This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Don't use special locale collation for C locale
[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;
488 return;
489 }
490
d35fca5f 491 /* If this is not the same locale as currently, set the new one up */
98994639
HS
492 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
493 ++PL_collation_ix;
494 Safefree(PL_collation_name);
495 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 496 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
497 if (PL_collation_standard) {
498 goto is_standard_collation;
499 }
98994639
HS
500
501 {
d35fca5f
KW
502 /* A locale collation definition includes primary, secondary,
503 * tertiary, etc. weights for each character. To sort, the primary
504 * weights are used, and only if they compare equal, then the
505 * secondary weights are used, and only if they compare equal, then
506 * the tertiary, etc. strxfrm() works by taking the input string,
507 * say ABC, and creating an output string consisting of first the
508 * primary weights, A¹B¹C¹ followed by the secondary ones, A²B²C²;
509 * and then the tertiary, etc, yielding A¹B¹C¹A²B²C²A³B³C³....
510 * Some characters may not have weights at every level. In our
511 * example, let's say B doesn't have a tertiary weight, and A
512 * doesn't have a secondary weight. The constructed string is then
513 * going to be A¹B¹C¹B²C²A³C³.... This has the desired
514 * characteristics that strcmp() will look at the secondary or
515 * tertiary weights only if the strings compare equal at all higher
516 * priority weights. The length of the transformed string is
517 * roughly a linear function of the input string. It's not exactly
518 * linear because some characters don't have weights at all levels,
519 * and there are some complications, so there is often per-string
520 * overhead. When we call strxfrm() we have to allocate some
521 * memory to hold the transformed string. The calculations below
522 * try to find constants for this locale 'm' and 'b' so that m*x +
523 * b equals how much space we need given the size of the input
524 * string in 'x'. If we calculate too small, we increase the size
525 * as needed, and call strxfrm() again, but it is better to get it
526 * right the first time to avoid wasted expensive string
527 * transformations. */
98994639
HS
528 /* 2: at most so many chars ('a', 'b'). */
529 /* 50: surely no system expands a char more. */
530#define XFRMBUFSIZE (2 * 50)
531 char xbuf[XFRMBUFSIZE];
8772537c
AL
532 const Size_t fa = strxfrm(xbuf, "a", XFRMBUFSIZE);
533 const Size_t fb = strxfrm(xbuf, "ab", XFRMBUFSIZE);
534 const SSize_t mult = fb - fa;
e0601d3c 535 if (mult < 1 && !(fa == 0 && fb == 0))
ad49ad39
NC
536 Perl_croak(aTHX_ "panic: strxfrm() gets absurd - a => %"UVuf", ab => %"UVuf,
537 (UV) fa, (UV) fb);
eb160463 538 PL_collxfrm_base = (fa > (Size_t)mult) ? (fa - mult) : 0;
98994639
HS
539 PL_collxfrm_mult = mult;
540 }
541 }
542
f2ce9e1c
JH
543#else
544 PERL_UNUSED_ARG(newcoll);
98994639
HS
545#endif /* USE_LOCALE_COLLATE */
546}
547
b385bb4d
KW
548#ifdef WIN32
549
550char *
551Perl_my_setlocale(pTHX_ int category, const char* locale)
552{
553 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
554 * difference unless the input locale is "", which means on Windows to get
555 * the machine default, which is set via the computer's "Regional and
556 * Language Options" (or its current equivalent). In POSIX, it instead
557 * means to find the locale from the user's environment. This routine
558 * looks in the environment, and, if anything is found, uses that instead
559 * of going to the machine default. If there is no environment override,
560 * the machine default is used, as normal, by calling the real setlocale()
561 * with "". The POSIX behavior is to use the LC_ALL variable if set;
562 * otherwise to use the particular category's variable if set; otherwise to
563 * use the LANG variable. */
564
175c4cf9 565 bool override_LC_ALL = FALSE;
89f7b9aa
KW
566 char * result;
567
b385bb4d
KW
568 if (locale && strEQ(locale, "")) {
569# ifdef LC_ALL
570 locale = PerlEnv_getenv("LC_ALL");
571 if (! locale) {
572#endif
573 switch (category) {
574# ifdef LC_ALL
575 case LC_ALL:
481465ea 576 override_LC_ALL = TRUE;
b385bb4d
KW
577 break; /* We already know its variable isn't set */
578# endif
579# ifdef USE_LOCALE_TIME
580 case LC_TIME:
581 locale = PerlEnv_getenv("LC_TIME");
582 break;
583# endif
584# ifdef USE_LOCALE_CTYPE
585 case LC_CTYPE:
586 locale = PerlEnv_getenv("LC_CTYPE");
587 break;
588# endif
589# ifdef USE_LOCALE_COLLATE
590 case LC_COLLATE:
591 locale = PerlEnv_getenv("LC_COLLATE");
592 break;
593# endif
594# ifdef USE_LOCALE_MONETARY
595 case LC_MONETARY:
596 locale = PerlEnv_getenv("LC_MONETARY");
597 break;
598# endif
599# ifdef USE_LOCALE_NUMERIC
600 case LC_NUMERIC:
601 locale = PerlEnv_getenv("LC_NUMERIC");
602 break;
603# endif
604# ifdef USE_LOCALE_MESSAGES
605 case LC_MESSAGES:
606 locale = PerlEnv_getenv("LC_MESSAGES");
607 break;
608# endif
609 default:
610 /* This is a category, like PAPER_SIZE that we don't
611 * know about; and so can't provide a wrapper. */
612 break;
613 }
614 if (! locale) {
615 locale = PerlEnv_getenv("LANG");
481465ea 616 if (! locale) {
b385bb4d
KW
617 locale = "";
618 }
619 }
620# ifdef LC_ALL
621 }
622# endif
623 }
624
89f7b9aa 625 result = setlocale(category, locale);
bbc98134
KW
626 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
627 _setlocale_debug_string(category, locale, result)));
89f7b9aa 628
481465ea 629 if (! override_LC_ALL) {
89f7b9aa
KW
630 return result;
631 }
632
dfd77d7a 633 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
634 * LANG variable or the system default if there is no LANG. But these have
635 * lower priority than the other LC_foo variables, so override it for each
636 * one that is set. (If they are set to "", it means to use the same thing
637 * we just set LC_ALL to, so can skip) */
89f7b9aa
KW
638# ifdef USE_LOCALE_TIME
639 result = PerlEnv_getenv("LC_TIME");
730252b2 640 if (result && strNE(result, "")) {
89f7b9aa 641 setlocale(LC_TIME, result);
bbc98134
KW
642 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
643 __FILE__, __LINE__,
644 _setlocale_debug_string(LC_TIME, result, "not captured")));
89f7b9aa
KW
645 }
646# endif
647# ifdef USE_LOCALE_CTYPE
648 result = PerlEnv_getenv("LC_CTYPE");
730252b2 649 if (result && strNE(result, "")) {
89f7b9aa 650 setlocale(LC_CTYPE, result);
bbc98134
KW
651 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
652 __FILE__, __LINE__,
653 _setlocale_debug_string(LC_CTYPE, result, "not captured")));
89f7b9aa
KW
654 }
655# endif
656# ifdef USE_LOCALE_COLLATE
657 result = PerlEnv_getenv("LC_COLLATE");
730252b2 658 if (result && strNE(result, "")) {
89f7b9aa 659 setlocale(LC_COLLATE, result);
bbc98134
KW
660 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
661 __FILE__, __LINE__,
662 _setlocale_debug_string(LC_COLLATE, result, "not captured")));
89f7b9aa
KW
663 }
664# endif
665# ifdef USE_LOCALE_MONETARY
666 result = PerlEnv_getenv("LC_MONETARY");
730252b2 667 if (result && strNE(result, "")) {
89f7b9aa 668 setlocale(LC_MONETARY, result);
bbc98134
KW
669 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
670 __FILE__, __LINE__,
671 _setlocale_debug_string(LC_MONETARY, result, "not captured")));
89f7b9aa
KW
672 }
673# endif
674# ifdef USE_LOCALE_NUMERIC
675 result = PerlEnv_getenv("LC_NUMERIC");
730252b2 676 if (result && strNE(result, "")) {
89f7b9aa 677 setlocale(LC_NUMERIC, result);
bbc98134
KW
678 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
679 __FILE__, __LINE__,
680 _setlocale_debug_string(LC_NUMERIC, result, "not captured")));
89f7b9aa
KW
681 }
682# endif
683# ifdef USE_LOCALE_MESSAGES
684 result = PerlEnv_getenv("LC_MESSAGES");
730252b2 685 if (result && strNE(result, "")) {
89f7b9aa 686 setlocale(LC_MESSAGES, result);
bbc98134
KW
687 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
688 __FILE__, __LINE__,
689 _setlocale_debug_string(LC_MESSAGES, result, "not captured")));
89f7b9aa
KW
690 }
691# endif
692
bbc98134
KW
693 result = setlocale(LC_ALL, NULL);
694 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
695 __FILE__, __LINE__,
696 _setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 697
bbc98134 698 return result;
b385bb4d
KW
699}
700
701#endif
702
703
98994639
HS
704/*
705 * Initialize locale awareness.
706 */
707int
708Perl_init_i18nl10n(pTHX_ int printwarn)
709{
0e92a118
KW
710 /* printwarn is
711 *
712 * 0 if not to output warning when setup locale is bad
713 * 1 if to output warning based on value of PERL_BADLANG
714 * >1 if to output regardless of PERL_BADLANG
715 *
716 * returns
98994639 717 * 1 = set ok or not applicable,
0e92a118
KW
718 * 0 = fallback to a locale of lower priority
719 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
720 *
721 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
722 * set, debugging information is output.
723 *
724 * This looks more complicated than it is, mainly due to the #ifdefs.
725 *
726 * We try to set LC_ALL to the value determined by the environment. If
727 * there is no LC_ALL on this platform, we try the individual categories we
728 * know about. If this works, we are done.
729 *
730 * But if it doesn't work, we have to do something else. We search the
731 * environment variables ourselves instead of relying on the system to do
732 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
733 * think there is one), and the ultimate fallback "C". This is all done in
734 * the same loop as above to avoid duplicating code, but it makes things
735 * more complex. After the original failure, we add the fallback
736 * possibilities to the list of locales to try, and iterate the loop
737 * through them all until one succeeds.
738 *
739 * On Ultrix, the locale MUST come from the environment, so there is
740 * preliminary code to set it. I (khw) am not sure that it is necessary,
741 * and that this couldn't be folded into the loop, but barring any real
742 * platforms to test on, it's staying as-is
743 *
744 * A slight complication is that in embedded Perls, the locale may already
745 * be set-up, and we don't want to get it from the normal environment
746 * variables. This is handled by having a special environment variable
747 * indicate we're in this situation. We simply set setlocale's 2nd
748 * parameter to be a NULL instead of "". That indicates to setlocale that
749 * it is not to change anything, but to return the current value,
750 * effectively initializing perl's db to what the locale already is.
751 *
752 * We play the same trick with NULL if a LC_ALL succeeds. We call
753 * setlocale() on the individual categores with NULL to get their existing
754 * values for our db, instead of trying to change them.
755 * */
98994639 756
0e92a118
KW
757 int ok = 1;
758
98994639 759#if defined(USE_LOCALE)
98994639
HS
760#ifdef USE_LOCALE_CTYPE
761 char *curctype = NULL;
762#endif /* USE_LOCALE_CTYPE */
763#ifdef USE_LOCALE_COLLATE
764 char *curcoll = NULL;
765#endif /* USE_LOCALE_COLLATE */
766#ifdef USE_LOCALE_NUMERIC
767 char *curnum = NULL;
768#endif /* USE_LOCALE_NUMERIC */
769#ifdef __GLIBC__
175c4cf9 770 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
98994639 771#endif
65ebb059 772
ccd65d51
KW
773 /* NULL uses the existing already set up locale */
774 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
775 ? NULL
776 : "";
5d1187d1
KW
777#ifdef DEBUGGING
778 const bool debug = (PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT"))
779 ? TRUE
780 : FALSE;
781# define DEBUG_LOCALE_INIT(category, locale, result) \
782 STMT_START { \
783 if (debug) { \
784 PerlIO_printf(Perl_debug_log, \
785 "%s:%d: %s\n", \
786 __FILE__, __LINE__, \
787 _setlocale_debug_string(category, \
788 locale, \
789 result)); \
790 } \
791 } STMT_END
792#else
793# define DEBUG_LOCALE_INIT(a,b,c)
794#endif
c3fcd832
KW
795 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
796 unsigned int trial_locales_count;
175c4cf9
KW
797 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
798 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 799 bool setlocale_failure = FALSE;
65ebb059
KW
800 unsigned int i;
801 char *p;
175c4cf9
KW
802
803 /* A later getenv() could zap this, so only use here */
804 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
805
806 const bool locwarn = (printwarn > 1
807 || (printwarn
808 && (! bad_lang_use_once
22ff3130
HS
809 || (
810 /* disallow with "" or "0" */
811 *bad_lang_use_once
812 && strNE("0", bad_lang_use_once)))));
0e92a118 813 bool done = FALSE;
5d1187d1
KW
814 char * sl_result; /* return from setlocale() */
815 char * locale_param;
6bce99ee
JH
816#ifdef WIN32
817 /* In some systems you can find out the system default locale
818 * and use that as the fallback locale. */
819# define SYSTEM_DEFAULT_LOCALE
820#endif
821#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059 822 const char *system_default_locale = NULL;
6bce99ee 823#endif
98994639 824
0e92a118
KW
825#ifndef LOCALE_ENVIRON_REQUIRED
826 PERL_UNUSED_VAR(done);
5d1187d1 827 PERL_UNUSED_VAR(locale_param);
0e92a118 828#else
98994639
HS
829
830 /*
831 * Ultrix setlocale(..., "") fails if there are no environment
832 * variables from which to get a locale name.
833 */
834
b3e384bf 835# ifdef LC_ALL
98994639 836 if (lang) {
5d1187d1
KW
837 sl_result = my_setlocale(LC_ALL, setlocale_init);
838 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result);
839 if (sl_result)
98994639
HS
840 done = TRUE;
841 else
842 setlocale_failure = TRUE;
843 }
5d1187d1 844 if (! setlocale_failure) {
b3e384bf 845# ifdef USE_LOCALE_CTYPE
5d1187d1
KW
846 locale_param = (! done && (lang || PerlEnv_getenv("LC_CTYPE")))
847 ? setlocale_init
848 : NULL;
849 curctype = my_setlocale(LC_CTYPE, locale_param);
850 DEBUG_LOCALE_INIT(LC_CTYPE, locale_param, sl_result);
851 if (! curctype)
98994639
HS
852 setlocale_failure = TRUE;
853 else
854 curctype = savepv(curctype);
b3e384bf
KW
855# endif /* USE_LOCALE_CTYPE */
856# ifdef USE_LOCALE_COLLATE
5d1187d1
KW
857 locale_param = (! done && (lang || PerlEnv_getenv("LC_COLLATE")))
858 ? setlocale_init
859 : NULL;
860 curcoll = my_setlocale(LC_COLLATE, locale_param);
861 DEBUG_LOCALE_INIT(LC_COLLATE, locale_param, sl_result);
862 if (! curcoll)
98994639
HS
863 setlocale_failure = TRUE;
864 else
865 curcoll = savepv(curcoll);
b3e384bf
KW
866# endif /* USE_LOCALE_COLLATE */
867# ifdef USE_LOCALE_NUMERIC
5d1187d1
KW
868 locale_param = (! done && (lang || PerlEnv_getenv("LC_NUMERIC")))
869 ? setlocale_init
870 : NULL;
871 curnum = my_setlocale(LC_NUMERIC, locale_param);
872 DEBUG_LOCALE_INIT(LC_NUMERIC, locale_param, sl_result);
873 if (! curnum)
98994639
HS
874 setlocale_failure = TRUE;
875 else
876 curnum = savepv(curnum);
b3e384bf 877# endif /* USE_LOCALE_NUMERIC */
a782673d 878# ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
879 locale_param = (! done && (lang || PerlEnv_getenv("LC_MESSAGES")))
880 ? setlocale_init
881 : NULL;
882 sl_result = my_setlocale(LC_MESSAGES, locale_param);
883 DEBUG_LOCALE_INIT(LC_MESSAGES, locale_param, sl_result);
884 if (! sl_result)
a782673d
KW
885 setlocale_failure = TRUE;
886 }
887# endif /* USE_LOCALE_MESSAGES */
c835d6be 888# ifdef USE_LOCALE_MONETARY
5d1187d1
KW
889 locale_param = (! done && (lang || PerlEnv_getenv("LC_MONETARY")))
890 ? setlocale_init
891 : NULL;
892 sl_result = my_setlocale(LC_MONETARY, locale_param);
893 DEBUG_LOCALE_INIT(LC_MONETARY, locale_param, sl_result);
894 if (! sl_result) {
c835d6be
KW
895 setlocale_failure = TRUE;
896 }
897# endif /* USE_LOCALE_MONETARY */
98994639
HS
898 }
899
b3e384bf 900# endif /* LC_ALL */
98994639
HS
901
902#endif /* !LOCALE_ENVIRON_REQUIRED */
903
65ebb059 904 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
905 * the list. Normally the loop is executed just once. But if setting the
906 * locale fails, inside the loop we add fallback trials to the array and so
907 * will execute the loop multiple times */
c3fcd832
KW
908 trial_locales[0] = setlocale_init;
909 trial_locales_count = 1;
65ebb059
KW
910 for (i= 0; i < trial_locales_count; i++) {
911 const char * trial_locale = trial_locales[i];
912
913 if (i > 0) {
914
915 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
916 * when i==0, but I (khw) don't think that behavior makes much
917 * sense */
918 setlocale_failure = FALSE;
919
6bce99ee
JH
920#ifdef SYSTEM_DEFAULT_LOCALE
921# ifdef WIN32
65ebb059
KW
922 /* On Windows machines, an entry of "" after the 0th means to use
923 * the system default locale, which we now proceed to get. */
924 if (strEQ(trial_locale, "")) {
925 unsigned int j;
926
927 /* Note that this may change the locale, but we are going to do
928 * that anyway just below */
929 system_default_locale = setlocale(LC_ALL, "");
5d1187d1 930 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059
KW
931
932 /* Skip if invalid or it's already on the list of locales to
933 * try */
934 if (! system_default_locale) {
935 goto next_iteration;
936 }
937 for (j = 0; j < trial_locales_count; j++) {
938 if (strEQ(system_default_locale, trial_locales[j])) {
939 goto next_iteration;
940 }
941 }
942
943 trial_locale = system_default_locale;
944 }
6bce99ee
JH
945# endif /* WIN32 */
946#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
947 }
948
98994639 949#ifdef LC_ALL
5d1187d1
KW
950 sl_result = my_setlocale(LC_ALL, trial_locale);
951 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result);
952 if (! sl_result) {
49c85077 953 setlocale_failure = TRUE;
7cd8b568
KW
954 }
955 else {
956 /* Since LC_ALL succeeded, it should have changed all the other
957 * categories it can to its value; so we massage things so that the
958 * setlocales below just return their category's current values.
959 * This adequately handles the case in NetBSD where LC_COLLATE may
960 * not be defined for a locale, and setting it individually will
961 * fail, whereas setting LC_ALL suceeds, leaving LC_COLLATE set to
962 * the POSIX locale. */
963 trial_locale = NULL;
964 }
98994639
HS
965#endif /* LC_ALL */
966
49c85077 967 if (!setlocale_failure) {
98994639 968#ifdef USE_LOCALE_CTYPE
49c85077 969 Safefree(curctype);
5d1187d1
KW
970 curctype = my_setlocale(LC_CTYPE, trial_locale);
971 DEBUG_LOCALE_INIT(LC_CTYPE, trial_locale, curctype);
972 if (! curctype)
49c85077
KW
973 setlocale_failure = TRUE;
974 else
975 curctype = savepv(curctype);
98994639
HS
976#endif /* USE_LOCALE_CTYPE */
977#ifdef USE_LOCALE_COLLATE
49c85077 978 Safefree(curcoll);
5d1187d1
KW
979 curcoll = my_setlocale(LC_COLLATE, trial_locale);
980 DEBUG_LOCALE_INIT(LC_COLLATE, trial_locale, curcoll);
981 if (! curcoll)
49c85077
KW
982 setlocale_failure = TRUE;
983 else
984 curcoll = savepv(curcoll);
98994639
HS
985#endif /* USE_LOCALE_COLLATE */
986#ifdef USE_LOCALE_NUMERIC
49c85077 987 Safefree(curnum);
5d1187d1
KW
988 curnum = my_setlocale(LC_NUMERIC, trial_locale);
989 DEBUG_LOCALE_INIT(LC_NUMERIC, trial_locale, curnum);
990 if (! curnum)
49c85077
KW
991 setlocale_failure = TRUE;
992 else
993 curnum = savepv(curnum);
98994639 994#endif /* USE_LOCALE_NUMERIC */
a782673d 995#ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
996 sl_result = my_setlocale(LC_MESSAGES, trial_locale);
997 DEBUG_LOCALE_INIT(LC_MESSAGES, trial_locale, sl_result);
998 if (! (sl_result))
a782673d
KW
999 setlocale_failure = TRUE;
1000#endif /* USE_LOCALE_MESSAGES */
c835d6be 1001#ifdef USE_LOCALE_MONETARY
5d1187d1
KW
1002 sl_result = my_setlocale(LC_MONETARY, trial_locale);
1003 DEBUG_LOCALE_INIT(LC_MONETARY, trial_locale, sl_result);
1004 if (! (sl_result))
c835d6be
KW
1005 setlocale_failure = TRUE;
1006#endif /* USE_LOCALE_MONETARY */
1007
49c85077
KW
1008 if (! setlocale_failure) { /* Success */
1009 break;
1010 }
65ebb059 1011 }
98994639 1012
49c85077
KW
1013 /* Here, something failed; will need to try a fallback. */
1014 ok = 0;
65ebb059 1015
49c85077
KW
1016 if (i == 0) {
1017 unsigned int j;
98994639 1018
65ebb059 1019 if (locwarn) { /* Output failure info only on the first one */
98994639
HS
1020#ifdef LC_ALL
1021
49c85077
KW
1022 PerlIO_printf(Perl_error_log,
1023 "perl: warning: Setting locale failed.\n");
98994639
HS
1024
1025#else /* !LC_ALL */
1026
49c85077
KW
1027 PerlIO_printf(Perl_error_log,
1028 "perl: warning: Setting locale failed for the categories:\n\t");
20a240df 1029# ifdef USE_LOCALE_CTYPE
49c85077
KW
1030 if (! curctype)
1031 PerlIO_printf(Perl_error_log, "LC_CTYPE ");
20a240df
KW
1032# endif /* USE_LOCALE_CTYPE */
1033# ifdef USE_LOCALE_COLLATE
49c85077
KW
1034 if (! curcoll)
1035 PerlIO_printf(Perl_error_log, "LC_COLLATE ");
20a240df
KW
1036# endif /* USE_LOCALE_COLLATE */
1037# ifdef USE_LOCALE_NUMERIC
49c85077
KW
1038 if (! curnum)
1039 PerlIO_printf(Perl_error_log, "LC_NUMERIC ");
20a240df 1040# endif /* USE_LOCALE_NUMERIC */
a782673d 1041 PerlIO_printf(Perl_error_log, "and possibly others\n");
98994639
HS
1042
1043#endif /* LC_ALL */
1044
49c85077
KW
1045 PerlIO_printf(Perl_error_log,
1046 "perl: warning: Please check that your locale settings:\n");
98994639
HS
1047
1048#ifdef __GLIBC__
49c85077
KW
1049 PerlIO_printf(Perl_error_log,
1050 "\tLANGUAGE = %c%s%c,\n",
1051 language ? '"' : '(',
1052 language ? language : "unset",
1053 language ? '"' : ')');
98994639
HS
1054#endif
1055
49c85077
KW
1056 PerlIO_printf(Perl_error_log,
1057 "\tLC_ALL = %c%s%c,\n",
1058 lc_all ? '"' : '(',
1059 lc_all ? lc_all : "unset",
1060 lc_all ? '"' : ')');
98994639
HS
1061
1062#if defined(USE_ENVIRON_ARRAY)
49c85077
KW
1063 {
1064 char **e;
1065 for (e = environ; *e; e++) {
1066 if (strnEQ(*e, "LC_", 3)
1067 && strnNE(*e, "LC_ALL=", 7)
1068 && (p = strchr(*e, '=')))
1069 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
1070 (int)(p - *e), *e, p + 1);
1071 }
1072 }
98994639 1073#else
49c85077
KW
1074 PerlIO_printf(Perl_error_log,
1075 "\t(possibly more locale environment variables)\n");
98994639
HS
1076#endif
1077
49c85077
KW
1078 PerlIO_printf(Perl_error_log,
1079 "\tLANG = %c%s%c\n",
1080 lang ? '"' : '(',
1081 lang ? lang : "unset",
1082 lang ? '"' : ')');
98994639 1083
49c85077
KW
1084 PerlIO_printf(Perl_error_log,
1085 " are supported and installed on your system.\n");
1086 }
98994639 1087
65ebb059 1088 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 1089 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
1090 * usually change the upper bound of the loop we are in.
1091 *
1092 * Since the system's default way of setting the locale has not
1093 * found one that works, We use Perl's defined ordering: LC_ALL,
1094 * LANG, and the C locale. We don't try the same locale twice, so
1095 * don't add to the list if already there. (On POSIX systems, the
1096 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
1097 * but there's no harm done by doing it explicitly.
1098 *
1099 * Note that this tries the LC_ALL environment variable even on
1100 * systems which have no LC_ALL locale setting. This may or may
1101 * not have been originally intentional, but there's no real need
1102 * to change the behavior. */
65ebb059
KW
1103 if (lc_all) {
1104 for (j = 0; j < trial_locales_count; j++) {
1105 if (strEQ(lc_all, trial_locales[j])) {
1106 goto done_lc_all;
1107 }
1108 }
1109 trial_locales[trial_locales_count++] = lc_all;
1110 }
1111 done_lc_all:
98994639 1112
65ebb059
KW
1113 if (lang) {
1114 for (j = 0; j < trial_locales_count; j++) {
1115 if (strEQ(lang, trial_locales[j])) {
1116 goto done_lang;
1117 }
1118 }
1119 trial_locales[trial_locales_count++] = lang;
1120 }
1121 done_lang:
1122
1123#if defined(WIN32) && defined(LC_ALL)
1124 /* For Windows, we also try the system default locale before "C".
1125 * (If there exists a Windows without LC_ALL we skip this because
1126 * it gets too complicated. For those, the "C" is the next
1127 * fallback possibility). The "" is the same as the 0th element of
1128 * the array, but the code at the loop above knows to treat it
1129 * differently when not the 0th */
1130 trial_locales[trial_locales_count++] = "";
1131#endif
1132
1133 for (j = 0; j < trial_locales_count; j++) {
1134 if (strEQ("C", trial_locales[j])) {
1135 goto done_C;
1136 }
1137 }
1138 trial_locales[trial_locales_count++] = "C";
98994639 1139
65ebb059
KW
1140 done_C: ;
1141 } /* end of first time through the loop */
98994639 1142
65ebb059
KW
1143#ifdef WIN32
1144 next_iteration: ;
1145#endif
1146
1147 } /* end of looping through the trial locales */
1148
1149 if (ok < 1) { /* If we tried to fallback */
1150 const char* msg;
1151 if (! setlocale_failure) { /* fallback succeeded */
1152 msg = "Falling back to";
1153 }
1154 else { /* fallback failed */
98994639 1155
65ebb059
KW
1156 /* We dropped off the end of the loop, so have to decrement i to
1157 * get back to the value the last time through */
1158 i--;
98994639 1159
65ebb059
KW
1160 ok = -1;
1161 msg = "Failed to fall back to";
1162
1163 /* To continue, we should use whatever values we've got */
98994639 1164#ifdef USE_LOCALE_CTYPE
49c85077
KW
1165 Safefree(curctype);
1166 curctype = savepv(setlocale(LC_CTYPE, NULL));
5d1187d1 1167 DEBUG_LOCALE_INIT(LC_CTYPE, NULL, curctype);
98994639
HS
1168#endif /* USE_LOCALE_CTYPE */
1169#ifdef USE_LOCALE_COLLATE
49c85077
KW
1170 Safefree(curcoll);
1171 curcoll = savepv(setlocale(LC_COLLATE, NULL));
5d1187d1 1172 DEBUG_LOCALE_INIT(LC_COLLATE, NULL, curcoll);
98994639
HS
1173#endif /* USE_LOCALE_COLLATE */
1174#ifdef USE_LOCALE_NUMERIC
49c85077
KW
1175 Safefree(curnum);
1176 curnum = savepv(setlocale(LC_NUMERIC, NULL));
5d1187d1 1177 DEBUG_LOCALE_INIT(LC_NUMERIC, NULL, curnum);
98994639 1178#endif /* USE_LOCALE_NUMERIC */
65ebb059
KW
1179 }
1180
1181 if (locwarn) {
1182 const char * description;
1183 const char * name = "";
1184 if (strEQ(trial_locales[i], "C")) {
1185 description = "the standard locale";
1186 name = "C";
1187 }
6bce99ee 1188#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059
KW
1189 else if (strEQ(trial_locales[i], "")) {
1190 description = "the system default locale";
1191 if (system_default_locale) {
1192 name = system_default_locale;
1193 }
1194 }
6bce99ee 1195#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
1196 else {
1197 description = "a fallback locale";
1198 name = trial_locales[i];
1199 }
1200 if (name && strNE(name, "")) {
1201 PerlIO_printf(Perl_error_log,
1202 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
1203 }
1204 else {
1205 PerlIO_printf(Perl_error_log,
1206 "perl: warning: %s %s.\n", msg, description);
1207 }
1208 }
1209 } /* End of tried to fallback */
98994639
HS
1210
1211#ifdef USE_LOCALE_CTYPE
1212 new_ctype(curctype);
1213#endif /* USE_LOCALE_CTYPE */
1214
1215#ifdef USE_LOCALE_COLLATE
1216 new_collate(curcoll);
1217#endif /* USE_LOCALE_COLLATE */
1218
1219#ifdef USE_LOCALE_NUMERIC
1220 new_numeric(curnum);
1221#endif /* USE_LOCALE_NUMERIC */
b310b053 1222
8ef6e574 1223#if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
49c85077
KW
1224 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
1225 * locale is UTF-8. If PL_utf8locale and PL_unicode (set by -C or by
1226 * $ENV{PERL_UNICODE}) are true, perl.c:S_parse_body() will turn on the
1227 * PerlIO :utf8 layer on STDIN, STDOUT, STDERR, _and_ the default open
1228 * discipline. */
c1284011 1229 PL_utf8locale = _is_cur_LC_category_utf8(LC_CTYPE);
49c85077 1230
a05d7ebb 1231 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
1232 This is an alternative to using the -C command line switch
1233 (the -C if present will override this). */
1234 {
dd374669 1235 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 1236 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
1237 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
1238 PL_utf8cache = -1;
b310b053 1239 }
ec71e770 1240#endif
b310b053 1241
98994639 1242#ifdef USE_LOCALE_CTYPE
43c5f42d 1243 Safefree(curctype);
98994639
HS
1244#endif /* USE_LOCALE_CTYPE */
1245#ifdef USE_LOCALE_COLLATE
43c5f42d 1246 Safefree(curcoll);
98994639
HS
1247#endif /* USE_LOCALE_COLLATE */
1248#ifdef USE_LOCALE_NUMERIC
43c5f42d 1249 Safefree(curnum);
98994639 1250#endif /* USE_LOCALE_NUMERIC */
8ef6e574 1251
175c4cf9
KW
1252#ifdef __GLIBC__
1253 Safefree(language);
1254#endif
1255
1256 Safefree(lc_all);
1257 Safefree(lang);
1258
e3305790
KW
1259#else /* !USE_LOCALE */
1260 PERL_UNUSED_ARG(printwarn);
1261#endif /* USE_LOCALE */
1262
98994639
HS
1263 return ok;
1264}
1265
49c85077 1266
98994639
HS
1267#ifdef USE_LOCALE_COLLATE
1268
1269/*
1270 * mem_collxfrm() is a bit like strxfrm() but with two important
1271 * differences. First, it handles embedded NULs. Second, it allocates
1272 * a bit more memory than needed for the transformed data itself.
1273 * The real transformed data begins at offset sizeof(collationix).
d35fca5f
KW
1274 * *xlen is set to the length of that, and doesn't include the collation index
1275 * size.
98994639
HS
1276 * Please see sv_collxfrm() to see how this is used.
1277 */
166f8a29 1278
98994639
HS
1279char *
1280Perl_mem_collxfrm(pTHX_ const char *s, STRLEN len, STRLEN *xlen)
1281{
1282 char *xbuf;
1283 STRLEN xAlloc, xin, xout; /* xalloc is a reserved word in VC */
1284
7918f24d
NC
1285 PERL_ARGS_ASSERT_MEM_COLLXFRM;
1286
98994639
HS
1287 /* the first sizeof(collationix) bytes are used by sv_collxfrm(). */
1288 /* the +1 is for the terminating NUL. */
1289
1290 xAlloc = sizeof(PL_collation_ix) + PL_collxfrm_base + (PL_collxfrm_mult * len) + 1;
a02a5408 1291 Newx(xbuf, xAlloc, char);
bb0f664e 1292 if (UNLIKELY(! xbuf))
98994639
HS
1293 goto bad;
1294
d35fca5f 1295 /* Store the collation id */
98994639
HS
1296 *(U32*)xbuf = PL_collation_ix;
1297 xout = sizeof(PL_collation_ix);
d35fca5f
KW
1298
1299 /* Then the transformation of the input. We loop until successful, or we
1300 * give up */
98994639 1301 for (xin = 0; xin < len; ) {
224e8ef5 1302 Size_t xused;
98994639
HS
1303
1304 for (;;) {
1305 xused = strxfrm(xbuf + xout, s + xin, xAlloc - xout);
d35fca5f
KW
1306
1307 /* If the transformed string occupies less space than we told
1308 * strxfrm() was available, it means it successfully transformed
1309 * the whole string. */
eb160463 1310 if ((STRLEN)xused < xAlloc - xout)
98994639 1311 break;
bb0f664e
KW
1312
1313 if (UNLIKELY(xused >= PERL_INT_MAX))
1314 goto bad;
d35fca5f
KW
1315
1316 /* Otherwise it should be that the transformation stopped in the
1317 * middle because it ran out of space. Malloc more, and try again.
1318 * */
98994639
HS
1319 xAlloc = (2 * xAlloc) + 1;
1320 Renew(xbuf, xAlloc, char);
bb0f664e 1321 if (UNLIKELY(! xbuf))
98994639
HS
1322 goto bad;
1323 }
1324
1325 xin += strlen(s + xin) + 1;
1326 xout += xused;
1327
1328 /* Embedded NULs are understood but silently skipped
1329 * because they make no sense in locale collation. */
1330 }
1331
1332 xbuf[xout] = '\0';
1333 *xlen = xout - sizeof(PL_collation_ix);
1334 return xbuf;
1335
1336 bad:
1337 Safefree(xbuf);
1338 *xlen = 0;
1339 return NULL;
1340}
1341
1342#endif /* USE_LOCALE_COLLATE */
1343
8ef6e574
KW
1344#ifdef USE_LOCALE
1345
c1284011
KW
1346bool
1347Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
1348{
1349 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
1350 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 1351 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
1352 * could give the wrong result. The result will very likely be correct for
1353 * languages that have commonly used non-ASCII characters, but for notably
1354 * English, it comes down to if the locale's name ends in something like
1355 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61
KW
1356
1357 char *save_input_locale = NULL;
7d74bb61
KW
1358 STRLEN final_pos;
1359
8ef6e574 1360#ifdef LC_ALL
7d74bb61 1361 assert(category != LC_ALL);
8ef6e574 1362#endif
7d74bb61
KW
1363
1364 /* First dispose of the trivial cases */
b07fffd1 1365 save_input_locale = setlocale(category, NULL);
7d74bb61 1366 if (! save_input_locale) {
69014004
KW
1367 DEBUG_L(PerlIO_printf(Perl_debug_log,
1368 "Could not find current locale for category %d\n",
1369 category));
7d74bb61
KW
1370 return FALSE; /* XXX maybe should croak */
1371 }
b07fffd1 1372 save_input_locale = stdize_locale(savepv(save_input_locale));
a39edc4c 1373 if (isNAME_C_OR_POSIX(save_input_locale)) {
69014004
KW
1374 DEBUG_L(PerlIO_printf(Perl_debug_log,
1375 "Current locale for category %d is %s\n",
1376 category, save_input_locale));
b07fffd1 1377 Safefree(save_input_locale);
7d74bb61
KW
1378 return FALSE;
1379 }
1380
1d958db2
KW
1381#if defined(USE_LOCALE_CTYPE) \
1382 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 1383
1d958db2 1384 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
1385
1386 char *save_ctype_locale = NULL;
119ee68b 1387 bool is_utf8;
7d74bb61 1388
119ee68b 1389 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
1390
1391 /* Get the current LC_CTYPE locale */
4f72bb37 1392 save_ctype_locale = setlocale(LC_CTYPE, NULL);
7d74bb61 1393 if (! save_ctype_locale) {
69014004
KW
1394 DEBUG_L(PerlIO_printf(Perl_debug_log,
1395 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
1396 goto cant_use_nllanginfo;
1397 }
4f72bb37 1398 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
1399
1400 /* If LC_CTYPE and the desired category use the same locale, this
1401 * means that finding the value for LC_CTYPE is the same as finding
1402 * the value for the desired category. Otherwise, switch LC_CTYPE
1403 * to the desired category's locale */
1404 if (strEQ(save_ctype_locale, save_input_locale)) {
1405 Safefree(save_ctype_locale);
1406 save_ctype_locale = NULL;
1407 }
1408 else if (! setlocale(LC_CTYPE, save_input_locale)) {
69014004
KW
1409 DEBUG_L(PerlIO_printf(Perl_debug_log,
1410 "Could not change LC_CTYPE locale to %s\n",
1411 save_input_locale));
7d74bb61
KW
1412 Safefree(save_ctype_locale);
1413 goto cant_use_nllanginfo;
1414 }
1415 }
1416
69014004
KW
1417 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
1418 save_input_locale));
1419
7d74bb61 1420 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
1421 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
1422 * should give the correct results */
119ee68b 1423
1d958db2
KW
1424# if defined(HAS_NL_LANGINFO) && defined(CODESET)
1425 {
4f72bb37 1426 char *codeset = nl_langinfo(CODESET);
1d958db2 1427 if (codeset && strNE(codeset, "")) {
4f72bb37 1428 codeset = savepv(codeset);
119ee68b 1429
1d958db2
KW
1430 /* If we switched LC_CTYPE, switch back */
1431 if (save_ctype_locale) {
1432 setlocale(LC_CTYPE, save_ctype_locale);
1433 Safefree(save_ctype_locale);
1434 }
1435
1436 is_utf8 = foldEQ(codeset, STR_WITH_LEN("UTF-8"))
1437 || foldEQ(codeset, STR_WITH_LEN("UTF8"));
1438
69014004
KW
1439 DEBUG_L(PerlIO_printf(Perl_debug_log,
1440 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
1441 codeset, is_utf8));
1d958db2
KW
1442 Safefree(codeset);
1443 Safefree(save_input_locale);
1444 return is_utf8;
1445 }
119ee68b
KW
1446 }
1447
1d958db2
KW
1448# endif
1449# ifdef MB_CUR_MAX
1450
1451 /* Here, either we don't have nl_langinfo, or it didn't return a
1452 * codeset. Try MB_CUR_MAX */
1453
119ee68b
KW
1454 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
1455 * Unicode code point. Since UTF-8 is the only non-single byte
1456 * encoding we handle, we just say any such encoding is UTF-8, and if
1457 * turns out to be wrong, other things will fail */
1458 is_utf8 = MB_CUR_MAX >= 4;
1459
69014004
KW
1460 DEBUG_L(PerlIO_printf(Perl_debug_log,
1461 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
1462 (int) MB_CUR_MAX, is_utf8));
1463
119ee68b
KW
1464 Safefree(save_input_locale);
1465
1466# ifdef HAS_MBTOWC
1467
1468 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
1469 * since they are both in the C99 standard. We can feed a known byte
1470 * string to the latter function, and check that it gives the expected
1471 * result */
1472 if (is_utf8) {
1473 wchar_t wc;
856b881c 1474 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 1475 errno = 0;
f019f68f 1476 if ((size_t)mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8))
119ee68b
KW
1477 != strlen(HYPHEN_UTF8)
1478 || wc != (wchar_t) 0x2010)
1479 {
1480 is_utf8 = FALSE;
abdcbdb8 1481 DEBUG_L(PerlIO_printf(Perl_debug_log, "\thyphen=U+%x\n", (unsigned int)wc));
69014004
KW
1482 DEBUG_L(PerlIO_printf(Perl_debug_log,
1483 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
1484 mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8)), errno));
119ee68b
KW
1485 }
1486 }
119ee68b
KW
1487# endif
1488
1d958db2
KW
1489 /* If we switched LC_CTYPE, switch back */
1490 if (save_ctype_locale) {
1491 setlocale(LC_CTYPE, save_ctype_locale);
1492 Safefree(save_ctype_locale);
119ee68b 1493 }
7d74bb61 1494
1d958db2 1495 return is_utf8;
119ee68b 1496# endif
7d74bb61 1497 }
119ee68b 1498
7d74bb61
KW
1499 cant_use_nllanginfo:
1500
0080c90a
KW
1501#else /* nl_langinfo should work if available, so don't bother compiling this
1502 fallback code. The final fallback of looking at the name is
1503 compiled, and will be executed if nl_langinfo fails */
7d74bb61 1504
97f4de96
KW
1505 /* nl_langinfo not available or failed somehow. Next try looking at the
1506 * currency symbol to see if it disambiguates things. Often that will be
1507 * in the native script, and if the symbol isn't in UTF-8, we know that the
1508 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
1509 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
1510 * */
fa9b773e
KW
1511
1512#ifdef HAS_LOCALECONV
fa9b773e 1513# ifdef USE_LOCALE_MONETARY
fa9b773e
KW
1514 {
1515 char *save_monetary_locale = NULL;
fa9b773e 1516 bool only_ascii = FALSE;
13542a67
KW
1517 bool is_utf8 = FALSE;
1518 struct lconv* lc;
fa9b773e 1519
97f4de96
KW
1520 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
1521 * the desired category, if it isn't that locale already */
1522
fa9b773e
KW
1523 if (category != LC_MONETARY) {
1524
4f72bb37 1525 save_monetary_locale = setlocale(LC_MONETARY, NULL);
fa9b773e 1526 if (! save_monetary_locale) {
69014004
KW
1527 DEBUG_L(PerlIO_printf(Perl_debug_log,
1528 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
1529 goto cant_use_monetary;
1530 }
4f72bb37 1531 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 1532
13542a67
KW
1533 if (strEQ(save_monetary_locale, save_input_locale)) {
1534 Safefree(save_monetary_locale);
1535 save_monetary_locale = NULL;
1536 }
1537 else if (! setlocale(LC_MONETARY, save_input_locale)) {
59c234b4
KW
1538 DEBUG_L(PerlIO_printf(Perl_debug_log,
1539 "Could not change LC_MONETARY locale to %s\n",
1540 save_input_locale));
1541 Safefree(save_monetary_locale);
1542 goto cant_use_monetary;
fa9b773e
KW
1543 }
1544 }
1545
1546 /* Here the current LC_MONETARY is set to the locale of the category
1547 * whose information is desired. */
1548
13542a67
KW
1549 lc = localeconv();
1550 if (! lc
1551 || ! lc->currency_symbol
9f10db87 1552 || is_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
1553 {
1554 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));
1555 only_ascii = TRUE;
1556 }
1557 else {
1558 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
1559 }
1560
1561 /* If we changed it, restore LC_MONETARY to its original locale */
1562 if (save_monetary_locale) {
1563 setlocale(LC_MONETARY, save_monetary_locale);
1564 Safefree(save_monetary_locale);
1565 }
1566
13542a67 1567 if (! only_ascii) {
fa9b773e 1568
59c234b4
KW
1569 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
1570 * otherwise assume the locale is UTF-8 if and only if the symbol
1571 * is non-ascii UTF-8. */
1572 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
1573 save_input_locale, is_utf8));
1574 Safefree(save_input_locale);
1575 return is_utf8;
13542a67 1576 }
fa9b773e
KW
1577 }
1578 cant_use_monetary:
1579
1580# endif /* USE_LOCALE_MONETARY */
1581#endif /* HAS_LOCALECONV */
1582
15f7e74e
KW
1583#if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
1584
1585/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
1586 * the names of the months and weekdays, timezone, and am/pm indicator */
1587 {
1588 char *save_time_locale = NULL;
1589 int hour = 10;
1590 bool is_dst = FALSE;
1591 int dom = 1;
1592 int month = 0;
1593 int i;
1594 char * formatted_time;
1595
1596
1597 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
1598 * desired category, if it isn't that locale already */
1599
1600 if (category != LC_TIME) {
1601
1602 save_time_locale = setlocale(LC_TIME, NULL);
1603 if (! save_time_locale) {
1604 DEBUG_L(PerlIO_printf(Perl_debug_log,
1605 "Could not find current locale for LC_TIME\n"));
1606 goto cant_use_time;
1607 }
1608 save_time_locale = stdize_locale(savepv(save_time_locale));
1609
1610 if (strEQ(save_time_locale, save_input_locale)) {
1611 Safefree(save_time_locale);
1612 save_time_locale = NULL;
1613 }
1614 else if (! setlocale(LC_TIME, save_input_locale)) {
1615 DEBUG_L(PerlIO_printf(Perl_debug_log,
1616 "Could not change LC_TIME locale to %s\n",
1617 save_input_locale));
1618 Safefree(save_time_locale);
1619 goto cant_use_time;
1620 }
1621 }
1622
1623 /* Here the current LC_TIME is set to the locale of the category
1624 * whose information is desired. Look at all the days of the week and
9f10db87 1625 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
1626 * characters. The first such a one found will tell us if the locale
1627 * is UTF-8 or not */
1628
1629 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
1630 formatted_time = my_strftime("%A %B %Z %p",
1631 0, 0, hour, dom, month, 112, 0, 0, is_dst);
9f10db87 1632 if (! formatted_time || is_invariant_string((U8 *) formatted_time, 0)) {
15f7e74e
KW
1633
1634 /* Here, we didn't find a non-ASCII. Try the next time through
1635 * with the complemented dst and am/pm, and try with the next
1636 * weekday. After we have gotten all weekdays, try the next
1637 * month */
1638 is_dst = ! is_dst;
1639 hour = (hour + 12) % 24;
1640 dom++;
1641 if (i > 6) {
1642 month++;
1643 }
1644 continue;
1645 }
1646
1647 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
1648 * false otherwise. But first, restore LC_TIME to its original
1649 * locale if we changed it */
1650 if (save_time_locale) {
1651 setlocale(LC_TIME, save_time_locale);
1652 Safefree(save_time_locale);
1653 }
1654
1655 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
1656 save_input_locale,
1657 is_utf8_string((U8 *) formatted_time, 0)));
1658 Safefree(save_input_locale);
1659 return is_utf8_string((U8 *) formatted_time, 0);
1660 }
1661
1662 /* Falling off the end of the loop indicates all the names were just
1663 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
1664 * to its original locale */
1665 if (save_time_locale) {
1666 setlocale(LC_TIME, save_time_locale);
1667 Safefree(save_time_locale);
1668 }
1669 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));
1670 }
1671 cant_use_time:
1672
1673#endif
1674
1675#if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
1676
1677/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
1678 * on our dromedary test machine, which has over 700 locales. There, this
1679 * added no value to looking at the currency symbol and the time strings. I
1680 * left it in so as to avoid rewriting it if real-world experience indicates
1681 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
1682 * haven't found illegal utf8, we continue on and examine all the strerror()
1683 * messages on the platform for utf8ness. If all are ASCII, we still don't
1684 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
1685 * utf8ness. The reason this doesn't help much is that the messages may not
1686 * have been translated into the locale. The currency symbol and time strings
1687 * are much more likely to have been translated. */
1688 {
855aeb93 1689 int e;
5857e934
KW
1690 bool is_utf8 = FALSE;
1691 bool non_ascii = FALSE;
855aeb93 1692 char *save_messages_locale = NULL;
5857e934 1693 const char * errmsg = NULL;
855aeb93 1694
5857e934
KW
1695 /* Like above, we set LC_MESSAGES to the locale of the desired
1696 * category, if it isn't that locale already */
855aeb93
JH
1697
1698 if (category != LC_MESSAGES) {
1699
5857e934 1700 save_messages_locale = setlocale(LC_MESSAGES, NULL);
855aeb93 1701 if (! save_messages_locale) {
5857e934
KW
1702 DEBUG_L(PerlIO_printf(Perl_debug_log,
1703 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
1704 goto cant_use_messages;
1705 }
5857e934 1706 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
1707
1708 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
1709 Safefree(save_messages_locale);
1710 save_messages_locale = NULL;
855aeb93
JH
1711 }
1712 else if (! setlocale(LC_MESSAGES, save_input_locale)) {
5857e934
KW
1713 DEBUG_L(PerlIO_printf(Perl_debug_log,
1714 "Could not change LC_MESSAGES locale to %s\n",
1715 save_input_locale));
855aeb93
JH
1716 Safefree(save_messages_locale);
1717 goto cant_use_messages;
1718 }
1719 }
1720
1721 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
1722 * whose information is desired. Look through all the messages. We
1723 * can't use Strerror() here because it may expand to code that
1724 * segfaults in miniperl */
855aeb93 1725
5857e934
KW
1726 for (e = 0; e <= sys_nerr; e++) {
1727 errno = 0;
1728 errmsg = sys_errlist[e];
1729 if (errno || !errmsg) {
855aeb93
JH
1730 break;
1731 }
5857e934 1732 errmsg = savepv(errmsg);
9f10db87 1733 if (! is_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
1734 non_ascii = TRUE;
1735 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
1736 break;
855aeb93
JH
1737 }
1738 }
5857e934 1739 Safefree(errmsg);
855aeb93
JH
1740
1741 /* And, if we changed it, restore LC_MESSAGES to its original locale */
1742 if (save_messages_locale) {
1743 setlocale(LC_MESSAGES, save_messages_locale);
1744 Safefree(save_messages_locale);
1745 }
1746
5857e934
KW
1747 if (non_ascii) {
1748
1749 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
1750 * any non-ascii means it is one; otherwise we assume it isn't */
1751 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
1752 save_input_locale,
1753 is_utf8));
1754 Safefree(save_input_locale);
1755 return is_utf8;
1756 }
855aeb93 1757
5857e934 1758 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
1759 }
1760 cant_use_messages:
1761
1762#endif
fa9b773e 1763
0080c90a
KW
1764#endif /* the code that is compiled when no nl_langinfo */
1765
92c0a900
KW
1766#ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
1767 UTF-8 locale */
97f4de96
KW
1768 /* As a last resort, look at the locale name to see if it matches
1769 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
1770 * return of setlocale(), is actually defined to be opaque, so we can't
1771 * really rely on the absence of various substrings in the name to indicate
1772 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
1773 * be a UTF-8 locale. Similarly for the other common names */
1774
1775 final_pos = strlen(save_input_locale) - 1;
1776 if (final_pos >= 3) {
1777 char *name = save_input_locale;
1778
1779 /* Find next 'U' or 'u' and look from there */
1780 while ((name += strcspn(name, "Uu") + 1)
1781 <= save_input_locale + final_pos - 2)
1782 {
305b8651
KW
1783 if (!isALPHA_FOLD_NE(*name, 't')
1784 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
1785 {
1786 continue;
1787 }
1788 name += 2;
1789 if (*(name) == '-') {
1790 if ((name > save_input_locale + final_pos - 1)) {
1791 break;
1792 }
1793 name++;
1794 }
1795 if (*(name) == '8') {
97f4de96
KW
1796 DEBUG_L(PerlIO_printf(Perl_debug_log,
1797 "Locale %s ends with UTF-8 in name\n",
1798 save_input_locale));
00c54b9c 1799 Safefree(save_input_locale);
97f4de96
KW
1800 return TRUE;
1801 }
1802 }
1803 DEBUG_L(PerlIO_printf(Perl_debug_log,
1804 "Locale %s doesn't end with UTF-8 in name\n",
1805 save_input_locale));
1806 }
92c0a900 1807#endif
97f4de96
KW
1808
1809#ifdef WIN32
1810 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
1811 if (final_pos >= 4
1812 && *(save_input_locale + final_pos - 0) == '1'
1813 && *(save_input_locale + final_pos - 1) == '0'
1814 && *(save_input_locale + final_pos - 2) == '0'
1815 && *(save_input_locale + final_pos - 3) == '5'
1816 && *(save_input_locale + final_pos - 4) == '6')
1817 {
1818 DEBUG_L(PerlIO_printf(Perl_debug_log,
1819 "Locale %s ends with 10056 in name, is UTF-8 locale\n",
1820 save_input_locale));
1821 Safefree(save_input_locale);
1822 return TRUE;
1823 }
1824#endif
1825
1826 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
1827 * since we are about to return FALSE anyway, there is no point in doing
1828 * this extra work */
1829#if 0
1830 if (instr(save_input_locale, "8859")) {
1831 DEBUG_L(PerlIO_printf(Perl_debug_log,
1832 "Locale %s has 8859 in name, not UTF-8 locale\n",
1833 save_input_locale));
1834 Safefree(save_input_locale);
1835 return FALSE;
1836 }
1837#endif
1838
69014004
KW
1839 DEBUG_L(PerlIO_printf(Perl_debug_log,
1840 "Assuming locale %s is not a UTF-8 locale\n",
1841 save_input_locale));
fa9b773e 1842 Safefree(save_input_locale);
7d74bb61
KW
1843 return FALSE;
1844}
1845
8ef6e574 1846#endif
7d74bb61 1847
d6ded950
KW
1848
1849bool
1850Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
1851{
1a4f13e1 1852 dVAR;
d6ded950
KW
1853 /* Internal function which returns if we are in the scope of a pragma that
1854 * enables the locale category 'category'. 'compiling' should indicate if
1855 * this is during the compilation phase (TRUE) or not (FALSE). */
1856
1857 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
1858
1859 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
1860 if (! categories || categories == &PL_sv_placeholder) {
1861 return FALSE;
1862 }
1863
1864 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
1865 * a valid unsigned */
1866 assert(category >= -1);
1867 return cBOOL(SvUV(categories) & (1U << (category + 1)));
1868}
1869
2c6ee1a7
KW
1870char *
1871Perl_my_strerror(pTHX_ const int errnum) {
a0b53297 1872 dVAR;
2c6ee1a7
KW
1873
1874 /* Uses C locale for the error text unless within scope of 'use locale' for
1875 * LC_MESSAGES */
1876
1877#ifdef USE_LOCALE_MESSAGES
8ee4b769 1878 if (! IN_LC(LC_MESSAGES)) {
a0b53297
KW
1879 char * save_locale;
1880
1881 /* We have a critical section to prevent another thread from changing
1882 * the locale out from under us (or zapping the buffer returned from
1883 * setlocale() ) */
1884 LOCALE_LOCK;
1885
1886 save_locale = setlocale(LC_MESSAGES, NULL);
a39edc4c 1887 if (! isNAME_C_OR_POSIX(save_locale)) {
2c6ee1a7
KW
1888 char *errstr;
1889
1890 /* The next setlocale likely will zap this, so create a copy */
1891 save_locale = savepv(save_locale);
1892
1893 setlocale(LC_MESSAGES, "C");
1894
1895 /* This points to the static space in Strerror, with all its
1896 * limitations */
1897 errstr = Strerror(errnum);
1898
1899 setlocale(LC_MESSAGES, save_locale);
1900 Safefree(save_locale);
a0b53297
KW
1901
1902 LOCALE_UNLOCK;
1903
2c6ee1a7
KW
1904 return errstr;
1905 }
a0b53297
KW
1906
1907 LOCALE_UNLOCK;
2c6ee1a7
KW
1908 }
1909#endif
1910
1911 return Strerror(errnum);
1912}
1913
66610fdd 1914/*
747c467a
KW
1915
1916=head1 Locale-related functions and macros
1917
1918=for apidoc sync_locale
1919
1920Changing the program's locale should be avoided by XS code. Nevertheless,
1921certain non-Perl libraries called from XS, such as C<Gtk> do so. When this
1922happens, Perl needs to be told that the locale has changed. Use this function
1923to do so, before returning to Perl.
1924
1925=cut
1926*/
1927
1928void
1929Perl_sync_locale(pTHX)
1930{
1931
1932#ifdef USE_LOCALE_CTYPE
1933 new_ctype(setlocale(LC_CTYPE, NULL));
1934#endif /* USE_LOCALE_CTYPE */
1935
1936#ifdef USE_LOCALE_COLLATE
1937 new_collate(setlocale(LC_COLLATE, NULL));
1938#endif
1939
1940#ifdef USE_LOCALE_NUMERIC
1941 set_numeric_local(); /* Switch from "C" to underlying LC_NUMERIC */
1942 new_numeric(setlocale(LC_NUMERIC, NULL));
1943#endif /* USE_LOCALE_NUMERIC */
1944
1945}
1946
5d1187d1
KW
1947#if defined(DEBUGGING) && defined(USE_LOCALE)
1948
1949char *
1950Perl__setlocale_debug_string(const int category, /* category number,
1951 like LC_ALL */
1952 const char* const locale, /* locale name */
1953
1954 /* return value from setlocale() when attempting to
1955 * set 'category' to 'locale' */
1956 const char* const retval)
1957{
1958 /* Returns a pointer to a NUL-terminated string in static storage with
1959 * added text about the info passed in. This is not thread safe and will
1960 * be overwritten by the next call, so this should be used just to
fa07b8e5 1961 * formulate a string to immediately print or savepv() on. */
5d1187d1 1962
398a990f
DM
1963 /* initialise to a non-null value to keep it out of BSS and so keep
1964 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
1965 static char ret[128] = "x";
5d1187d1 1966
fa07b8e5 1967 my_strlcpy(ret, "setlocale(", sizeof(ret));
5d1187d1
KW
1968
1969 switch (category) {
1970 default:
fa07b8e5 1971 my_snprintf(ret, sizeof(ret), "%s? %d", ret, category);
5d1187d1
KW
1972 break;
1973# ifdef LC_ALL
1974 case LC_ALL:
fa07b8e5 1975 my_strlcat(ret, "LC_ALL", sizeof(ret));
5d1187d1
KW
1976 break;
1977# endif
1978# ifdef LC_CTYPE
1979 case LC_CTYPE:
fa07b8e5 1980 my_strlcat(ret, "LC_CTYPE", sizeof(ret));
5d1187d1
KW
1981 break;
1982# endif
1983# ifdef LC_NUMERIC
1984 case LC_NUMERIC:
fa07b8e5 1985 my_strlcat(ret, "LC_NUMERIC", sizeof(ret));
5d1187d1
KW
1986 break;
1987# endif
1988# ifdef LC_COLLATE
1989 case LC_COLLATE:
fa07b8e5 1990 my_strlcat(ret, "LC_COLLATE", sizeof(ret));
5d1187d1
KW
1991 break;
1992# endif
1993# ifdef LC_TIME
1994 case LC_TIME:
fa07b8e5 1995 my_strlcat(ret, "LC_TIME", sizeof(ret));
5d1187d1
KW
1996 break;
1997# endif
1998# ifdef LC_MONETARY
1999 case LC_MONETARY:
fa07b8e5 2000 my_strlcat(ret, "LC_MONETARY", sizeof(ret));
5d1187d1
KW
2001 break;
2002# endif
2003# ifdef LC_MESSAGES
2004 case LC_MESSAGES:
fa07b8e5 2005 my_strlcat(ret, "LC_MESSAGES", sizeof(ret));
5d1187d1
KW
2006 break;
2007# endif
2008 }
2009
fa07b8e5 2010 my_strlcat(ret, ", ", sizeof(ret));
5d1187d1
KW
2011
2012 if (locale) {
fa07b8e5
KW
2013 my_strlcat(ret, "\"", sizeof(ret));
2014 my_strlcat(ret, locale, sizeof(ret));
2015 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2016 }
2017 else {
fa07b8e5 2018 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2019 }
2020
fa07b8e5 2021 my_strlcat(ret, ") returned ", sizeof(ret));
5d1187d1
KW
2022
2023 if (retval) {
fa07b8e5
KW
2024 my_strlcat(ret, "\"", sizeof(ret));
2025 my_strlcat(ret, retval, sizeof(ret));
2026 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2027 }
2028 else {
fa07b8e5 2029 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2030 }
2031
2032 assert(strlen(ret) < sizeof(ret));
2033
2034 return ret;
2035}
2036
2037#endif
747c467a
KW
2038
2039
2040/*
14d04a33 2041 * ex: set ts=8 sts=4 sw=4 et:
37442d52 2042 */