symbol
9f10db87 1570 || is_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
1571 {
1572 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));
1573 only_ascii = TRUE;
1574 }
1575 else {
1576 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
1577 }
1578
1579 /* If we changed it, restore LC_MONETARY to its original locale */
1580 if (save_monetary_locale) {
1581 setlocale(LC_MONETARY, save_monetary_locale);
1582 Safefree(save_monetary_locale);
1583 }
1584
13542a67 1585 if (! only_ascii) {
fa9b773e 1586
59c234b4
KW
1587 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
1588 * otherwise assume the locale is UTF-8 if and only if the symbol
1589 * is non-ascii UTF-8. */
1590 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
1591 save_input_locale, is_utf8));
1592 Safefree(save_input_locale);
1593 return is_utf8;
13542a67 1594 }
fa9b773e
KW
1595 }
1596 cant_use_monetary:
1597
1598# endif /* USE_LOCALE_MONETARY */
1599#endif /* HAS_LOCALECONV */
1600
15f7e74e
KW
1601#if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
1602
1603/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
1604 * the names of the months and weekdays, timezone, and am/pm indicator */
1605 {
1606 char *save_time_locale = NULL;
1607 int hour = 10;
1608 bool is_dst = FALSE;
1609 int dom = 1;
1610 int month = 0;
1611 int i;
1612 char * formatted_time;
1613
1614
1615 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
1616 * desired category, if it isn't that locale already */
1617
1618 if (category != LC_TIME) {
1619
1620 save_time_locale = setlocale(LC_TIME, NULL);
1621 if (! save_time_locale) {
1622 DEBUG_L(PerlIO_printf(Perl_debug_log,
1623 "Could not find current locale for LC_TIME\n"));
1624 goto cant_use_time;
1625 }
1626 save_time_locale = stdize_locale(savepv(save_time_locale));
1627
1628 if (strEQ(save_time_locale, save_input_locale)) {
1629 Safefree(save_time_locale);
1630 save_time_locale = NULL;
1631 }
1632 else if (! setlocale(LC_TIME, save_input_locale)) {
1633 DEBUG_L(PerlIO_printf(Perl_debug_log,
1634 "Could not change LC_TIME locale to %s\n",
1635 save_input_locale));
1636 Safefree(save_time_locale);
1637 goto cant_use_time;
1638 }
1639 }
1640
1641 /* Here the current LC_TIME is set to the locale of the category
1642 * whose information is desired. Look at all the days of the week and
9f10db87 1643 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
1644 * characters. The first such a one found will tell us if the locale
1645 * is UTF-8 or not */
1646
1647 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
1648 formatted_time = my_strftime("%A %B %Z %p",
1649 0, 0, hour, dom, month, 112, 0, 0, is_dst);
9f10db87 1650 if (! formatted_time || is_invariant_string((U8 *) formatted_time, 0)) {
15f7e74e
KW
1651
1652 /* Here, we didn't find a non-ASCII. Try the next time through
1653 * with the complemented dst and am/pm, and try with the next
1654 * weekday. After we have gotten all weekdays, try the next
1655 * month */
1656 is_dst = ! is_dst;
1657 hour = (hour + 12) % 24;
1658 dom++;
1659 if (i > 6) {
1660 month++;
1661 }
1662 continue;
1663 }
1664
1665 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
1666 * false otherwise. But first, restore LC_TIME to its original
1667 * locale if we changed it */
1668 if (save_time_locale) {
1669 setlocale(LC_TIME, save_time_locale);
1670 Safefree(save_time_locale);
1671 }
1672
1673 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
1674 save_input_locale,
1675 is_utf8_string((U8 *) formatted_time, 0)));
1676 Safefree(save_input_locale);
1677 return is_utf8_string((U8 *) formatted_time, 0);
1678 }
1679
1680 /* Falling off the end of the loop indicates all the names were just
1681 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
1682 * to its original locale */
1683 if (save_time_locale) {
1684 setlocale(LC_TIME, save_time_locale);
1685 Safefree(save_time_locale);
1686 }
1687 DEBUG_L(PerlIO_printf(Perl_debug_log, "All time-related words for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
1688 }
1689 cant_use_time:
1690
1691#endif
1692
1693#if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
1694
1695/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
1696 * on our dromedary test machine, which has over 700 locales. There, this
1697 * added no value to looking at the currency symbol and the time strings. I
1698 * left it in so as to avoid rewriting it if real-world experience indicates
1699 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
1700 * haven't found illegal utf8, we continue on and examine all the strerror()
1701 * messages on the platform for utf8ness. If all are ASCII, we still don't
1702 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
1703 * utf8ness. The reason this doesn't help much is that the messages may not
1704 * have been translated into the locale. The currency symbol and time strings
1705 * are much more likely to have been translated. */
1706 {
855aeb93 1707 int e;
5857e934
KW
1708 bool is_utf8 = FALSE;
1709 bool non_ascii = FALSE;
855aeb93 1710 char *save_messages_locale = NULL;
5857e934 1711 const char * errmsg = NULL;
855aeb93 1712
5857e934
KW
1713 /* Like above, we set LC_MESSAGES to the locale of the desired
1714 * category, if it isn't that locale already */
855aeb93
JH
1715
1716 if (category != LC_MESSAGES) {
1717
5857e934 1718 save_messages_locale = setlocale(LC_MESSAGES, NULL);
855aeb93 1719 if (! save_messages_locale) {
5857e934
KW
1720 DEBUG_L(PerlIO_printf(Perl_debug_log,
1721 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
1722 goto cant_use_messages;
1723 }
5857e934 1724 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
1725
1726 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
1727 Safefree(save_messages_locale);
1728 save_messages_locale = NULL;
855aeb93
JH
1729 }
1730 else if (! setlocale(LC_MESSAGES, save_input_locale)) {
5857e934
KW
1731 DEBUG_L(PerlIO_printf(Perl_debug_log,
1732 "Could not change LC_MESSAGES locale to %s\n",
1733 save_input_locale));
855aeb93
JH
1734 Safefree(save_messages_locale);
1735 goto cant_use_messages;
1736 }
1737 }
1738
1739 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
1740 * whose information is desired. Look through all the messages. We
1741 * can't use Strerror() here because it may expand to code that
1742 * segfaults in miniperl */
855aeb93 1743
5857e934
KW
1744 for (e = 0; e <= sys_nerr; e++) {
1745 errno = 0;
1746 errmsg = sys_errlist[e];
1747 if (errno || !errmsg) {
855aeb93
JH
1748 break;
1749 }
5857e934 1750 errmsg = savepv(errmsg);
9f10db87 1751 if (! is_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
1752 non_ascii = TRUE;
1753 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
1754 break;
855aeb93
JH
1755 }
1756 }
5857e934 1757 Safefree(errmsg);
855aeb93
JH
1758
1759 /* And, if we changed it, restore LC_MESSAGES to its original locale */
1760 if (save_messages_locale) {
1761 setlocale(LC_MESSAGES, save_messages_locale);
1762 Safefree(save_messages_locale);
1763 }
1764
5857e934
KW
1765 if (non_ascii) {
1766
1767 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
1768 * any non-ascii means it is one; otherwise we assume it isn't */
1769 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
1770 save_input_locale,
1771 is_utf8));
1772 Safefree(save_input_locale);
1773 return is_utf8;
1774 }
855aeb93 1775
5857e934 1776 DEBUG_L(PerlIO_printf(Perl_debug_log, "All error messages for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
855aeb93
JH
1777 }
1778 cant_use_messages:
1779
1780#endif
fa9b773e 1781
0080c90a
KW
1782#endif /* the code that is compiled when no nl_langinfo */
1783
92c0a900
KW
1784#ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
1785 UTF-8 locale */
97f4de96
KW
1786 /* As a last resort, look at the locale name to see if it matches
1787 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
1788 * return of setlocale(), is actually defined to be opaque, so we can't
1789 * really rely on the absence of various substrings in the name to indicate
1790 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
1791 * be a UTF-8 locale. Similarly for the other common names */
1792
1793 final_pos = strlen(save_input_locale) - 1;
1794 if (final_pos >= 3) {
1795 char *name = save_input_locale;
1796
1797 /* Find next 'U' or 'u' and look from there */
1798 while ((name += strcspn(name, "Uu") + 1)
1799 <= save_input_locale + final_pos - 2)
1800 {
305b8651
KW
1801 if (!isALPHA_FOLD_NE(*name, 't')
1802 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
1803 {
1804 continue;
1805 }
1806 name += 2;
1807 if (*(name) == '-') {
1808 if ((name > save_input_locale + final_pos - 1)) {
1809 break;
1810 }
1811 name++;
1812 }
1813 if (*(name) == '8') {
97f4de96
KW
1814 DEBUG_L(PerlIO_printf(Perl_debug_log,
1815 "Locale %s ends with UTF-8 in name\n",
1816 save_input_locale));
00c54b9c 1817 Safefree(save_input_locale);
97f4de96
KW
1818 return TRUE;
1819 }
1820 }
1821 DEBUG_L(PerlIO_printf(Perl_debug_log,
1822 "Locale %s doesn't end with UTF-8 in name\n",
1823 save_input_locale));
1824 }
92c0a900 1825#endif
97f4de96
KW
1826
1827#ifdef WIN32
1828 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
1829 if (final_pos >= 4
1830 && *(save_input_locale + final_pos - 0) == '1'
1831 && *(save_input_locale + final_pos - 1) == '0'
1832 && *(save_input_locale + final_pos - 2) == '0'
1833 && *(save_input_locale + final_pos - 3) == '5'
1834 && *(save_input_locale + final_pos - 4) == '6')
1835 {
1836 DEBUG_L(PerlIO_printf(Perl_debug_log,
1837 "Locale %s ends with 10056 in name, is UTF-8 locale\n",
1838 save_input_locale));
1839 Safefree(save_input_locale);
1840 return TRUE;
1841 }
1842#endif
1843
1844 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
1845 * since we are about to return FALSE anyway, there is no point in doing
1846 * this extra work */
1847#if 0
1848 if (instr(save_input_locale, "8859")) {
1849 DEBUG_L(PerlIO_printf(Perl_debug_log,
1850 "Locale %s has 8859 in name, not UTF-8 locale\n",
1851 save_input_locale));
1852 Safefree(save_input_locale);
1853 return FALSE;
1854 }
1855#endif
1856
69014004
KW
1857 DEBUG_L(PerlIO_printf(Perl_debug_log,
1858 "Assuming locale %s is not a UTF-8 locale\n",
1859 save_input_locale));
fa9b773e 1860 Safefree(save_input_locale);
7d74bb61
KW
1861 return FALSE;
1862}
1863
8ef6e574 1864#endif
7d74bb61 1865
d6ded950
KW
1866
1867bool
1868Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
1869{
1a4f13e1 1870 dVAR;
d6ded950
KW
1871 /* Internal function which returns if we are in the scope of a pragma that
1872 * enables the locale category 'category'. 'compiling' should indicate if
1873 * this is during the compilation phase (TRUE) or not (FALSE). */
1874
1875 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
1876
1877 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
1878 if (! categories || categories == &PL_sv_placeholder) {
1879 return FALSE;
1880 }
1881
1882 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
1883 * a valid unsigned */
1884 assert(category >= -1);
1885 return cBOOL(SvUV(categories) & (1U << (category + 1)));
1886}
1887
2c6ee1a7
KW
1888char *
1889Perl_my_strerror(pTHX_ const int errnum) {
a0b53297 1890 dVAR;
2c6ee1a7
KW
1891
1892 /* Uses C locale for the error text unless within scope of 'use locale' for
1893 * LC_MESSAGES */
1894
1895#ifdef USE_LOCALE_MESSAGES
8ee4b769 1896 if (! IN_LC(LC_MESSAGES)) {
a0b53297
KW
1897 char * save_locale;
1898
1899 /* We have a critical section to prevent another thread from changing
1900 * the locale out from under us (or zapping the buffer returned from
1901 * setlocale() ) */
1902 LOCALE_LOCK;
1903
1904 save_locale = setlocale(LC_MESSAGES, NULL);
a39edc4c 1905 if (! isNAME_C_OR_POSIX(save_locale)) {
2c6ee1a7
KW
1906 char *errstr;
1907
1908 /* The next setlocale likely will zap this, so create a copy */
1909 save_locale = savepv(save_locale);
1910
1911 setlocale(LC_MESSAGES, "C");
1912
1913 /* This points to the static space in Strerror, with all its
1914 * limitations */
1915 errstr = Strerror(errnum);
1916
1917 setlocale(LC_MESSAGES, save_locale);
1918 Safefree(save_locale);
a0b53297
KW
1919
1920 LOCALE_UNLOCK;
1921
2c6ee1a7
KW
1922 return errstr;
1923 }
a0b53297
KW
1924
1925 LOCALE_UNLOCK;
2c6ee1a7
KW
1926 }
1927#endif
1928
1929 return Strerror(errnum);
1930}
1931
66610fdd 1932/*
747c467a
KW
1933
1934=head1 Locale-related functions and macros
1935
1936=for apidoc sync_locale
1937
1938Changing the program's locale should be avoided by XS code. Nevertheless,
1939certain non-Perl libraries called from XS, such as C<Gtk> do so. When this
1940happens, Perl needs to be told that the locale has changed. Use this function
1941to do so, before returning to Perl.
1942
1943=cut
1944*/
1945
1946void
1947Perl_sync_locale(pTHX)
1948{
1949
1950#ifdef USE_LOCALE_CTYPE
1951 new_ctype(setlocale(LC_CTYPE, NULL));
1952#endif /* USE_LOCALE_CTYPE */
1953
1954#ifdef USE_LOCALE_COLLATE
1955 new_collate(setlocale(LC_COLLATE, NULL));
1956#endif
1957
1958#ifdef USE_LOCALE_NUMERIC
1959 set_numeric_local(); /* Switch from "C" to underlying LC_NUMERIC */
1960 new_numeric(setlocale(LC_NUMERIC, NULL));
1961#endif /* USE_LOCALE_NUMERIC */
1962
1963}
1964
5d1187d1
KW
1965#if defined(DEBUGGING) && defined(USE_LOCALE)
1966
1967char *
1968Perl__setlocale_debug_string(const int category, /* category number,
1969 like LC_ALL */
1970 const char* const locale, /* locale name */
1971
1972 /* return value from setlocale() when attempting to
1973 * set 'category' to 'locale' */
1974 const char* const retval)
1975{
1976 /* Returns a pointer to a NUL-terminated string in static storage with
1977 * added text about the info passed in. This is not thread safe and will
1978 * be overwritten by the next call, so this should be used just to
fa07b8e5 1979 * formulate a string to immediately print or savepv() on. */
5d1187d1 1980
398a990f
DM
1981 /* initialise to a non-null value to keep it out of BSS and so keep
1982 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
1983 static char ret[128] = "x";
5d1187d1 1984
fa07b8e5 1985 my_strlcpy(ret, "setlocale(", sizeof(ret));
5d1187d1
KW
1986
1987 switch (category) {
1988 default:
fa07b8e5 1989 my_snprintf(ret, sizeof(ret), "%s? %d", ret, category);
5d1187d1
KW
1990 break;
1991# ifdef LC_ALL
1992 case LC_ALL:
fa07b8e5 1993 my_strlcat(ret, "LC_ALL", sizeof(ret));
5d1187d1
KW
1994 break;
1995# endif
1996# ifdef LC_CTYPE
1997 case LC_CTYPE:
fa07b8e5 1998 my_strlcat(ret, "LC_CTYPE", sizeof(ret));
5d1187d1
KW
1999 break;
2000# endif
2001# ifdef LC_NUMERIC
2002 case LC_NUMERIC:
fa07b8e5 2003 my_strlcat(ret, "LC_NUMERIC", sizeof(ret));
5d1187d1
KW
2004 break;
2005# endif
2006# ifdef LC_COLLATE
2007 case LC_COLLATE:
fa07b8e5 2008 my_strlcat(ret, "LC_COLLATE", sizeof(ret));
5d1187d1
KW
2009 break;
2010# endif
2011# ifdef LC_TIME
2012 case LC_TIME:
fa07b8e5 2013 my_strlcat(ret, "LC_TIME", sizeof(ret));
5d1187d1
KW
2014 break;
2015# endif
2016# ifdef LC_MONETARY
2017 case LC_MONETARY:
fa07b8e5 2018 my_strlcat(ret, "LC_MONETARY", sizeof(ret));
5d1187d1
KW
2019 break;
2020# endif
2021# ifdef LC_MESSAGES
2022 case LC_MESSAGES:
fa07b8e5 2023 my_strlcat(ret, "LC_MESSAGES", sizeof(ret));
5d1187d1
KW
2024 break;
2025# endif
2026 }
2027
fa07b8e5 2028 my_strlcat(ret, ", ", sizeof(ret));
5d1187d1
KW
2029
2030 if (locale) {
fa07b8e5
KW
2031 my_strlcat(ret, "\"", sizeof(ret));
2032 my_strlcat(ret, locale, sizeof(ret));
2033 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2034 }
2035 else {
fa07b8e5 2036 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2037 }
2038
fa07b8e5 2039 my_strlcat(ret, ") returned ", sizeof(ret));
5d1187d1
KW
2040
2041 if (retval) {
fa07b8e5
KW
2042 my_strlcat(ret, "\"", sizeof(ret));
2043 my_strlcat(ret, retval, sizeof(ret));
2044 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2045 }
2046 else {
fa07b8e5 2047 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2048 }
2049
2050 assert(strlen(ret) < sizeof(ret));
2051
2052 return ret;
2053}
2054
2055#endif
747c467a
KW
2056
2057
2058/*
14d04a33 2059 * ex: set ts=8 sts=4 sw=4 et:
37442d52 2060 */