This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Change algorithm for strxfrm() trials
[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;
6696cfa7 489 *PL_strxfrm_min_char = '\0';
98994639
HS
490 return;
491 }
492
d35fca5f 493 /* If this is not the same locale as currently, set the new one up */
98994639
HS
494 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
495 ++PL_collation_ix;
496 Safefree(PL_collation_name);
497 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 498 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
499 if (PL_collation_standard) {
500 goto is_standard_collation;
501 }
98994639 502
165a1c52 503 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
6696cfa7 504 *PL_strxfrm_min_char = '\0';
165a1c52 505
59c018b9
KW
506 /* A locale collation definition includes primary, secondary, tertiary,
507 * etc. weights for each character. To sort, the primary weights are
508 * used, and only if they compare equal, then the secondary weights are
509 * used, and only if they compare equal, then the tertiary, etc.
510 *
511 * strxfrm() works by taking the input string, say ABC, and creating an
512 * output transformed string consisting of first the primary weights,
513 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
514 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
515 * may not have weights at every level. In our example, let's say B
516 * doesn't have a tertiary weight, and A doesn't have a secondary
517 * weight. The constructed string is then going to be
518 * A¹B¹C¹ B²C² A³C³ ....
519 * This has the desired effect that strcmp() will look at the secondary
520 * or tertiary weights only if the strings compare equal at all higher
521 * priority weights. The spaces shown here, like in
522 * "A¹B¹C¹ * A²B²C² "
523 * are not just for readability. In the general case, these must
524 * actually be bytes, which we will call here 'separator weights'; and
525 * they must be smaller than any other weight value, but since these
526 * are C strings, only the terminating one can be a NUL (some
527 * implementations may include a non-NUL separator weight just before
528 * the NUL). Implementations tend to reserve 01 for the separator
529 * weights. They are needed so that a shorter string's secondary
530 * weights won't be misconstrued as primary weights of a longer string,
531 * etc. By making them smaller than any other weight, the shorter
532 * string will sort first. (Actually, if all secondary weights are
533 * smaller than all primary ones, there is no need for a separator
534 * weight between those two levels, etc.)
535 *
536 * The length of the transformed string is roughly a linear function of
537 * the input string. It's not exactly linear because some characters
538 * don't have weights at all levels. When we call strxfrm() we have to
539 * allocate some memory to hold the transformed string. The
540 * calculations below try to find coefficients 'm' and 'b' for this
541 * locale so that m*x + b equals how much space we need, given the size
542 * of the input string in 'x'. If we calculate too small, we increase
543 * the size as needed, and call strxfrm() again, but it is better to
544 * get it right the first time to avoid wasted expensive string
545 * transformations. */
546
98994639
HS
547 {
548 /* 2: at most so many chars ('a', 'b'). */
549 /* 50: surely no system expands a char more. */
550#define XFRMBUFSIZE (2 * 50)
551 char xbuf[XFRMBUFSIZE];
8772537c
AL
552 const Size_t fa = strxfrm(xbuf, "a", XFRMBUFSIZE);
553 const Size_t fb = strxfrm(xbuf, "ab", XFRMBUFSIZE);
554 const SSize_t mult = fb - fa;
e0601d3c 555 if (mult < 1 && !(fa == 0 && fb == 0))
ad49ad39
NC
556 Perl_croak(aTHX_ "panic: strxfrm() gets absurd - a => %"UVuf", ab => %"UVuf,
557 (UV) fa, (UV) fb);
eb160463 558 PL_collxfrm_base = (fa > (Size_t)mult) ? (fa - mult) : 0;
98994639
HS
559 PL_collxfrm_mult = mult;
560 }
561 }
562
f2ce9e1c
JH
563#else
564 PERL_UNUSED_ARG(newcoll);
98994639
HS
565#endif /* USE_LOCALE_COLLATE */
566}
567
b385bb4d
KW
568#ifdef WIN32
569
570char *
571Perl_my_setlocale(pTHX_ int category, const char* locale)
572{
573 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
574 * difference unless the input locale is "", which means on Windows to get
575 * the machine default, which is set via the computer's "Regional and
576 * Language Options" (or its current equivalent). In POSIX, it instead
577 * means to find the locale from the user's environment. This routine
578 * looks in the environment, and, if anything is found, uses that instead
579 * of going to the machine default. If there is no environment override,
580 * the machine default is used, as normal, by calling the real setlocale()
581 * with "". The POSIX behavior is to use the LC_ALL variable if set;
582 * otherwise to use the particular category's variable if set; otherwise to
583 * use the LANG variable. */
584
175c4cf9 585 bool override_LC_ALL = FALSE;
89f7b9aa
KW
586 char * result;
587
b385bb4d
KW
588 if (locale && strEQ(locale, "")) {
589# ifdef LC_ALL
590 locale = PerlEnv_getenv("LC_ALL");
591 if (! locale) {
592#endif
593 switch (category) {
594# ifdef LC_ALL
595 case LC_ALL:
481465ea 596 override_LC_ALL = TRUE;
b385bb4d
KW
597 break; /* We already know its variable isn't set */
598# endif
599# ifdef USE_LOCALE_TIME
600 case LC_TIME:
601 locale = PerlEnv_getenv("LC_TIME");
602 break;
603# endif
604# ifdef USE_LOCALE_CTYPE
605 case LC_CTYPE:
606 locale = PerlEnv_getenv("LC_CTYPE");
607 break;
608# endif
609# ifdef USE_LOCALE_COLLATE
610 case LC_COLLATE:
611 locale = PerlEnv_getenv("LC_COLLATE");
612 break;
613# endif
614# ifdef USE_LOCALE_MONETARY
615 case LC_MONETARY:
616 locale = PerlEnv_getenv("LC_MONETARY");
617 break;
618# endif
619# ifdef USE_LOCALE_NUMERIC
620 case LC_NUMERIC:
621 locale = PerlEnv_getenv("LC_NUMERIC");
622 break;
623# endif
624# ifdef USE_LOCALE_MESSAGES
625 case LC_MESSAGES:
626 locale = PerlEnv_getenv("LC_MESSAGES");
627 break;
628# endif
629 default:
630 /* This is a category, like PAPER_SIZE that we don't
631 * know about; and so can't provide a wrapper. */
632 break;
633 }
634 if (! locale) {
635 locale = PerlEnv_getenv("LANG");
481465ea 636 if (! locale) {
b385bb4d
KW
637 locale = "";
638 }
639 }
640# ifdef LC_ALL
641 }
642# endif
643 }
644
89f7b9aa 645 result = setlocale(category, locale);
bbc98134
KW
646 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
647 _setlocale_debug_string(category, locale, result)));
89f7b9aa 648
481465ea 649 if (! override_LC_ALL) {
89f7b9aa
KW
650 return result;
651 }
652
dfd77d7a 653 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
654 * LANG variable or the system default if there is no LANG. But these have
655 * lower priority than the other LC_foo variables, so override it for each
656 * one that is set. (If they are set to "", it means to use the same thing
657 * we just set LC_ALL to, so can skip) */
89f7b9aa
KW
658# ifdef USE_LOCALE_TIME
659 result = PerlEnv_getenv("LC_TIME");
730252b2 660 if (result && strNE(result, "")) {
89f7b9aa 661 setlocale(LC_TIME, result);
bbc98134
KW
662 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
663 __FILE__, __LINE__,
664 _setlocale_debug_string(LC_TIME, result, "not captured")));
89f7b9aa
KW
665 }
666# endif
667# ifdef USE_LOCALE_CTYPE
668 result = PerlEnv_getenv("LC_CTYPE");
730252b2 669 if (result && strNE(result, "")) {
89f7b9aa 670 setlocale(LC_CTYPE, result);
bbc98134
KW
671 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
672 __FILE__, __LINE__,
673 _setlocale_debug_string(LC_CTYPE, result, "not captured")));
89f7b9aa
KW
674 }
675# endif
676# ifdef USE_LOCALE_COLLATE
677 result = PerlEnv_getenv("LC_COLLATE");
730252b2 678 if (result && strNE(result, "")) {
89f7b9aa 679 setlocale(LC_COLLATE, result);
bbc98134
KW
680 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
681 __FILE__, __LINE__,
682 _setlocale_debug_string(LC_COLLATE, result, "not captured")));
89f7b9aa
KW
683 }
684# endif
685# ifdef USE_LOCALE_MONETARY
686 result = PerlEnv_getenv("LC_MONETARY");
730252b2 687 if (result && strNE(result, "")) {
89f7b9aa 688 setlocale(LC_MONETARY, result);
bbc98134
KW
689 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
690 __FILE__, __LINE__,
691 _setlocale_debug_string(LC_MONETARY, result, "not captured")));
89f7b9aa
KW
692 }
693# endif
694# ifdef USE_LOCALE_NUMERIC
695 result = PerlEnv_getenv("LC_NUMERIC");
730252b2 696 if (result && strNE(result, "")) {
89f7b9aa 697 setlocale(LC_NUMERIC, result);
bbc98134
KW
698 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
699 __FILE__, __LINE__,
700 _setlocale_debug_string(LC_NUMERIC, result, "not captured")));
89f7b9aa
KW
701 }
702# endif
703# ifdef USE_LOCALE_MESSAGES
704 result = PerlEnv_getenv("LC_MESSAGES");
730252b2 705 if (result && strNE(result, "")) {
89f7b9aa 706 setlocale(LC_MESSAGES, result);
bbc98134
KW
707 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
708 __FILE__, __LINE__,
709 _setlocale_debug_string(LC_MESSAGES, result, "not captured")));
89f7b9aa
KW
710 }
711# endif
712
bbc98134
KW
713 result = setlocale(LC_ALL, NULL);
714 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
715 __FILE__, __LINE__,
716 _setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 717
bbc98134 718 return result;
b385bb4d
KW
719}
720
721#endif
722
723
98994639
HS
724/*
725 * Initialize locale awareness.
726 */
727int
728Perl_init_i18nl10n(pTHX_ int printwarn)
729{
0e92a118
KW
730 /* printwarn is
731 *
732 * 0 if not to output warning when setup locale is bad
733 * 1 if to output warning based on value of PERL_BADLANG
734 * >1 if to output regardless of PERL_BADLANG
735 *
736 * returns
98994639 737 * 1 = set ok or not applicable,
0e92a118
KW
738 * 0 = fallback to a locale of lower priority
739 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
740 *
741 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
742 * set, debugging information is output.
743 *
744 * This looks more complicated than it is, mainly due to the #ifdefs.
745 *
746 * We try to set LC_ALL to the value determined by the environment. If
747 * there is no LC_ALL on this platform, we try the individual categories we
748 * know about. If this works, we are done.
749 *
750 * But if it doesn't work, we have to do something else. We search the
751 * environment variables ourselves instead of relying on the system to do
752 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
753 * think there is one), and the ultimate fallback "C". This is all done in
754 * the same loop as above to avoid duplicating code, but it makes things
755 * more complex. After the original failure, we add the fallback
756 * possibilities to the list of locales to try, and iterate the loop
757 * through them all until one succeeds.
758 *
759 * On Ultrix, the locale MUST come from the environment, so there is
760 * preliminary code to set it. I (khw) am not sure that it is necessary,
761 * and that this couldn't be folded into the loop, but barring any real
762 * platforms to test on, it's staying as-is
763 *
764 * A slight complication is that in embedded Perls, the locale may already
765 * be set-up, and we don't want to get it from the normal environment
766 * variables. This is handled by having a special environment variable
767 * indicate we're in this situation. We simply set setlocale's 2nd
768 * parameter to be a NULL instead of "". That indicates to setlocale that
769 * it is not to change anything, but to return the current value,
770 * effectively initializing perl's db to what the locale already is.
771 *
772 * We play the same trick with NULL if a LC_ALL succeeds. We call
773 * setlocale() on the individual categores with NULL to get their existing
774 * values for our db, instead of trying to change them.
775 * */
98994639 776
0e92a118
KW
777 int ok = 1;
778
98994639 779#if defined(USE_LOCALE)
98994639
HS
780#ifdef USE_LOCALE_CTYPE
781 char *curctype = NULL;
782#endif /* USE_LOCALE_CTYPE */
783#ifdef USE_LOCALE_COLLATE
784 char *curcoll = NULL;
785#endif /* USE_LOCALE_COLLATE */
786#ifdef USE_LOCALE_NUMERIC
787 char *curnum = NULL;
788#endif /* USE_LOCALE_NUMERIC */
789#ifdef __GLIBC__
175c4cf9 790 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
98994639 791#endif
65ebb059 792
ccd65d51
KW
793 /* NULL uses the existing already set up locale */
794 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
795 ? NULL
796 : "";
5d1187d1
KW
797#ifdef DEBUGGING
798 const bool debug = (PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT"))
799 ? TRUE
800 : FALSE;
801# define DEBUG_LOCALE_INIT(category, locale, result) \
802 STMT_START { \
803 if (debug) { \
804 PerlIO_printf(Perl_debug_log, \
805 "%s:%d: %s\n", \
806 __FILE__, __LINE__, \
807 _setlocale_debug_string(category, \
808 locale, \
809 result)); \
810 } \
811 } STMT_END
812#else
813# define DEBUG_LOCALE_INIT(a,b,c)
814#endif
c3fcd832
KW
815 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
816 unsigned int trial_locales_count;
175c4cf9
KW
817 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
818 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 819 bool setlocale_failure = FALSE;
65ebb059
KW
820 unsigned int i;
821 char *p;
175c4cf9
KW
822
823 /* A later getenv() could zap this, so only use here */
824 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
825
826 const bool locwarn = (printwarn > 1
827 || (printwarn
828 && (! bad_lang_use_once
22ff3130
HS
829 || (
830 /* disallow with "" or "0" */
831 *bad_lang_use_once
832 && strNE("0", bad_lang_use_once)))));
0e92a118 833 bool done = FALSE;
5d1187d1
KW
834 char * sl_result; /* return from setlocale() */
835 char * locale_param;
6bce99ee
JH
836#ifdef WIN32
837 /* In some systems you can find out the system default locale
838 * and use that as the fallback locale. */
839# define SYSTEM_DEFAULT_LOCALE
840#endif
841#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059 842 const char *system_default_locale = NULL;
6bce99ee 843#endif
98994639 844
0e92a118
KW
845#ifndef LOCALE_ENVIRON_REQUIRED
846 PERL_UNUSED_VAR(done);
5d1187d1 847 PERL_UNUSED_VAR(locale_param);
0e92a118 848#else
98994639
HS
849
850 /*
851 * Ultrix setlocale(..., "") fails if there are no environment
852 * variables from which to get a locale name.
853 */
854
b3e384bf 855# ifdef LC_ALL
98994639 856 if (lang) {
5d1187d1
KW
857 sl_result = my_setlocale(LC_ALL, setlocale_init);
858 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result);
859 if (sl_result)
98994639
HS
860 done = TRUE;
861 else
862 setlocale_failure = TRUE;
863 }
5d1187d1 864 if (! setlocale_failure) {
b3e384bf 865# ifdef USE_LOCALE_CTYPE
5d1187d1
KW
866 locale_param = (! done && (lang || PerlEnv_getenv("LC_CTYPE")))
867 ? setlocale_init
868 : NULL;
869 curctype = my_setlocale(LC_CTYPE, locale_param);
870 DEBUG_LOCALE_INIT(LC_CTYPE, locale_param, sl_result);
871 if (! curctype)
98994639
HS
872 setlocale_failure = TRUE;
873 else
874 curctype = savepv(curctype);
b3e384bf
KW
875# endif /* USE_LOCALE_CTYPE */
876# ifdef USE_LOCALE_COLLATE
5d1187d1
KW
877 locale_param = (! done && (lang || PerlEnv_getenv("LC_COLLATE")))
878 ? setlocale_init
879 : NULL;
880 curcoll = my_setlocale(LC_COLLATE, locale_param);
881 DEBUG_LOCALE_INIT(LC_COLLATE, locale_param, sl_result);
882 if (! curcoll)
98994639
HS
883 setlocale_failure = TRUE;
884 else
885 curcoll = savepv(curcoll);
b3e384bf
KW
886# endif /* USE_LOCALE_COLLATE */
887# ifdef USE_LOCALE_NUMERIC
5d1187d1
KW
888 locale_param = (! done && (lang || PerlEnv_getenv("LC_NUMERIC")))
889 ? setlocale_init
890 : NULL;
891 curnum = my_setlocale(LC_NUMERIC, locale_param);
892 DEBUG_LOCALE_INIT(LC_NUMERIC, locale_param, sl_result);
893 if (! curnum)
98994639
HS
894 setlocale_failure = TRUE;
895 else
896 curnum = savepv(curnum);
b3e384bf 897# endif /* USE_LOCALE_NUMERIC */
a782673d 898# ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
899 locale_param = (! done && (lang || PerlEnv_getenv("LC_MESSAGES")))
900 ? setlocale_init
901 : NULL;
902 sl_result = my_setlocale(LC_MESSAGES, locale_param);
903 DEBUG_LOCALE_INIT(LC_MESSAGES, locale_param, sl_result);
904 if (! sl_result)
a782673d
KW
905 setlocale_failure = TRUE;
906 }
907# endif /* USE_LOCALE_MESSAGES */
c835d6be 908# ifdef USE_LOCALE_MONETARY
5d1187d1
KW
909 locale_param = (! done && (lang || PerlEnv_getenv("LC_MONETARY")))
910 ? setlocale_init
911 : NULL;
912 sl_result = my_setlocale(LC_MONETARY, locale_param);
913 DEBUG_LOCALE_INIT(LC_MONETARY, locale_param, sl_result);
914 if (! sl_result) {
c835d6be
KW
915 setlocale_failure = TRUE;
916 }
917# endif /* USE_LOCALE_MONETARY */
98994639
HS
918 }
919
b3e384bf 920# endif /* LC_ALL */
98994639
HS
921
922#endif /* !LOCALE_ENVIRON_REQUIRED */
923
65ebb059 924 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
925 * the list. Normally the loop is executed just once. But if setting the
926 * locale fails, inside the loop we add fallback trials to the array and so
927 * will execute the loop multiple times */
c3fcd832
KW
928 trial_locales[0] = setlocale_init;
929 trial_locales_count = 1;
65ebb059
KW
930 for (i= 0; i < trial_locales_count; i++) {
931 const char * trial_locale = trial_locales[i];
932
933 if (i > 0) {
934
935 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
936 * when i==0, but I (khw) don't think that behavior makes much
937 * sense */
938 setlocale_failure = FALSE;
939
6bce99ee
JH
940#ifdef SYSTEM_DEFAULT_LOCALE
941# ifdef WIN32
65ebb059
KW
942 /* On Windows machines, an entry of "" after the 0th means to use
943 * the system default locale, which we now proceed to get. */
944 if (strEQ(trial_locale, "")) {
945 unsigned int j;
946
947 /* Note that this may change the locale, but we are going to do
948 * that anyway just below */
949 system_default_locale = setlocale(LC_ALL, "");
5d1187d1 950 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059
KW
951
952 /* Skip if invalid or it's already on the list of locales to
953 * try */
954 if (! system_default_locale) {
955 goto next_iteration;
956 }
957 for (j = 0; j < trial_locales_count; j++) {
958 if (strEQ(system_default_locale, trial_locales[j])) {
959 goto next_iteration;
960 }
961 }
962
963 trial_locale = system_default_locale;
964 }
6bce99ee
JH
965# endif /* WIN32 */
966#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
967 }
968
98994639 969#ifdef LC_ALL
5d1187d1
KW
970 sl_result = my_setlocale(LC_ALL, trial_locale);
971 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result);
972 if (! sl_result) {
49c85077 973 setlocale_failure = TRUE;
7cd8b568
KW
974 }
975 else {
976 /* Since LC_ALL succeeded, it should have changed all the other
977 * categories it can to its value; so we massage things so that the
978 * setlocales below just return their category's current values.
979 * This adequately handles the case in NetBSD where LC_COLLATE may
980 * not be defined for a locale, and setting it individually will
981 * fail, whereas setting LC_ALL suceeds, leaving LC_COLLATE set to
982 * the POSIX locale. */
983 trial_locale = NULL;
984 }
98994639
HS
985#endif /* LC_ALL */
986
49c85077 987 if (!setlocale_failure) {
98994639 988#ifdef USE_LOCALE_CTYPE
49c85077 989 Safefree(curctype);
5d1187d1
KW
990 curctype = my_setlocale(LC_CTYPE, trial_locale);
991 DEBUG_LOCALE_INIT(LC_CTYPE, trial_locale, curctype);
992 if (! curctype)
49c85077
KW
993 setlocale_failure = TRUE;
994 else
995 curctype = savepv(curctype);
98994639
HS
996#endif /* USE_LOCALE_CTYPE */
997#ifdef USE_LOCALE_COLLATE
49c85077 998 Safefree(curcoll);
5d1187d1
KW
999 curcoll = my_setlocale(LC_COLLATE, trial_locale);
1000 DEBUG_LOCALE_INIT(LC_COLLATE, trial_locale, curcoll);
1001 if (! curcoll)
49c85077
KW
1002 setlocale_failure = TRUE;
1003 else
1004 curcoll = savepv(curcoll);
98994639
HS
1005#endif /* USE_LOCALE_COLLATE */
1006#ifdef USE_LOCALE_NUMERIC
49c85077 1007 Safefree(curnum);
5d1187d1
KW
1008 curnum = my_setlocale(LC_NUMERIC, trial_locale);
1009 DEBUG_LOCALE_INIT(LC_NUMERIC, trial_locale, curnum);
1010 if (! curnum)
49c85077
KW
1011 setlocale_failure = TRUE;
1012 else
1013 curnum = savepv(curnum);
98994639 1014#endif /* USE_LOCALE_NUMERIC */
a782673d 1015#ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
1016 sl_result = my_setlocale(LC_MESSAGES, trial_locale);
1017 DEBUG_LOCALE_INIT(LC_MESSAGES, trial_locale, sl_result);
1018 if (! (sl_result))
a782673d
KW
1019 setlocale_failure = TRUE;
1020#endif /* USE_LOCALE_MESSAGES */
c835d6be 1021#ifdef USE_LOCALE_MONETARY
5d1187d1
KW
1022 sl_result = my_setlocale(LC_MONETARY, trial_locale);
1023 DEBUG_LOCALE_INIT(LC_MONETARY, trial_locale, sl_result);
1024 if (! (sl_result))
c835d6be
KW
1025 setlocale_failure = TRUE;
1026#endif /* USE_LOCALE_MONETARY */
1027
49c85077
KW
1028 if (! setlocale_failure) { /* Success */
1029 break;
1030 }
65ebb059 1031 }
98994639 1032
49c85077
KW
1033 /* Here, something failed; will need to try a fallback. */
1034 ok = 0;
65ebb059 1035
49c85077
KW
1036 if (i == 0) {
1037 unsigned int j;
98994639 1038
65ebb059 1039 if (locwarn) { /* Output failure info only on the first one */
98994639
HS
1040#ifdef LC_ALL
1041
49c85077
KW
1042 PerlIO_printf(Perl_error_log,
1043 "perl: warning: Setting locale failed.\n");
98994639
HS
1044
1045#else /* !LC_ALL */
1046
49c85077
KW
1047 PerlIO_printf(Perl_error_log,
1048 "perl: warning: Setting locale failed for the categories:\n\t");
20a240df 1049# ifdef USE_LOCALE_CTYPE
49c85077
KW
1050 if (! curctype)
1051 PerlIO_printf(Perl_error_log, "LC_CTYPE ");
20a240df
KW
1052# endif /* USE_LOCALE_CTYPE */
1053# ifdef USE_LOCALE_COLLATE
49c85077
KW
1054 if (! curcoll)
1055 PerlIO_printf(Perl_error_log, "LC_COLLATE ");
20a240df
KW
1056# endif /* USE_LOCALE_COLLATE */
1057# ifdef USE_LOCALE_NUMERIC
49c85077
KW
1058 if (! curnum)
1059 PerlIO_printf(Perl_error_log, "LC_NUMERIC ");
20a240df 1060# endif /* USE_LOCALE_NUMERIC */
a782673d 1061 PerlIO_printf(Perl_error_log, "and possibly others\n");
98994639
HS
1062
1063#endif /* LC_ALL */
1064
49c85077
KW
1065 PerlIO_printf(Perl_error_log,
1066 "perl: warning: Please check that your locale settings:\n");
98994639
HS
1067
1068#ifdef __GLIBC__
49c85077
KW
1069 PerlIO_printf(Perl_error_log,
1070 "\tLANGUAGE = %c%s%c,\n",
1071 language ? '"' : '(',
1072 language ? language : "unset",
1073 language ? '"' : ')');
98994639
HS
1074#endif
1075
49c85077
KW
1076 PerlIO_printf(Perl_error_log,
1077 "\tLC_ALL = %c%s%c,\n",
1078 lc_all ? '"' : '(',
1079 lc_all ? lc_all : "unset",
1080 lc_all ? '"' : ')');
98994639
HS
1081
1082#if defined(USE_ENVIRON_ARRAY)
49c85077
KW
1083 {
1084 char **e;
1085 for (e = environ; *e; e++) {
1086 if (strnEQ(*e, "LC_", 3)
1087 && strnNE(*e, "LC_ALL=", 7)
1088 && (p = strchr(*e, '=')))
1089 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
1090 (int)(p - *e), *e, p + 1);
1091 }
1092 }
98994639 1093#else
49c85077
KW
1094 PerlIO_printf(Perl_error_log,
1095 "\t(possibly more locale environment variables)\n");
98994639
HS
1096#endif
1097
49c85077
KW
1098 PerlIO_printf(Perl_error_log,
1099 "\tLANG = %c%s%c\n",
1100 lang ? '"' : '(',
1101 lang ? lang : "unset",
1102 lang ? '"' : ')');
98994639 1103
49c85077
KW
1104 PerlIO_printf(Perl_error_log,
1105 " are supported and installed on your system.\n");
1106 }
98994639 1107
65ebb059 1108 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 1109 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
1110 * usually change the upper bound of the loop we are in.
1111 *
1112 * Since the system's default way of setting the locale has not
1113 * found one that works, We use Perl's defined ordering: LC_ALL,
1114 * LANG, and the C locale. We don't try the same locale twice, so
1115 * don't add to the list if already there. (On POSIX systems, the
1116 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
1117 * but there's no harm done by doing it explicitly.
1118 *
1119 * Note that this tries the LC_ALL environment variable even on
1120 * systems which have no LC_ALL locale setting. This may or may
1121 * not have been originally intentional, but there's no real need
1122 * to change the behavior. */
65ebb059
KW
1123 if (lc_all) {
1124 for (j = 0; j < trial_locales_count; j++) {
1125 if (strEQ(lc_all, trial_locales[j])) {
1126 goto done_lc_all;
1127 }
1128 }
1129 trial_locales[trial_locales_count++] = lc_all;
1130 }
1131 done_lc_all:
98994639 1132
65ebb059
KW
1133 if (lang) {
1134 for (j = 0; j < trial_locales_count; j++) {
1135 if (strEQ(lang, trial_locales[j])) {
1136 goto done_lang;
1137 }
1138 }
1139 trial_locales[trial_locales_count++] = lang;
1140 }
1141 done_lang:
1142
1143#if defined(WIN32) && defined(LC_ALL)
1144 /* For Windows, we also try the system default locale before "C".
1145 * (If there exists a Windows without LC_ALL we skip this because
1146 * it gets too complicated. For those, the "C" is the next
1147 * fallback possibility). The "" is the same as the 0th element of
1148 * the array, but the code at the loop above knows to treat it
1149 * differently when not the 0th */
1150 trial_locales[trial_locales_count++] = "";
1151#endif
1152
1153 for (j = 0; j < trial_locales_count; j++) {
1154 if (strEQ("C", trial_locales[j])) {
1155 goto done_C;
1156 }
1157 }
1158 trial_locales[trial_locales_count++] = "C";
98994639 1159
65ebb059
KW
1160 done_C: ;
1161 } /* end of first time through the loop */
98994639 1162
65ebb059
KW
1163#ifdef WIN32
1164 next_iteration: ;
1165#endif
1166
1167 } /* end of looping through the trial locales */
1168
1169 if (ok < 1) { /* If we tried to fallback */
1170 const char* msg;
1171 if (! setlocale_failure) { /* fallback succeeded */
1172 msg = "Falling back to";
1173 }
1174 else { /* fallback failed */
98994639 1175
65ebb059
KW
1176 /* We dropped off the end of the loop, so have to decrement i to
1177 * get back to the value the last time through */
1178 i--;
98994639 1179
65ebb059
KW
1180 ok = -1;
1181 msg = "Failed to fall back to";
1182
1183 /* To continue, we should use whatever values we've got */
98994639 1184#ifdef USE_LOCALE_CTYPE
49c85077
KW
1185 Safefree(curctype);
1186 curctype = savepv(setlocale(LC_CTYPE, NULL));
5d1187d1 1187 DEBUG_LOCALE_INIT(LC_CTYPE, NULL, curctype);
98994639
HS
1188#endif /* USE_LOCALE_CTYPE */
1189#ifdef USE_LOCALE_COLLATE
49c85077
KW
1190 Safefree(curcoll);
1191 curcoll = savepv(setlocale(LC_COLLATE, NULL));
5d1187d1 1192 DEBUG_LOCALE_INIT(LC_COLLATE, NULL, curcoll);
98994639
HS
1193#endif /* USE_LOCALE_COLLATE */
1194#ifdef USE_LOCALE_NUMERIC
49c85077
KW
1195 Safefree(curnum);
1196 curnum = savepv(setlocale(LC_NUMERIC, NULL));
5d1187d1 1197 DEBUG_LOCALE_INIT(LC_NUMERIC, NULL, curnum);
98994639 1198#endif /* USE_LOCALE_NUMERIC */
65ebb059
KW
1199 }
1200
1201 if (locwarn) {
1202 const char * description;
1203 const char * name = "";
1204 if (strEQ(trial_locales[i], "C")) {
1205 description = "the standard locale";
1206 name = "C";
1207 }
6bce99ee 1208#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059
KW
1209 else if (strEQ(trial_locales[i], "")) {
1210 description = "the system default locale";
1211 if (system_default_locale) {
1212 name = system_default_locale;
1213 }
1214 }
6bce99ee 1215#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
1216 else {
1217 description = "a fallback locale";
1218 name = trial_locales[i];
1219 }
1220 if (name && strNE(name, "")) {
1221 PerlIO_printf(Perl_error_log,
1222 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
1223 }
1224 else {
1225 PerlIO_printf(Perl_error_log,
1226 "perl: warning: %s %s.\n", msg, description);
1227 }
1228 }
1229 } /* End of tried to fallback */
98994639
HS
1230
1231#ifdef USE_LOCALE_CTYPE
1232 new_ctype(curctype);
1233#endif /* USE_LOCALE_CTYPE */
1234
1235#ifdef USE_LOCALE_COLLATE
1236 new_collate(curcoll);
1237#endif /* USE_LOCALE_COLLATE */
1238
1239#ifdef USE_LOCALE_NUMERIC
1240 new_numeric(curnum);
1241#endif /* USE_LOCALE_NUMERIC */
b310b053 1242
8ef6e574 1243#if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
49c85077
KW
1244 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
1245 * locale is UTF-8. If PL_utf8locale and PL_unicode (set by -C or by
1246 * $ENV{PERL_UNICODE}) are true, perl.c:S_parse_body() will turn on the
1247 * PerlIO :utf8 layer on STDIN, STDOUT, STDERR, _and_ the default open
1248 * discipline. */
c1284011 1249 PL_utf8locale = _is_cur_LC_category_utf8(LC_CTYPE);
49c85077 1250
a05d7ebb 1251 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
1252 This is an alternative to using the -C command line switch
1253 (the -C if present will override this). */
1254 {
dd374669 1255 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 1256 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
1257 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
1258 PL_utf8cache = -1;
b310b053 1259 }
ec71e770 1260#endif
b310b053 1261
98994639 1262#ifdef USE_LOCALE_CTYPE
43c5f42d 1263 Safefree(curctype);
98994639
HS
1264#endif /* USE_LOCALE_CTYPE */
1265#ifdef USE_LOCALE_COLLATE
43c5f42d 1266 Safefree(curcoll);
98994639
HS
1267#endif /* USE_LOCALE_COLLATE */
1268#ifdef USE_LOCALE_NUMERIC
43c5f42d 1269 Safefree(curnum);
98994639 1270#endif /* USE_LOCALE_NUMERIC */
8ef6e574 1271
175c4cf9
KW
1272#ifdef __GLIBC__
1273 Safefree(language);
1274#endif
1275
1276 Safefree(lc_all);
1277 Safefree(lang);
1278
e3305790
KW
1279#else /* !USE_LOCALE */
1280 PERL_UNUSED_ARG(printwarn);
1281#endif /* USE_LOCALE */
1282
98994639
HS
1283 return ok;
1284}
1285
49c85077 1286
98994639
HS
1287#ifdef USE_LOCALE_COLLATE
1288
1289/*
1290 * mem_collxfrm() is a bit like strxfrm() but with two important
1291 * differences. First, it handles embedded NULs. Second, it allocates
1292 * a bit more memory than needed for the transformed data itself.
1293 * The real transformed data begins at offset sizeof(collationix).
d35fca5f
KW
1294 * *xlen is set to the length of that, and doesn't include the collation index
1295 * size.
98994639
HS
1296 * Please see sv_collxfrm() to see how this is used.
1297 */
166f8a29 1298
98994639 1299char *
6696cfa7
KW
1300Perl_mem_collxfrm(pTHX_ const char *input_string,
1301 STRLEN len,
1302 STRLEN *xlen
1303 )
98994639 1304{
6696cfa7
KW
1305 char * s = (char *) input_string;
1306 STRLEN s_strlen = strlen(input_string);
98994639 1307 char *xbuf;
6696cfa7 1308 STRLEN xAlloc, xout; /* xalloc is a reserved word in VC */
c664130f 1309 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 1310
7918f24d
NC
1311 PERL_ARGS_ASSERT_MEM_COLLXFRM;
1312
6696cfa7
KW
1313 /* Replace any embedded NULs with the control that sorts before any others.
1314 * This will give as good as possible results on strings that don't
1315 * otherwise contain that character, but otherwise there may be
1316 * less-than-perfect results with that character and NUL. This is
1317 * unavoidable unless we replace strxfrm with our own implementation.
1318 *
1319 * XXX This code may be overkill. khw wrote it before realizing that if
1320 * you change a NUL into some other character, that that may change the
1321 * strxfrm results if that character is part of a sequence with other
1322 * characters for weight calculations. To minimize the chances of this,
1323 * now the replacement is restricted to another control (likely to be
1324 * \001). But the full generality has been retained.
1325 *
1326 * This is one of the few places in the perl core, where we can use
1327 * standard functions like strlen() and strcat(). It's because we're
1328 * looking for NULs. */
1329 if (s_strlen < len) {
1330 char * e = s + len;
1331 char * sans_nuls;
1332 STRLEN cur_min_char_len;
1333
1334 /* If we don't know what control character sorts lowest for this
1335 * locale, find it */
1336 if (*PL_strxfrm_min_char == '\0') {
1337 int j;
1338 char * cur_min_x = NULL; /* Cur cp's xfrm, (except it also
1339 includes the collation index
1340 prefixed. */
1341
1342 /* Look through all legal code points (NUL isn't) */
1343 for (j = 1; j < 256; j++) {
1344 char * x; /* j's xfrm plus collation index */
1345 STRLEN x_len; /* length of 'x' */
1346 STRLEN trial_len = 1;
1347
1348 /* Create a 1 byte string of the current code point, but with
1349 * room to be 2 bytes */
1350 char cur_source[] = { (char) j, '\0' , '\0' };
1351
1352 if (PL_in_utf8_COLLATE_locale) {
1353 if (! isCNTRL_L1(j)) {
1354 continue;
1355 }
1356
1357 /* If needs to be 2 bytes, find them */
1358 if (! UVCHR_IS_INVARIANT(j)) {
1359 continue; /* Can't handle variants yet */
1360 }
1361 }
1362 else if (! isCNTRL_LC(j)) {
1363 continue;
1364 }
1365
1366 /* Then transform it */
1367 x = mem_collxfrm(cur_source, trial_len, &x_len);
1368
1369 /* If something went wrong (which it shouldn't), just
1370 * ignore this code point */
1371 if ( x_len == 0
1372 || strlen(x + sizeof(PL_collation_ix)) < x_len)
1373 {
1374 continue;
1375 }
1376
1377 /* If this character's transformation is lower than
1378 * the current lowest, this one becomes the lowest */
1379 if ( cur_min_x == NULL
1380 || strLT(x + sizeof(PL_collation_ix),
1381 cur_min_x + sizeof(PL_collation_ix)))
1382 {
1383 strcpy(PL_strxfrm_min_char, cur_source);
1384 cur_min_x = x;
1385 }
1386 else {
1387 Safefree(x);
1388 }
1389 } /* end of loop through all bytes */
1390
1391 /* Unlikely, but possible, if there aren't any controls in the
1392 * locale, arbitrarily use \001 */
1393 if (cur_min_x == NULL) {
1394 STRLEN x_len; /* temporary */
1395 cur_min_x = mem_collxfrm("\001", 1, &x_len);
1396 /* cur_min_cp was already initialized to 1 */
1397 }
1398
1399 Safefree(cur_min_x);
1400 }
1401
1402 /* The worst case length for the replaced string would be if every
1403 * character in it is NUL. Multiply that by the length of each
1404 * replacement, and allow for a trailing NUL */
1405 cur_min_char_len = strlen(PL_strxfrm_min_char);
1406 Newx(sans_nuls, (len * cur_min_char_len) + 1, char);
1407 *sans_nuls = '\0';
1408
1409
1410 /* Replace each NUL with the lowest collating control. Loop until have
1411 * exhausted all the NULs */
1412 while (s + s_strlen < e) {
1413 strcat(sans_nuls, s);
1414
1415 /* Do the actual replacement */
1416 strcat(sans_nuls, PL_strxfrm_min_char);
1417
1418 /* Move past the input NUL */
1419 s += s_strlen + 1;
1420 s_strlen = strlen(s);
1421 }
1422
1423 /* And add anything that trails the final NUL */
1424 strcat(sans_nuls, s);
1425
1426 /* Switch so below we transform this modified string */
1427 s = sans_nuls;
1428 len = strlen(s);
1429 }
1430
59c018b9
KW
1431 /* The first element in the output is the collation id, used by
1432 * sv_collxfrm(); then comes the space for the transformed string. The
1433 * equation should give us a good estimate as to how much is needed */
98994639 1434 xAlloc = sizeof(PL_collation_ix) + PL_collxfrm_base + (PL_collxfrm_mult * len) + 1;
a02a5408 1435 Newx(xbuf, xAlloc, char);
bb0f664e 1436 if (UNLIKELY(! xbuf))
98994639
HS
1437 goto bad;
1438
d35fca5f 1439 /* Store the collation id */
98994639
HS
1440 *(U32*)xbuf = PL_collation_ix;
1441 xout = sizeof(PL_collation_ix);
d35fca5f
KW
1442
1443 /* Then the transformation of the input. We loop until successful, or we
1444 * give up */
4ebeff16
KW
1445 for (;;) {
1446 STRLEN xused = strxfrm(xbuf + xout, s, xAlloc - xout);
1447
1448 /* If the transformed string occupies less space than we told strxfrm()
1449 * was available, it means it successfully transformed the whole
1450 * string. */
1451 if (xused < xAlloc - xout) {
1452 xout += xused;
1453 break;
1454 }
bb0f664e 1455
4ebeff16
KW
1456 if (UNLIKELY(xused >= PERL_INT_MAX))
1457 goto bad;
d35fca5f 1458
c664130f
KW
1459 /* A well-behaved strxfrm() returns exactly how much space it needs
1460 * (not including the trailing NUL) when it fails due to not enough
1461 * space being provided. Assume that this is the case unless it's been
1462 * proven otherwise */
1463 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
1464 xAlloc = xused + sizeof(PL_collation_ix) + 1;
1465 }
1466 else { /* Here, either:
1467 * 1) The strxfrm() has previously shown bad behavior; or
1468 * 2) It isn't the first time through the loop, which means
1469 * that the strxfrm() is now showing bad behavior, because
1470 * we gave it what it said was needed in the previous
1471 * iteration, and it came back saying it needed still more.
1472 * (Many versions of cygwin fit this. When the buffer size
1473 * isn't sufficient, they return the input size instead of
1474 * how much is needed.)
1475 * Increase the buffer size by a fixed percentage and try again. */
1476 xAlloc = (2 * xAlloc) + 1;
1477 PL_strxfrm_is_behaved = FALSE;
1478 }
1479
1480
4ebeff16
KW
1481 Renew(xbuf, xAlloc, char);
1482 if (UNLIKELY(! xbuf))
1483 goto bad;
c664130f
KW
1484
1485 first_time = FALSE;
4ebeff16 1486 }
98994639 1487
6696cfa7
KW
1488 *xlen = xout - sizeof(PL_collation_ix);
1489
3c5f993e
KW
1490 /* Free up unneeded space; retain ehough for trailing NUL */
1491 Renew(xbuf, xout + 1, char);
98994639 1492
6696cfa7
KW
1493 if (s != input_string) {
1494 Safefree(s);
98994639
HS
1495 }
1496
98994639
HS
1497 return xbuf;
1498
1499 bad:
1500 Safefree(xbuf);
6696cfa7
KW
1501 if (s != input_string) {
1502 Safefree(s);
1503 }
98994639
HS
1504 *xlen = 0;
1505 return NULL;
1506}
1507
1508#endif /* USE_LOCALE_COLLATE */
1509
8ef6e574
KW
1510#ifdef USE_LOCALE
1511
c1284011
KW
1512bool
1513Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
1514{
1515 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
1516 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 1517 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
1518 * could give the wrong result. The result will very likely be correct for
1519 * languages that have commonly used non-ASCII characters, but for notably
1520 * English, it comes down to if the locale's name ends in something like
1521 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61
KW
1522
1523 char *save_input_locale = NULL;
7d74bb61
KW
1524 STRLEN final_pos;
1525
8ef6e574 1526#ifdef LC_ALL
7d74bb61 1527 assert(category != LC_ALL);
8ef6e574 1528#endif
7d74bb61
KW
1529
1530 /* First dispose of the trivial cases */
b07fffd1 1531 save_input_locale = setlocale(category, NULL);
7d74bb61 1532 if (! save_input_locale) {
69014004
KW
1533 DEBUG_L(PerlIO_printf(Perl_debug_log,
1534 "Could not find current locale for category %d\n",
1535 category));
7d74bb61
KW
1536 return FALSE; /* XXX maybe should croak */
1537 }
b07fffd1 1538 save_input_locale = stdize_locale(savepv(save_input_locale));
a39edc4c 1539 if (isNAME_C_OR_POSIX(save_input_locale)) {
69014004
KW
1540 DEBUG_L(PerlIO_printf(Perl_debug_log,
1541 "Current locale for category %d is %s\n",
1542 category, save_input_locale));
b07fffd1 1543 Safefree(save_input_locale);
7d74bb61
KW
1544 return FALSE;
1545 }
1546
1d958db2
KW
1547#if defined(USE_LOCALE_CTYPE) \
1548 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 1549
1d958db2 1550 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
1551
1552 char *save_ctype_locale = NULL;
119ee68b 1553 bool is_utf8;
7d74bb61 1554
119ee68b 1555 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
1556
1557 /* Get the current LC_CTYPE locale */
4f72bb37 1558 save_ctype_locale = setlocale(LC_CTYPE, NULL);
7d74bb61 1559 if (! save_ctype_locale) {
69014004
KW
1560 DEBUG_L(PerlIO_printf(Perl_debug_log,
1561 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
1562 goto cant_use_nllanginfo;
1563 }
4f72bb37 1564 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
1565
1566 /* If LC_CTYPE and the desired category use the same locale, this
1567 * means that finding the value for LC_CTYPE is the same as finding
1568 * the value for the desired category. Otherwise, switch LC_CTYPE
1569 * to the desired category's locale */
1570 if (strEQ(save_ctype_locale, save_input_locale)) {
1571 Safefree(save_ctype_locale);
1572 save_ctype_locale = NULL;
1573 }
1574 else if (! setlocale(LC_CTYPE, save_input_locale)) {
69014004
KW
1575 DEBUG_L(PerlIO_printf(Perl_debug_log,
1576 "Could not change LC_CTYPE locale to %s\n",
1577 save_input_locale));
7d74bb61
KW
1578 Safefree(save_ctype_locale);
1579 goto cant_use_nllanginfo;
1580 }
1581 }
1582
69014004
KW
1583 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
1584 save_input_locale));
1585
7d74bb61 1586 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
1587 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
1588 * should give the correct results */
119ee68b 1589
1d958db2
KW
1590# if defined(HAS_NL_LANGINFO) && defined(CODESET)
1591 {
4f72bb37 1592 char *codeset = nl_langinfo(CODESET);
1d958db2 1593 if (codeset && strNE(codeset, "")) {
4f72bb37 1594 codeset = savepv(codeset);
119ee68b 1595
1d958db2
KW
1596 /* If we switched LC_CTYPE, switch back */
1597 if (save_ctype_locale) {
1598 setlocale(LC_CTYPE, save_ctype_locale);
1599 Safefree(save_ctype_locale);
1600 }
1601
1602 is_utf8 = foldEQ(codeset, STR_WITH_LEN("UTF-8"))
1603 || foldEQ(codeset, STR_WITH_LEN("UTF8"));
1604
69014004
KW
1605 DEBUG_L(PerlIO_printf(Perl_debug_log,
1606 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
1607 codeset, is_utf8));
1d958db2
KW
1608 Safefree(codeset);
1609 Safefree(save_input_locale);
1610 return is_utf8;
1611 }
119ee68b
KW
1612 }
1613
1d958db2
KW
1614# endif
1615# ifdef MB_CUR_MAX
1616
1617 /* Here, either we don't have nl_langinfo, or it didn't return a
1618 * codeset. Try MB_CUR_MAX */
1619
119ee68b
KW
1620 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
1621 * Unicode code point. Since UTF-8 is the only non-single byte
1622 * encoding we handle, we just say any such encoding is UTF-8, and if
1623 * turns out to be wrong, other things will fail */
1624 is_utf8 = MB_CUR_MAX >= 4;
1625
69014004
KW
1626 DEBUG_L(PerlIO_printf(Perl_debug_log,
1627 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
1628 (int) MB_CUR_MAX, is_utf8));
1629
119ee68b
KW
1630 Safefree(save_input_locale);
1631
1632# ifdef HAS_MBTOWC
1633
1634 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
1635 * since they are both in the C99 standard. We can feed a known byte
1636 * string to the latter function, and check that it gives the expected
1637 * result */
1638 if (is_utf8) {
1639 wchar_t wc;
856b881c 1640 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 1641 errno = 0;
f019f68f 1642 if ((size_t)mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8))
119ee68b
KW
1643 != strlen(HYPHEN_UTF8)
1644 || wc != (wchar_t) 0x2010)
1645 {
1646 is_utf8 = FALSE;
abdcbdb8 1647 DEBUG_L(PerlIO_printf(Perl_debug_log, "\thyphen=U+%x\n", (unsigned int)wc));
69014004
KW
1648 DEBUG_L(PerlIO_printf(Perl_debug_log,
1649 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
1650 mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8)), errno));
119ee68b
KW
1651 }
1652 }
119ee68b
KW
1653# endif
1654
1d958db2
KW
1655 /* If we switched LC_CTYPE, switch back */
1656 if (save_ctype_locale) {
1657 setlocale(LC_CTYPE, save_ctype_locale);
1658 Safefree(save_ctype_locale);
119ee68b 1659 }
7d74bb61 1660
1d958db2 1661 return is_utf8;
119ee68b 1662# endif
7d74bb61 1663 }
119ee68b 1664
7d74bb61
KW
1665 cant_use_nllanginfo:
1666
0080c90a
KW
1667#else /* nl_langinfo should work if available, so don't bother compiling this
1668 fallback code. The final fallback of looking at the name is
1669 compiled, and will be executed if nl_langinfo fails */
7d74bb61 1670
97f4de96
KW
1671 /* nl_langinfo not available or failed somehow. Next try looking at the
1672 * currency symbol to see if it disambiguates things. Often that will be
1673 * in the native script, and if the symbol isn't in UTF-8, we know that the
1674 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
1675 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
1676 * */
fa9b773e
KW
1677
1678#ifdef HAS_LOCALECONV
fa9b773e 1679# ifdef USE_LOCALE_MONETARY
fa9b773e
KW
1680 {
1681 char *save_monetary_locale = NULL;
fa9b773e 1682 bool only_ascii = FALSE;
13542a67
KW
1683 bool is_utf8 = FALSE;
1684 struct lconv* lc;
fa9b773e 1685
97f4de96
KW
1686 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
1687 * the desired category, if it isn't that locale already */
1688
fa9b773e
KW
1689 if (category != LC_MONETARY) {
1690
4f72bb37 1691 save_monetary_locale = setlocale(LC_MONETARY, NULL);
fa9b773e 1692 if (! save_monetary_locale) {
69014004
KW
1693 DEBUG_L(PerlIO_printf(Perl_debug_log,
1694 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
1695 goto cant_use_monetary;
1696 }
4f72bb37 1697 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 1698
13542a67
KW
1699 if (strEQ(save_monetary_locale, save_input_locale)) {
1700 Safefree(save_monetary_locale);
1701 save_monetary_locale = NULL;
1702 }
1703 else if (! setlocale(LC_MONETARY, save_input_locale)) {
59c234b4
KW
1704 DEBUG_L(PerlIO_printf(Perl_debug_log,
1705 "Could not change LC_MONETARY locale to %s\n",
1706 save_input_locale));
1707 Safefree(save_monetary_locale);
1708 goto cant_use_monetary;
fa9b773e
KW
1709 }
1710 }
1711
1712 /* Here the current LC_MONETARY is set to the locale of the category
1713 * whose information is desired. */
1714
13542a67
KW
1715 lc = localeconv();
1716 if (! lc
1717 || ! lc->currency_symbol
9f10db87 1718 || is_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
1719 {
1720 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));
1721 only_ascii = TRUE;
1722 }
1723 else {
1724 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
1725 }
1726
1727 /* If we changed it, restore LC_MONETARY to its original locale */
1728 if (save_monetary_locale) {
1729 setlocale(LC_MONETARY, save_monetary_locale);
1730 Safefree(save_monetary_locale);
1731 }
1732
13542a67 1733 if (! only_ascii) {
fa9b773e 1734
59c234b4
KW
1735 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
1736 * otherwise assume the locale is UTF-8 if and only if the symbol
1737 * is non-ascii UTF-8. */
1738 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
1739 save_input_locale, is_utf8));
1740 Safefree(save_input_locale);
1741 return is_utf8;
13542a67 1742 }
fa9b773e
KW
1743 }
1744 cant_use_monetary:
1745
1746# endif /* USE_LOCALE_MONETARY */
1747#endif /* HAS_LOCALECONV */
1748
15f7e74e
KW
1749#if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
1750
1751/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
1752 * the names of the months and weekdays, timezone, and am/pm indicator */
1753 {
1754 char *save_time_locale = NULL;
1755 int hour = 10;
1756 bool is_dst = FALSE;
1757 int dom = 1;
1758 int month = 0;
1759 int i;
1760 char * formatted_time;
1761
1762
1763 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
1764 * desired category, if it isn't that locale already */
1765
1766 if (category != LC_TIME) {
1767
1768 save_time_locale = setlocale(LC_TIME, NULL);
1769 if (! save_time_locale) {
1770 DEBUG_L(PerlIO_printf(Perl_debug_log,
1771 "Could not find current locale for LC_TIME\n"));
1772 goto cant_use_time;
1773 }
1774 save_time_locale = stdize_locale(savepv(save_time_locale));
1775
1776 if (strEQ(save_time_locale, save_input_locale)) {
1777 Safefree(save_time_locale);
1778 save_time_locale = NULL;
1779 }
1780 else if (! setlocale(LC_TIME, save_input_locale)) {
1781 DEBUG_L(PerlIO_printf(Perl_debug_log,
1782 "Could not change LC_TIME locale to %s\n",
1783 save_input_locale));
1784 Safefree(save_time_locale);
1785 goto cant_use_time;
1786 }
1787 }
1788
1789 /* Here the current LC_TIME is set to the locale of the category
1790 * whose information is desired. Look at all the days of the week and
9f10db87 1791 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
1792 * characters. The first such a one found will tell us if the locale
1793 * is UTF-8 or not */
1794
1795 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
1796 formatted_time = my_strftime("%A %B %Z %p",
1797 0, 0, hour, dom, month, 112, 0, 0, is_dst);
9f10db87 1798 if (! formatted_time || is_invariant_string((U8 *) formatted_time, 0)) {
15f7e74e
KW
1799
1800 /* Here, we didn't find a non-ASCII. Try the next time through
1801 * with the complemented dst and am/pm, and try with the next
1802 * weekday. After we have gotten all weekdays, try the next
1803 * month */
1804 is_dst = ! is_dst;
1805 hour = (hour + 12) % 24;
1806 dom++;
1807 if (i > 6) {
1808 month++;
1809 }
1810 continue;
1811 }
1812
1813 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
1814 * false otherwise. But first, restore LC_TIME to its original
1815 * locale if we changed it */
1816 if (save_time_locale) {
1817 setlocale(LC_TIME, save_time_locale);
1818 Safefree(save_time_locale);
1819 }
1820
1821 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
1822 save_input_locale,
1823 is_utf8_string((U8 *) formatted_time, 0)));
1824 Safefree(save_input_locale);
1825 return is_utf8_string((U8 *) formatted_time, 0);
1826 }
1827
1828 /* Falling off the end of the loop indicates all the names were just
1829 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
1830 * to its original locale */
1831 if (save_time_locale) {
1832 setlocale(LC_TIME, save_time_locale);
1833 Safefree(save_time_locale);
1834 }
1835 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));
1836 }
1837 cant_use_time:
1838
1839#endif
1840
1841#if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
1842
1843/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
1844 * on our dromedary test machine, which has over 700 locales. There, this
1845 * added no value to looking at the currency symbol and the time strings. I
1846 * left it in so as to avoid rewriting it if real-world experience indicates
1847 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
1848 * haven't found illegal utf8, we continue on and examine all the strerror()
1849 * messages on the platform for utf8ness. If all are ASCII, we still don't
1850 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
1851 * utf8ness. The reason this doesn't help much is that the messages may not
1852 * have been translated into the locale. The currency symbol and time strings
1853 * are much more likely to have been translated. */
1854 {
855aeb93 1855 int e;
5857e934
KW
1856 bool is_utf8 = FALSE;
1857 bool non_ascii = FALSE;
855aeb93 1858 char *save_messages_locale = NULL;
5857e934 1859 const char * errmsg = NULL;
855aeb93 1860
5857e934
KW
1861 /* Like above, we set LC_MESSAGES to the locale of the desired
1862 * category, if it isn't that locale already */
855aeb93
JH
1863
1864 if (category != LC_MESSAGES) {
1865
5857e934 1866 save_messages_locale = setlocale(LC_MESSAGES, NULL);
855aeb93 1867 if (! save_messages_locale) {
5857e934
KW
1868 DEBUG_L(PerlIO_printf(Perl_debug_log,
1869 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
1870 goto cant_use_messages;
1871 }
5857e934 1872 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
1873
1874 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
1875 Safefree(save_messages_locale);
1876 save_messages_locale = NULL;
855aeb93
JH
1877 }
1878 else if (! setlocale(LC_MESSAGES, save_input_locale)) {
5857e934
KW
1879 DEBUG_L(PerlIO_printf(Perl_debug_log,
1880 "Could not change LC_MESSAGES locale to %s\n",
1881 save_input_locale));
855aeb93
JH
1882 Safefree(save_messages_locale);
1883 goto cant_use_messages;
1884 }
1885 }
1886
1887 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
1888 * whose information is desired. Look through all the messages. We
1889 * can't use Strerror() here because it may expand to code that
1890 * segfaults in miniperl */
855aeb93 1891
5857e934
KW
1892 for (e = 0; e <= sys_nerr; e++) {
1893 errno = 0;
1894 errmsg = sys_errlist[e];
1895 if (errno || !errmsg) {
855aeb93
JH
1896 break;
1897 }
5857e934 1898 errmsg = savepv(errmsg);
9f10db87 1899 if (! is_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
1900 non_ascii = TRUE;
1901 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
1902 break;
855aeb93
JH
1903 }
1904 }
5857e934 1905 Safefree(errmsg);
855aeb93
JH
1906
1907 /* And, if we changed it, restore LC_MESSAGES to its original locale */
1908 if (save_messages_locale) {
1909 setlocale(LC_MESSAGES, save_messages_locale);
1910 Safefree(save_messages_locale);
1911 }
1912
5857e934
KW
1913 if (non_ascii) {
1914
1915 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
1916 * any non-ascii means it is one; otherwise we assume it isn't */
1917 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
1918 save_input_locale,
1919 is_utf8));
1920 Safefree(save_input_locale);
1921 return is_utf8;
1922 }
855aeb93 1923
5857e934 1924 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
1925 }
1926 cant_use_messages:
1927
1928#endif
fa9b773e 1929
0080c90a
KW
1930#endif /* the code that is compiled when no nl_langinfo */
1931
92c0a900
KW
1932#ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
1933 UTF-8 locale */
97f4de96
KW
1934 /* As a last resort, look at the locale name to see if it matches
1935 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
1936 * return of setlocale(), is actually defined to be opaque, so we can't
1937 * really rely on the absence of various substrings in the name to indicate
1938 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
1939 * be a UTF-8 locale. Similarly for the other common names */
1940
1941 final_pos = strlen(save_input_locale) - 1;
1942 if (final_pos >= 3) {
1943 char *name = save_input_locale;
1944
1945 /* Find next 'U' or 'u' and look from there */
1946 while ((name += strcspn(name, "Uu") + 1)
1947 <= save_input_locale + final_pos - 2)
1948 {
305b8651
KW
1949 if (!isALPHA_FOLD_NE(*name, 't')
1950 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
1951 {
1952 continue;
1953 }
1954 name += 2;
1955 if (*(name) == '-') {
1956 if ((name > save_input_locale + final_pos - 1)) {
1957 break;
1958 }
1959 name++;
1960 }
1961 if (*(name) == '8') {
97f4de96
KW
1962 DEBUG_L(PerlIO_printf(Perl_debug_log,
1963 "Locale %s ends with UTF-8 in name\n",
1964 save_input_locale));
00c54b9c 1965 Safefree(save_input_locale);
97f4de96
KW
1966 return TRUE;
1967 }
1968 }
1969 DEBUG_L(PerlIO_printf(Perl_debug_log,
1970 "Locale %s doesn't end with UTF-8 in name\n",
1971 save_input_locale));
1972 }
92c0a900 1973#endif
97f4de96
KW
1974
1975#ifdef WIN32
1976 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
1977 if (final_pos >= 4
1978 && *(save_input_locale + final_pos - 0) == '1'
1979 && *(save_input_locale + final_pos - 1) == '0'
1980 && *(save_input_locale + final_pos - 2) == '0'
1981 && *(save_input_locale + final_pos - 3) == '5'
1982 && *(save_input_locale + final_pos - 4) == '6')
1983 {
1984 DEBUG_L(PerlIO_printf(Perl_debug_log,
1985 "Locale %s ends with 10056 in name, is UTF-8 locale\n",
1986 save_input_locale));
1987 Safefree(save_input_locale);
1988 return TRUE;
1989 }
1990#endif
1991
1992 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
1993 * since we are about to return FALSE anyway, there is no point in doing
1994 * this extra work */
1995#if 0
1996 if (instr(save_input_locale, "8859")) {
1997 DEBUG_L(PerlIO_printf(Perl_debug_log,
1998 "Locale %s has 8859 in name, not UTF-8 locale\n",
1999 save_input_locale));
2000 Safefree(save_input_locale);
2001 return FALSE;
2002 }
2003#endif
2004
69014004
KW
2005 DEBUG_L(PerlIO_printf(Perl_debug_log,
2006 "Assuming locale %s is not a UTF-8 locale\n",
2007 save_input_locale));
fa9b773e 2008 Safefree(save_input_locale);
7d74bb61
KW
2009 return FALSE;
2010}
2011
8ef6e574 2012#endif
7d74bb61 2013
d6ded950
KW
2014
2015bool
2016Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
2017{
1a4f13e1 2018 dVAR;
d6ded950
KW
2019 /* Internal function which returns if we are in the scope of a pragma that
2020 * enables the locale category 'category'. 'compiling' should indicate if
2021 * this is during the compilation phase (TRUE) or not (FALSE). */
2022
2023 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
2024
2025 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
2026 if (! categories || categories == &PL_sv_placeholder) {
2027 return FALSE;
2028 }
2029
2030 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
2031 * a valid unsigned */
2032 assert(category >= -1);
2033 return cBOOL(SvUV(categories) & (1U << (category + 1)));
2034}
2035
2c6ee1a7
KW
2036char *
2037Perl_my_strerror(pTHX_ const int errnum) {
a0b53297 2038 dVAR;
2c6ee1a7
KW
2039
2040 /* Uses C locale for the error text unless within scope of 'use locale' for
2041 * LC_MESSAGES */
2042
2043#ifdef USE_LOCALE_MESSAGES
8ee4b769 2044 if (! IN_LC(LC_MESSAGES)) {
a0b53297
KW
2045 char * save_locale;
2046
2047 /* We have a critical section to prevent another thread from changing
2048 * the locale out from under us (or zapping the buffer returned from
2049 * setlocale() ) */
2050 LOCALE_LOCK;
2051
2052 save_locale = setlocale(LC_MESSAGES, NULL);
a39edc4c 2053 if (! isNAME_C_OR_POSIX(save_locale)) {
2c6ee1a7
KW
2054 char *errstr;
2055
2056 /* The next setlocale likely will zap this, so create a copy */
2057 save_locale = savepv(save_locale);
2058
2059 setlocale(LC_MESSAGES, "C");
2060
2061 /* This points to the static space in Strerror, with all its
2062 * limitations */
2063 errstr = Strerror(errnum);
2064
2065 setlocale(LC_MESSAGES, save_locale);
2066 Safefree(save_locale);
a0b53297
KW
2067
2068 LOCALE_UNLOCK;
2069
2c6ee1a7
KW
2070 return errstr;
2071 }
a0b53297
KW
2072
2073 LOCALE_UNLOCK;
2c6ee1a7
KW
2074 }
2075#endif
2076
2077 return Strerror(errnum);
2078}
2079
66610fdd 2080/*
747c467a
KW
2081
2082=head1 Locale-related functions and macros
2083
2084=for apidoc sync_locale
2085
2086Changing the program's locale should be avoided by XS code. Nevertheless,
2087certain non-Perl libraries called from XS, such as C<Gtk> do so. When this
2088happens, Perl needs to be told that the locale has changed. Use this function
2089to do so, before returning to Perl.
2090
2091=cut
2092*/
2093
2094void
2095Perl_sync_locale(pTHX)
2096{
2097
2098#ifdef USE_LOCALE_CTYPE
2099 new_ctype(setlocale(LC_CTYPE, NULL));
2100#endif /* USE_LOCALE_CTYPE */
2101
2102#ifdef USE_LOCALE_COLLATE
2103 new_collate(setlocale(LC_COLLATE, NULL));
2104#endif
2105
2106#ifdef USE_LOCALE_NUMERIC
2107 set_numeric_local(); /* Switch from "C" to underlying LC_NUMERIC */
2108 new_numeric(setlocale(LC_NUMERIC, NULL));
2109#endif /* USE_LOCALE_NUMERIC */
2110
2111}
2112
5d1187d1
KW
2113#if defined(DEBUGGING) && defined(USE_LOCALE)
2114
2115char *
2116Perl__setlocale_debug_string(const int category, /* category number,
2117 like LC_ALL */
2118 const char* const locale, /* locale name */
2119
2120 /* return value from setlocale() when attempting to
2121 * set 'category' to 'locale' */
2122 const char* const retval)
2123{
2124 /* Returns a pointer to a NUL-terminated string in static storage with
2125 * added text about the info passed in. This is not thread safe and will
2126 * be overwritten by the next call, so this should be used just to
fa07b8e5 2127 * formulate a string to immediately print or savepv() on. */
5d1187d1 2128
398a990f
DM
2129 /* initialise to a non-null value to keep it out of BSS and so keep
2130 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
2131 static char ret[128] = "x";
5d1187d1 2132
fa07b8e5 2133 my_strlcpy(ret, "setlocale(", sizeof(ret));
5d1187d1
KW
2134
2135 switch (category) {
2136 default:
fa07b8e5 2137 my_snprintf(ret, sizeof(ret), "%s? %d", ret, category);
5d1187d1
KW
2138 break;
2139# ifdef LC_ALL
2140 case LC_ALL:
fa07b8e5 2141 my_strlcat(ret, "LC_ALL", sizeof(ret));
5d1187d1
KW
2142 break;
2143# endif
2144# ifdef LC_CTYPE
2145 case LC_CTYPE:
fa07b8e5 2146 my_strlcat(ret, "LC_CTYPE", sizeof(ret));
5d1187d1
KW
2147 break;
2148# endif
2149# ifdef LC_NUMERIC
2150 case LC_NUMERIC:
fa07b8e5 2151 my_strlcat(ret, "LC_NUMERIC", sizeof(ret));
5d1187d1
KW
2152 break;
2153# endif
2154# ifdef LC_COLLATE
2155 case LC_COLLATE:
fa07b8e5 2156 my_strlcat(ret, "LC_COLLATE", sizeof(ret));
5d1187d1
KW
2157 break;
2158# endif
2159# ifdef LC_TIME
2160 case LC_TIME:
fa07b8e5 2161 my_strlcat(ret, "LC_TIME", sizeof(ret));
5d1187d1
KW
2162 break;
2163# endif
2164# ifdef LC_MONETARY
2165 case LC_MONETARY:
fa07b8e5 2166 my_strlcat(ret, "LC_MONETARY", sizeof(ret));
5d1187d1
KW
2167 break;
2168# endif
2169# ifdef LC_MESSAGES
2170 case LC_MESSAGES:
fa07b8e5 2171 my_strlcat(ret, "LC_MESSAGES", sizeof(ret));
5d1187d1
KW
2172 break;
2173# endif
2174 }
2175
fa07b8e5 2176 my_strlcat(ret, ", ", sizeof(ret));
5d1187d1
KW
2177
2178 if (locale) {
fa07b8e5
KW
2179 my_strlcat(ret, "\"", sizeof(ret));
2180 my_strlcat(ret, locale, sizeof(ret));
2181 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2182 }
2183 else {
fa07b8e5 2184 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2185 }
2186
fa07b8e5 2187 my_strlcat(ret, ") returned ", sizeof(ret));
5d1187d1
KW
2188
2189 if (retval) {
fa07b8e5
KW
2190 my_strlcat(ret, "\"", sizeof(ret));
2191 my_strlcat(ret, retval, sizeof(ret));
2192 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2193 }
2194 else {
fa07b8e5 2195 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2196 }
2197
2198 assert(strlen(ret) < sizeof(ret));
2199
2200 return ret;
2201}
2202
2203#endif
747c467a
KW
2204
2205
2206/*
14d04a33 2207 * ex: set ts=8 sts=4 sw=4 et:
37442d52 2208 */