This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
[MERGE] Rely on a C89 compiler
[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
f7416781 39#include "perl_langinfo.h"
98994639
HS
40#include "perl.h"
41
a4af207c
JH
42#include "reentr.h"
43
2fcc0ca9
KW
44/* If the environment says to, we can output debugging information during
45 * initialization. This is done before option parsing, and before any thread
46 * creation, so can be a file-level static */
47#ifdef DEBUGGING
27cdc72e
DM
48# ifdef PERL_GLOBAL_STRUCT
49 /* no global syms allowed */
50# define debug_initialization 0
51# define DEBUG_INITIALIZATION_set(v)
52# else
2fcc0ca9 53static bool debug_initialization = FALSE;
27cdc72e
DM
54# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
55# endif
2fcc0ca9
KW
56#endif
57
8ef6e574
KW
58#ifdef USE_LOCALE
59
98994639 60/*
0d071d52
KW
61 * Standardize the locale name from a string returned by 'setlocale', possibly
62 * modifying that string.
98994639 63 *
0ef2a2b2 64 * The typical return value of setlocale() is either
98994639
HS
65 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
66 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
67 * (the space-separated values represent the various sublocales,
0ef2a2b2 68 * in some unspecified order). This is not handled by this function.
98994639
HS
69 *
70 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
0ef2a2b2
KW
71 * which is harmful for further use of the string in setlocale(). This
72 * function removes the trailing new line and everything up through the '='
98994639
HS
73 *
74 */
75STATIC char *
76S_stdize_locale(pTHX_ char *locs)
77{
7452cf6a 78 const char * const s = strchr(locs, '=');
98994639
HS
79 bool okay = TRUE;
80
7918f24d
NC
81 PERL_ARGS_ASSERT_STDIZE_LOCALE;
82
8772537c
AL
83 if (s) {
84 const char * const t = strchr(s, '.');
98994639 85 okay = FALSE;
8772537c
AL
86 if (t) {
87 const char * const u = strchr(t, '\n');
88 if (u && (u[1] == 0)) {
89 const STRLEN len = u - s;
90 Move(s + 1, locs, len, char);
91 locs[len] = 0;
92 okay = TRUE;
98994639
HS
93 }
94 }
95 }
96
97 if (!okay)
98 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
99
100 return locs;
101}
102
8ef6e574
KW
103#endif
104
a4f00dcc
KW
105STATIC void
106S_set_numeric_radix(pTHX)
98994639
HS
107{
108#ifdef USE_LOCALE_NUMERIC
109# ifdef HAS_LOCALECONV
7452cf6a 110 const struct lconv* const lc = localeconv();
98994639 111
98994639
HS
112 if (lc && lc->decimal_point) {
113 if (lc->decimal_point[0] == '.' && lc->decimal_point[1] == 0) {
114 SvREFCNT_dec(PL_numeric_radix_sv);
a0714e2c 115 PL_numeric_radix_sv = NULL;
98994639
HS
116 }
117 else {
118 if (PL_numeric_radix_sv)
119 sv_setpv(PL_numeric_radix_sv, lc->decimal_point);
120 else
121 PL_numeric_radix_sv = newSVpv(lc->decimal_point, 0);
c5f058df 122 if (! is_utf8_invariant_string((U8 *) lc->decimal_point, 0)
28acfe03 123 && is_utf8_string((U8 *) lc->decimal_point, 0)
c1284011 124 && _is_cur_LC_category_utf8(LC_NUMERIC))
28acfe03
KW
125 {
126 SvUTF8_on(PL_numeric_radix_sv);
127 }
98994639
HS
128 }
129 }
130 else
a0714e2c 131 PL_numeric_radix_sv = NULL;
69014004 132
2fcc0ca9
KW
133#ifdef DEBUGGING
134 if (DEBUG_L_TEST || debug_initialization) {
135 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
69014004 136 (PL_numeric_radix_sv)
37b7e435
KW
137 ? SvPVX(PL_numeric_radix_sv)
138 : "NULL",
139 (PL_numeric_radix_sv)
39eb7305 140 ? cBOOL(SvUTF8(PL_numeric_radix_sv))
2fcc0ca9
KW
141 : 0);
142 }
143#endif
69014004 144
98994639
HS
145# endif /* HAS_LOCALECONV */
146#endif /* USE_LOCALE_NUMERIC */
147}
148
a39edc4c
KW
149/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
150 * return of setlocale(), then this is extremely likely to be the C or POSIX
151 * locale. However, the output of setlocale() is documented to be opaque, but
152 * the odds are extremely small that it would return these two strings for some
153 * other locale. Note that VMS in these two locales includes many non-ASCII
154 * characters as controls and punctuation (below are hex bytes):
155 * cntrl: 00-1F 7F 84-97 9B-9F
156 * 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
157 * Oddly, none there are listed as alphas, though some represent alphabetics
158 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
98b630b3
KW
159#define isNAME_C_OR_POSIX(name) ((name) != NULL \
160 && ((*(name) == 'C' && (*(name + 1)) == '\0') \
a39edc4c
KW
161 || strEQ((name), "POSIX")))
162
98994639 163void
8772537c 164Perl_new_numeric(pTHX_ const char *newnum)
98994639
HS
165{
166#ifdef USE_LOCALE_NUMERIC
0d071d52
KW
167
168 /* Called after all libc setlocale() calls affecting LC_NUMERIC, to tell
169 * core Perl this and that 'newnum' is the name of the new locale.
170 * It installs this locale as the current underlying default.
171 *
172 * The default locale and the C locale can be toggled between by use of the
173 * set_numeric_local() and set_numeric_standard() functions, which should
174 * probably not be called directly, but only via macros like
175 * SET_NUMERIC_STANDARD() in perl.h.
176 *
177 * The toggling is necessary mainly so that a non-dot radix decimal point
178 * character can be output, while allowing internal calculations to use a
179 * dot.
180 *
181 * This sets several interpreter-level variables:
bb304765 182 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
0d071d52 183 * PL_numeric_local A boolean indicating if the toggled state is such
7738054c
KW
184 * that the current locale is the program's underlying
185 * locale
186 * PL_numeric_standard An int indicating if the toggled state is such
187 * that the current locale is the C locale. If non-zero,
188 * it is in C; if > 1, it means it may not be toggled away
189 * from C.
0d071d52
KW
190 * Note that both of the last two variables can be true at the same time,
191 * if the underlying locale is C. (Toggling is a no-op under these
192 * circumstances.)
193 *
194 * Any code changing the locale (outside this file) should use
195 * POSIX::setlocale, which calls this function. Therefore this function
196 * should be called directly only from this file and from
197 * POSIX::setlocale() */
198
b03f34cf 199 char *save_newnum;
98994639
HS
200
201 if (! newnum) {
43c5f42d
NC
202 Safefree(PL_numeric_name);
203 PL_numeric_name = NULL;
98994639
HS
204 PL_numeric_standard = TRUE;
205 PL_numeric_local = TRUE;
206 return;
207 }
208
b03f34cf 209 save_newnum = stdize_locale(savepv(newnum));
abe1abcf
KW
210
211 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
212 PL_numeric_local = TRUE;
213
b03f34cf 214 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 215 Safefree(PL_numeric_name);
b03f34cf 216 PL_numeric_name = save_newnum;
b03f34cf 217 }
abe1abcf
KW
218 else {
219 Safefree(save_newnum);
220 }
4c28b29c
KW
221
222 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
223 * have to worry about the radix being a non-dot. (Core operations that
224 * need the underlying locale change to it temporarily). */
225 set_numeric_standard();
226
e19f01cb 227 set_numeric_radix();
6959d69d 228
f2ce9e1c
JH
229#else
230 PERL_UNUSED_ARG(newnum);
98994639
HS
231#endif /* USE_LOCALE_NUMERIC */
232}
233
234void
235Perl_set_numeric_standard(pTHX)
236{
237#ifdef USE_LOCALE_NUMERIC
28c1bf33
KW
238 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
239 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
240 * macro avoids calling this routine if toggling isn't necessary according
241 * to our records (which could be wrong if some XS code has changed the
242 * locale behind our back) */
0d071d52 243
a9b8c0d8
KW
244 setlocale(LC_NUMERIC, "C");
245 PL_numeric_standard = TRUE;
246 PL_numeric_local = isNAME_C_OR_POSIX(PL_numeric_name);
247 set_numeric_radix();
2fcc0ca9
KW
248#ifdef DEBUGGING
249 if (DEBUG_L_TEST || debug_initialization) {
250 PerlIO_printf(Perl_debug_log,
251 "Underlying LC_NUMERIC locale now is C\n");
252 }
253#endif
98994639
HS
254
255#endif /* USE_LOCALE_NUMERIC */
256}
257
258void
259Perl_set_numeric_local(pTHX)
260{
261#ifdef USE_LOCALE_NUMERIC
28c1bf33
KW
262 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
263 * code should use the macros like SET_NUMERIC_LOCAL() in perl.h instead of
264 * calling this directly. The macro avoids calling this routine if
265 * toggling isn't necessary according to our records (which could be wrong
266 * if some XS code has changed the locale behind our back) */
a9b8c0d8
KW
267
268 setlocale(LC_NUMERIC, PL_numeric_name);
269 PL_numeric_standard = isNAME_C_OR_POSIX(PL_numeric_name);
270 PL_numeric_local = TRUE;
271 set_numeric_radix();
2fcc0ca9
KW
272#ifdef DEBUGGING
273 if (DEBUG_L_TEST || debug_initialization) {
274 PerlIO_printf(Perl_debug_log,
69014004 275 "Underlying LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
276 PL_numeric_name);
277 }
278#endif
98994639
HS
279
280#endif /* USE_LOCALE_NUMERIC */
281}
282
283/*
284 * Set up for a new ctype locale.
285 */
a4f00dcc
KW
286STATIC void
287S_new_ctype(pTHX_ const char *newctype)
98994639
HS
288{
289#ifdef USE_LOCALE_CTYPE
0d071d52
KW
290
291 /* Called after all libc setlocale() calls affecting LC_CTYPE, to tell
292 * core Perl this and that 'newctype' is the name of the new locale.
293 *
294 * This function sets up the folding arrays for all 256 bytes, assuming
295 * that tofold() is tolc() since fold case is not a concept in POSIX,
296 *
297 * Any code changing the locale (outside this file) should use
298 * POSIX::setlocale, which calls this function. Therefore this function
299 * should be called directly only from this file and from
300 * POSIX::setlocale() */
301
27da23d5 302 dVAR;
68067e4e 303 UV i;
98994639 304
7918f24d
NC
305 PERL_ARGS_ASSERT_NEW_CTYPE;
306
215c5139
KW
307 /* We will replace any bad locale warning with 1) nothing if the new one is
308 * ok; or 2) a new warning for the bad new locale */
309 if (PL_warn_locale) {
310 SvREFCNT_dec_NN(PL_warn_locale);
311 PL_warn_locale = NULL;
312 }
313
c1284011 314 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
315
316 /* A UTF-8 locale gets standard rules. But note that code still has to
317 * handle this specially because of the three problematic code points */
318 if (PL_in_utf8_CTYPE_locale) {
319 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
320 }
321 else {
8c6180a9
KW
322 /* Assume enough space for every character being bad. 4 spaces each
323 * for the 94 printable characters that are output like "'x' "; and 5
324 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
325 * NUL */
326 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
327
cc9eaeb0
KW
328 /* Don't check for problems if we are suppressing the warnings */
329 bool check_for_problems = ckWARN_d(WARN_LOCALE)
330 || UNLIKELY(DEBUG_L_TEST);
8c6180a9
KW
331 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
332 to start */
333 unsigned int bad_count = 0; /* Count of bad characters */
334
baa60164
KW
335 for (i = 0; i < 256; i++) {
336 if (isUPPER_LC((U8) i))
337 PL_fold_locale[i] = (U8) toLOWER_LC((U8) i);
338 else if (isLOWER_LC((U8) i))
339 PL_fold_locale[i] = (U8) toUPPER_LC((U8) i);
340 else
341 PL_fold_locale[i] = (U8) i;
8c6180a9
KW
342
343 /* If checking for locale problems, see if the native ASCII-range
344 * printables plus \n and \t are in their expected categories in
345 * the new locale. If not, this could mean big trouble, upending
346 * Perl's and most programs' assumptions, like having a
347 * metacharacter with special meaning become a \w. Fortunately,
348 * it's very rare to find locales that aren't supersets of ASCII
349 * nowadays. It isn't a problem for most controls to be changed
350 * into something else; we check only \n and \t, though perhaps \r
351 * could be an issue as well. */
352 if (check_for_problems
353 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
354 {
355 if ((isALPHANUMERIC_A(i) && ! isALPHANUMERIC_LC(i))
356 || (isPUNCT_A(i) && ! isPUNCT_LC(i))
357 || (isBLANK_A(i) && ! isBLANK_LC(i))
358 || (i == '\n' && ! isCNTRL_LC(i)))
359 {
360 if (bad_count) { /* Separate multiple entries with a
361 blank */
362 bad_chars_list[bad_count++] = ' ';
363 }
364 bad_chars_list[bad_count++] = '\'';
365 if (isPRINT_A(i)) {
366 bad_chars_list[bad_count++] = (char) i;
367 }
368 else {
369 bad_chars_list[bad_count++] = '\\';
370 if (i == '\n') {
371 bad_chars_list[bad_count++] = 'n';
372 }
373 else {
374 assert(i == '\t');
375 bad_chars_list[bad_count++] = 't';
376 }
377 }
378 bad_chars_list[bad_count++] = '\'';
379 bad_chars_list[bad_count] = '\0';
380 }
381 }
382 }
383
384#ifdef MB_CUR_MAX
385 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 386 * this locale requires more than one byte, there are going to be
8c6180a9 387 * problems. */
9c8a6dc2
KW
388 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
389 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
390 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
391
ba1a4362
KW
392 if (check_for_problems && MB_CUR_MAX > 1
393
394 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
395 * locale. Just assume that the implementation for them (plus
396 * for POSIX) is correct and the > 1 value is spurious. (Since
397 * these are specially handled to never be considered UTF-8
398 * locales, as long as this is the only problem, everything
399 * should work fine */
400 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
401 {
8c6180a9
KW
402 multi_byte_locale = TRUE;
403 }
404#endif
405
406 if (bad_count || multi_byte_locale) {
780fcc9f 407 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 408 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 409 newctype,
8c6180a9
KW
410 (multi_byte_locale)
411 ? " Some characters in it are not recognized by"
412 " Perl."
413 : "",
414 (bad_count)
415 ? "\nThe following characters (and maybe others)"
416 " may not have the same meaning as the Perl"
417 " program expects:\n"
418 : "",
419 (bad_count)
420 ? bad_chars_list
421 : ""
422 );
cc9eaeb0 423 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
424 * output the message now. If not in that scope, we save the
425 * message to be output at the first operation using this locale,
426 * if that actually happens. Most programs don't use locales, so
427 * they are immune to bad ones. */
cc9eaeb0 428 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f
KW
429
430 /* We have to save 'newctype' because the setlocale() just
431 * below may destroy it. The next setlocale() further down
432 * should restore it properly so that the intermediate change
433 * here is transparent to this function's caller */
434 const char * const badlocale = savepv(newctype);
435
436 setlocale(LC_CTYPE, "C");
437
438 /* The '0' below suppresses a bogus gcc compiler warning */
439 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 440
780fcc9f 441 setlocale(LC_CTYPE, badlocale);
c0f3a893 442 Safefree(badlocale);
bddebb56
KW
443
444 if (IN_LC(LC_CTYPE)) {
445 SvREFCNT_dec_NN(PL_warn_locale);
446 PL_warn_locale = NULL;
447 }
780fcc9f 448 }
baa60164 449 }
31f05a37 450 }
98994639
HS
451
452#endif /* USE_LOCALE_CTYPE */
7918f24d 453 PERL_ARGS_ASSERT_NEW_CTYPE;
8772537c 454 PERL_UNUSED_ARG(newctype);
96a5add6 455 PERL_UNUSED_CONTEXT;
98994639
HS
456}
457
98994639 458void
2726666d
KW
459Perl__warn_problematic_locale()
460{
2726666d
KW
461
462#ifdef USE_LOCALE_CTYPE
463
5f04a188
KW
464 dTHX;
465
466 /* Internal-to-core function that outputs the message in PL_warn_locale,
467 * and then NULLS it. Should be called only through the macro
468 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
469
2726666d
KW
470 if (PL_warn_locale) {
471 /*GCC_DIAG_IGNORE(-Wformat-security); Didn't work */
472 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
473 SvPVX(PL_warn_locale),
474 0 /* dummy to avoid compiler warning */ );
475 /* GCC_DIAG_RESTORE; */
476 SvREFCNT_dec_NN(PL_warn_locale);
477 PL_warn_locale = NULL;
478 }
479
480#endif
481
482}
483
a4f00dcc
KW
484STATIC void
485S_new_collate(pTHX_ const char *newcoll)
98994639
HS
486{
487#ifdef USE_LOCALE_COLLATE
0d071d52
KW
488
489 /* Called after all libc setlocale() calls affecting LC_COLLATE, to tell
490 * core Perl this and that 'newcoll' is the name of the new locale.
491 *
492 * Any code changing the locale (outside this file) should use
493 * POSIX::setlocale, which calls this function. Therefore this function
494 * should be called directly only from this file and from
d35fca5f
KW
495 * POSIX::setlocale().
496 *
497 * The design of locale collation is that every locale change is given an
498 * index 'PL_collation_ix'. The first time a string particpates in an
499 * operation that requires collation while locale collation is active, it
500 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
501 * magic includes the collation index, and the transformation of the string
502 * by strxfrm(), q.v. That transformation is used when doing comparisons,
503 * instead of the string itself. If a string changes, the magic is
504 * cleared. The next time the locale changes, the index is incremented,
505 * and so we know during a comparison that the transformation is not
506 * necessarily still valid, and so is recomputed. Note that if the locale
507 * changes enough times, the index could wrap (a U32), and it is possible
508 * that a transformation would improperly be considered valid, leading to
509 * an unlikely bug */
0d071d52 510
98994639
HS
511 if (! newcoll) {
512 if (PL_collation_name) {
513 ++PL_collation_ix;
514 Safefree(PL_collation_name);
515 PL_collation_name = NULL;
516 }
517 PL_collation_standard = TRUE;
00bf60ca 518 is_standard_collation:
98994639
HS
519 PL_collxfrm_base = 0;
520 PL_collxfrm_mult = 2;
165a1c52 521 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 522 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 523 PL_strxfrm_max_cp = 0;
98994639
HS
524 return;
525 }
526
d35fca5f 527 /* If this is not the same locale as currently, set the new one up */
98994639
HS
528 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
529 ++PL_collation_ix;
530 Safefree(PL_collation_name);
531 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 532 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
533 if (PL_collation_standard) {
534 goto is_standard_collation;
535 }
98994639 536
165a1c52 537 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 538 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 539 PL_strxfrm_max_cp = 0;
165a1c52 540
59c018b9
KW
541 /* A locale collation definition includes primary, secondary, tertiary,
542 * etc. weights for each character. To sort, the primary weights are
543 * used, and only if they compare equal, then the secondary weights are
544 * used, and only if they compare equal, then the tertiary, etc.
545 *
546 * strxfrm() works by taking the input string, say ABC, and creating an
547 * output transformed string consisting of first the primary weights,
548 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
549 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
550 * may not have weights at every level. In our example, let's say B
551 * doesn't have a tertiary weight, and A doesn't have a secondary
552 * weight. The constructed string is then going to be
553 * A¹B¹C¹ B²C² A³C³ ....
554 * This has the desired effect that strcmp() will look at the secondary
555 * or tertiary weights only if the strings compare equal at all higher
556 * priority weights. The spaces shown here, like in
c342d20e 557 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
558 * are not just for readability. In the general case, these must
559 * actually be bytes, which we will call here 'separator weights'; and
560 * they must be smaller than any other weight value, but since these
561 * are C strings, only the terminating one can be a NUL (some
562 * implementations may include a non-NUL separator weight just before
563 * the NUL). Implementations tend to reserve 01 for the separator
564 * weights. They are needed so that a shorter string's secondary
565 * weights won't be misconstrued as primary weights of a longer string,
566 * etc. By making them smaller than any other weight, the shorter
567 * string will sort first. (Actually, if all secondary weights are
568 * smaller than all primary ones, there is no need for a separator
569 * weight between those two levels, etc.)
570 *
571 * The length of the transformed string is roughly a linear function of
572 * the input string. It's not exactly linear because some characters
573 * don't have weights at all levels. When we call strxfrm() we have to
574 * allocate some memory to hold the transformed string. The
575 * calculations below try to find coefficients 'm' and 'b' for this
576 * locale so that m*x + b equals how much space we need, given the size
577 * of the input string in 'x'. If we calculate too small, we increase
578 * the size as needed, and call strxfrm() again, but it is better to
579 * get it right the first time to avoid wasted expensive string
580 * transformations. */
581
98994639 582 {
79f120c8
KW
583 /* We use the string below to find how long the tranformation of it
584 * is. Almost all locales are supersets of ASCII, or at least the
585 * ASCII letters. We use all of them, half upper half lower,
586 * because if we used fewer, we might hit just the ones that are
587 * outliers in a particular locale. Most of the strings being
588 * collated will contain a preponderance of letters, and even if
589 * they are above-ASCII, they are likely to have the same number of
590 * weight levels as the ASCII ones. It turns out that digits tend
591 * to have fewer levels, and some punctuation has more, but those
592 * are relatively sparse in text, and khw believes this gives a
593 * reasonable result, but it could be changed if experience so
594 * dictates. */
595 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
596 char * x_longer; /* Transformed 'longer' */
597 Size_t x_len_longer; /* Length of 'x_longer' */
598
599 char * x_shorter; /* We also transform a substring of 'longer' */
600 Size_t x_len_shorter;
601
a4a439fb 602 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
603 * are interested only in its length). It is used because it has
604 * the intelligence to handle all cases, but to work, it needs some
605 * values of 'm' and 'b' to get it started. For the purposes of
606 * this calculation we use a very conservative estimate of 'm' and
607 * 'b'. This assumes a weight can be multiple bytes, enough to
608 * hold any UV on the platform, and there are 5 levels, 4 weight
609 * bytes, and a trailing NUL. */
610 PL_collxfrm_base = 5;
611 PL_collxfrm_mult = 5 * sizeof(UV);
612
613 /* Find out how long the transformation really is */
a4a439fb
KW
614 x_longer = _mem_collxfrm(longer,
615 sizeof(longer) - 1,
616 &x_len_longer,
617
618 /* We avoid converting to UTF-8 in the
619 * called function by telling it the
620 * string is in UTF-8 if the locale is a
621 * UTF-8 one. Since the string passed
622 * here is invariant under UTF-8, we can
623 * claim it's UTF-8 even though it isn't.
624 * */
625 PL_in_utf8_COLLATE_locale);
79f120c8
KW
626 Safefree(x_longer);
627
628 /* Find out how long the transformation of a substring of 'longer'
629 * is. Together the lengths of these transformations are
630 * sufficient to calculate 'm' and 'b'. The substring is all of
631 * 'longer' except the first character. This minimizes the chances
632 * of being swayed by outliers */
a4a439fb 633 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 634 sizeof(longer) - 2,
a4a439fb
KW
635 &x_len_shorter,
636 PL_in_utf8_COLLATE_locale);
79f120c8
KW
637 Safefree(x_shorter);
638
639 /* If the results are nonsensical for this simple test, the whole
640 * locale definition is suspect. Mark it so that locale collation
641 * is not active at all for it. XXX Should we warn? */
642 if ( x_len_shorter == 0
643 || x_len_longer == 0
644 || x_len_shorter >= x_len_longer)
645 {
646 PL_collxfrm_mult = 0;
647 PL_collxfrm_base = 0;
648 }
649 else {
650 SSize_t base; /* Temporary */
651
652 /* We have both: m * strlen(longer) + b = x_len_longer
653 * m * strlen(shorter) + b = x_len_shorter;
654 * subtracting yields:
655 * m * (strlen(longer) - strlen(shorter))
656 * = x_len_longer - x_len_shorter
657 * But we have set things up so that 'shorter' is 1 byte smaller
658 * than 'longer'. Hence:
659 * m = x_len_longer - x_len_shorter
660 *
661 * But if something went wrong, make sure the multiplier is at
662 * least 1.
663 */
664 if (x_len_longer > x_len_shorter) {
665 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
666 }
667 else {
668 PL_collxfrm_mult = 1;
669 }
670
671 /* mx + b = len
672 * so: b = len - mx
673 * but in case something has gone wrong, make sure it is
674 * non-negative */
675 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
676 if (base < 0) {
677 base = 0;
678 }
679
680 /* Add 1 for the trailing NUL */
681 PL_collxfrm_base = base + 1;
682 }
58eebef2
KW
683
684#ifdef DEBUGGING
685 if (DEBUG_L_TEST || debug_initialization) {
686 PerlIO_printf(Perl_debug_log,
b07929e4
KW
687 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
688 "x_len_longer=%zu,"
689 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
690 __FILE__, __LINE__,
691 PL_in_utf8_COLLATE_locale,
692 x_len_shorter, x_len_longer,
693 PL_collxfrm_mult, PL_collxfrm_base);
694 }
695#endif
98994639
HS
696 }
697 }
698
f2ce9e1c
JH
699#else
700 PERL_UNUSED_ARG(newcoll);
98994639
HS
701#endif /* USE_LOCALE_COLLATE */
702}
703
a4f00dcc 704#ifndef WIN32 /* No wrapper except on Windows */
b385bb4d 705
a4f00dcc
KW
706#define my_setlocale(a,b) setlocale(a,b)
707
708#else /* WIN32 */
709
710STATIC char *
711S_my_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
712{
713 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
714 * difference unless the input locale is "", which means on Windows to get
715 * the machine default, which is set via the computer's "Regional and
716 * Language Options" (or its current equivalent). In POSIX, it instead
717 * means to find the locale from the user's environment. This routine
718 * looks in the environment, and, if anything is found, uses that instead
719 * of going to the machine default. If there is no environment override,
720 * the machine default is used, as normal, by calling the real setlocale()
721 * with "". The POSIX behavior is to use the LC_ALL variable if set;
722 * otherwise to use the particular category's variable if set; otherwise to
723 * use the LANG variable. */
724
175c4cf9 725 bool override_LC_ALL = FALSE;
89f7b9aa
KW
726 char * result;
727
b385bb4d
KW
728 if (locale && strEQ(locale, "")) {
729# ifdef LC_ALL
730 locale = PerlEnv_getenv("LC_ALL");
731 if (! locale) {
732#endif
733 switch (category) {
734# ifdef LC_ALL
735 case LC_ALL:
481465ea 736 override_LC_ALL = TRUE;
b385bb4d
KW
737 break; /* We already know its variable isn't set */
738# endif
739# ifdef USE_LOCALE_TIME
740 case LC_TIME:
741 locale = PerlEnv_getenv("LC_TIME");
742 break;
743# endif
744# ifdef USE_LOCALE_CTYPE
745 case LC_CTYPE:
746 locale = PerlEnv_getenv("LC_CTYPE");
747 break;
748# endif
749# ifdef USE_LOCALE_COLLATE
750 case LC_COLLATE:
751 locale = PerlEnv_getenv("LC_COLLATE");
752 break;
753# endif
754# ifdef USE_LOCALE_MONETARY
755 case LC_MONETARY:
756 locale = PerlEnv_getenv("LC_MONETARY");
757 break;
758# endif
759# ifdef USE_LOCALE_NUMERIC
760 case LC_NUMERIC:
761 locale = PerlEnv_getenv("LC_NUMERIC");
762 break;
763# endif
764# ifdef USE_LOCALE_MESSAGES
765 case LC_MESSAGES:
766 locale = PerlEnv_getenv("LC_MESSAGES");
767 break;
768# endif
769 default:
770 /* This is a category, like PAPER_SIZE that we don't
771 * know about; and so can't provide a wrapper. */
772 break;
773 }
774 if (! locale) {
775 locale = PerlEnv_getenv("LANG");
481465ea 776 if (! locale) {
b385bb4d
KW
777 locale = "";
778 }
779 }
780# ifdef LC_ALL
781 }
782# endif
783 }
784
89f7b9aa 785 result = setlocale(category, locale);
bbc98134 786 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
a4f00dcc 787 setlocale_debug_string(category, locale, result)));
89f7b9aa 788
481465ea 789 if (! override_LC_ALL) {
89f7b9aa
KW
790 return result;
791 }
792
dfd77d7a 793 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
794 * LANG variable or the system default if there is no LANG. But these have
795 * lower priority than the other LC_foo variables, so override it for each
796 * one that is set. (If they are set to "", it means to use the same thing
797 * we just set LC_ALL to, so can skip) */
89f7b9aa
KW
798# ifdef USE_LOCALE_TIME
799 result = PerlEnv_getenv("LC_TIME");
730252b2 800 if (result && strNE(result, "")) {
89f7b9aa 801 setlocale(LC_TIME, result);
bbc98134
KW
802 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
803 __FILE__, __LINE__,
a4f00dcc 804 setlocale_debug_string(LC_TIME, result, "not captured")));
89f7b9aa
KW
805 }
806# endif
807# ifdef USE_LOCALE_CTYPE
808 result = PerlEnv_getenv("LC_CTYPE");
730252b2 809 if (result && strNE(result, "")) {
89f7b9aa 810 setlocale(LC_CTYPE, result);
bbc98134
KW
811 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
812 __FILE__, __LINE__,
a4f00dcc 813 setlocale_debug_string(LC_CTYPE, result, "not captured")));
89f7b9aa
KW
814 }
815# endif
816# ifdef USE_LOCALE_COLLATE
817 result = PerlEnv_getenv("LC_COLLATE");
730252b2 818 if (result && strNE(result, "")) {
89f7b9aa 819 setlocale(LC_COLLATE, result);
bbc98134
KW
820 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
821 __FILE__, __LINE__,
a4f00dcc 822 setlocale_debug_string(LC_COLLATE, result, "not captured")));
89f7b9aa
KW
823 }
824# endif
825# ifdef USE_LOCALE_MONETARY
826 result = PerlEnv_getenv("LC_MONETARY");
730252b2 827 if (result && strNE(result, "")) {
89f7b9aa 828 setlocale(LC_MONETARY, result);
bbc98134
KW
829 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
830 __FILE__, __LINE__,
a4f00dcc 831 setlocale_debug_string(LC_MONETARY, result, "not captured")));
89f7b9aa
KW
832 }
833# endif
834# ifdef USE_LOCALE_NUMERIC
835 result = PerlEnv_getenv("LC_NUMERIC");
730252b2 836 if (result && strNE(result, "")) {
89f7b9aa 837 setlocale(LC_NUMERIC, result);
bbc98134
KW
838 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
839 __FILE__, __LINE__,
a4f00dcc 840 setlocale_debug_string(LC_NUMERIC, result, "not captured")));
89f7b9aa
KW
841 }
842# endif
843# ifdef USE_LOCALE_MESSAGES
844 result = PerlEnv_getenv("LC_MESSAGES");
730252b2 845 if (result && strNE(result, "")) {
89f7b9aa 846 setlocale(LC_MESSAGES, result);
bbc98134
KW
847 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
848 __FILE__, __LINE__,
a4f00dcc 849 setlocale_debug_string(LC_MESSAGES, result, "not captured")));
89f7b9aa
KW
850 }
851# endif
852
bbc98134
KW
853 result = setlocale(LC_ALL, NULL);
854 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
855 __FILE__, __LINE__,
a4f00dcc 856 setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 857
bbc98134 858 return result;
b385bb4d
KW
859}
860
861#endif
862
a4f00dcc
KW
863char *
864Perl_setlocale(int category, const char * locale)
865{
866 /* This wraps POSIX::setlocale() */
867
868 char * retval;
869 dTHX;
870
871
872#ifdef USE_LOCALE_NUMERIC
873
874 /* A NULL locale means only query what the current one is. We
875 * have the LC_NUMERIC name saved, because we are normally switched
876 * into the C locale for it. Switch back so an LC_ALL query will yield
877 * the correct results; all other categories don't require special
878 * handling */
879 if (locale == NULL) {
880 if (category == LC_NUMERIC) {
881 return savepv(PL_numeric_name);
882 }
883
884# ifdef LC_ALL
885
886 else if (category == LC_ALL) {
887 SET_NUMERIC_UNDERLYING();
888 }
889
890# endif
891
892 }
893
894#endif
895
896 retval = my_setlocale(category, locale);
897
898 DEBUG_L(PerlIO_printf(Perl_debug_log,
899 "%s:%d: %s\n", __FILE__, __LINE__,
900 setlocale_debug_string(category, locale, retval)));
901 if (! retval) {
902 /* Should never happen that a query would return an error, but be
903 * sure and reset to C locale */
904 if (locale == 0) {
905 SET_NUMERIC_STANDARD();
906 }
907 return NULL;
908 }
909
910 /* Save retval since subsequent setlocale() calls may overwrite it. */
911 retval = savepv(retval);
912
913 /* If locale == NULL, we are just querying the state, but may have switched
914 * to NUMERIC_UNDERLYING. Switch back before returning. */
915 if (locale == NULL) {
916 SET_NUMERIC_STANDARD();
917 return retval;
918 }
919 else { /* Now that have switched locales, we have to update our records to
920 correspond */
921
922#ifdef USE_LOCALE_CTYPE
923
924 if ( category == LC_CTYPE
925
926# ifdef LC_ALL
927
928 || category == LC_ALL
929
930# endif
931
932 )
933 {
934 char *newctype;
935
936# ifdef LC_ALL
937
938 if (category == LC_ALL) {
939 newctype = setlocale(LC_CTYPE, NULL);
940 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
941 "%s:%d: %s\n", __FILE__, __LINE__,
942 setlocale_debug_string(LC_CTYPE, NULL, newctype)));
943 }
944 else
945
946# endif
947
948 newctype = retval;
949 new_ctype(newctype);
950 }
951
952#endif /* USE_LOCALE_CTYPE */
953
954#ifdef USE_LOCALE_COLLATE
955
956 if ( category == LC_COLLATE
957
958# ifdef LC_ALL
959
960 || category == LC_ALL
961
962# endif
963
964 )
965 {
966 char *newcoll;
967
968# ifdef LC_ALL
969
970 if (category == LC_ALL) {
971 newcoll = setlocale(LC_COLLATE, NULL);
972 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
973 "%s:%d: %s\n", __FILE__, __LINE__,
974 setlocale_debug_string(LC_COLLATE, NULL, newcoll)));
975 }
976 else
977
978# endif
979
980 newcoll = retval;
981 new_collate(newcoll);
982 }
983
984#endif /* USE_LOCALE_COLLATE */
985
986#ifdef USE_LOCALE_NUMERIC
987
988 if ( category == LC_NUMERIC
989
990# ifdef LC_ALL
991
992 || category == LC_ALL
993
994# endif
995
996 )
997 {
998 char *newnum;
999
1000# ifdef LC_ALL
1001
1002 if (category == LC_ALL) {
1003 newnum = setlocale(LC_NUMERIC, NULL);
1004 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1005 "%s:%d: %s\n", __FILE__, __LINE__,
1006 setlocale_debug_string(LC_NUMERIC, NULL, newnum)));
1007 }
1008 else
1009
1010# endif
1011
1012 newnum = retval;
1013 new_numeric(newnum);
1014 }
1015
1016#endif /* USE_LOCALE_NUMERIC */
1017
1018 }
1019
1020 return retval;
1021
f7416781
KW
1022
1023}
1024
1025PERL_STATIC_INLINE const char *
1026S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1027{
1028 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1029 * growing it if necessary */
1030
1031 const Size_t string_size = strlen(string) + offset + 1;
1032
1033 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1034
1035 if (*buf_size == 0) {
1036 Newx(*buf, string_size, char);
1037 *buf_size = string_size;
1038 }
1039 else if (string_size > *buf_size) {
1040 Renew(*buf, string_size, char);
1041 *buf_size = string_size;
1042 }
1043
1044 Copy(string, *buf + offset, string_size - offset, char);
1045 return *buf;
1046}
1047
1048/*
1049
1050=head1 Locale-related functions and macros
1051
1052=for apidoc Perl_langinfo
1053
1054This is an (almostª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
1055taking the same C<item> parameter values, and returning the same information.
1056But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1057of Perl's locale handling from your code, and can be used on systems that lack
1058a native C<nl_langinfo>.
1059
1060Expanding on these:
1061
1062=over
1063
1064=item *
1065
1066It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1067without you having to write extra code. The reason for the extra code would be
1068because these are from the C<LC_NUMERIC> locale category, which is normally
1069kept set to the C locale by Perl, no matter what the underlying locale is
1070supposed to be, and so to get the expected results, you have to temporarily
1071toggle into the underlying locale, and later toggle back. (You could use
1072plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1073but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1074keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1075expecting the radix (decimal point) character to be a dot.)
1076
1077=item *
1078
1079Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1080makes your code more portable. Of the fifty-some possible items specified by
1081the POSIX 2008 standard,
1082L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1083only two are completely unimplemented. It uses various techniques to recover
1084the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1085both of which are specified in C89, so should be always be available. Later
1086C<strftime()> versions have additional capabilities; C<""> is returned for
1087those not available on your system.
1088
1089The details for those items which may differ from what this emulation returns
1090and what a native C<nl_langinfo()> would return are:
1091
1092=over
1093
1094=item C<CODESET>
1095
1096=item C<ERA>
1097
1098Unimplemented, so returns C<"">.
1099
1100=item C<YESEXPR>
1101
1102=item C<NOEXPR>
1103
1104Only the values for English are returned. Earlier POSIX standards also
1105specified C<YESSTR> and C<NOSTR>, but these have been removed from POSIX 2008,
1106and aren't supported by C<Perl_langinfo>.
1107
1108=item C<D_FMT>
1109
1110Always evaluates to C<%x>, the locale's appropriate date representation.
1111
1112=item C<T_FMT>
1113
1114Always evaluates to C<%X>, the locale's appropriate time representation.
1115
1116=item C<D_T_FMT>
1117
1118Always evaluates to C<%c>, the locale's appropriate date and time
1119representation.
1120
1121=item C<CRNCYSTR>
1122
1123The return may be incorrect for those rare locales where the currency symbol
1124replaces the radix character.
1125Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1126to work differently.
1127
1128=item C<ALT_DIGITS>
1129
1130Currently this gives the same results as Linux does.
1131Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1132to work differently.
1133
1134=item C<ERA_D_FMT>
1135
1136=item C<ERA_T_FMT>
1137
1138=item C<ERA_D_T_FMT>
1139
1140=item C<T_FMT_AMPM>
1141
1142These are derived by using C<strftime()>, and not all versions of that function
1143know about them. C<""> is returned for these on such systems.
1144
1145=back
1146
1147When using C<Perl_langinfo> on systems that don't have a native
1148C<nl_langinfo()>, you must
1149
1150 #include "perl_langinfo.h"
1151
1152before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1153C<#include> with this one. (Doing it this way keeps out the symbols that plain
1154C<langinfo.h> imports into the namespace for code that doesn't need it.)
1155
1156You also should not use the bare C<langinfo.h> item names, but should preface
1157them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1158The C<PERL_I<foo>> versions will also work for this function on systems that do
1159have a native C<nl_langinfo>.
1160
1161=item *
1162
1163It is thread-friendly, returning its result in a buffer that won't be
1164overwritten by another thread, so you don't have to code for that possibility.
1165The buffer can be overwritten by the next call to C<nl_langinfo> or
1166C<Perl_langinfo> in the same thread.
1167
1168=item *
1169
1170ªIt returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
1171*>>, but you are (only by documentation) forbidden to write into the buffer.
1172By declaring this C<const>, the compiler enforces this restriction. The extra
1173C<const> is why this isn't an unequivocal drop-in replacement for
1174C<nl_langinfo>.
1175
1176=back
1177
1178The original impetus for C<Perl_langinfo()> was so that code that needs to
1179find out the current currency symbol, floating point radix character, or digit
1180grouping separator can use, on all systems, the simpler and more
1181thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1182pain to make thread-friendly. For other fields returned by C<localeconv>, it
1183is better to use the methods given in L<perlcall> to call
1184L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1185
1186=cut
1187
1188*/
1189
1190const char *
1191#ifdef HAS_NL_LANGINFO
1192Perl_langinfo(const nl_item item)
1193#else
1194Perl_langinfo(const int item)
1195#endif
1196{
1197 bool toggle = TRUE;
1198
1199#if defined(HAS_NL_LANGINFO)
1200# if ! defined(USE_ITHREADS)
1201
1202 /* Single-thread, and nl_langinfo() is available. Call it, switching to
1203 * underlying LC_NUMERIC for those items dependent on it */
1204
1205 const char * retval;
1206
1207 if (toggle) {
1208 if (item == PERL_RADIXCHAR || item == PERL_THOUSEP) {
1209 setlocale(LC_NUMERIC, PL_numeric_name);
1210 }
1211 else {
1212 toggle = FALSE;
1213 }
1214 }
1215
1216 retval = nl_langinfo(item);
1217
1218 if (toggle) {
1219 setlocale(LC_NUMERIC, "C");
1220 }
1221
1222 return retval;
1223
1224
1225# else
1226
1227 /* Multi-threaded, with native nl_langinfo(). Use it, copying result to
1228 * per-thread buffer, and toggling LC_NUMERIC if necessary, all within a
1229 * crtical section */
1230
1231 dTHX;
1232
1233 LOCALE_LOCK;
1234
1235 if (toggle) {
1236 if (item == PERL_RADIXCHAR || item == PERL_THOUSEP) {
1237 setlocale(LC_NUMERIC, PL_numeric_name);
1238 }
1239 else {
1240 toggle = FALSE;
1241 }
1242 }
1243
1244 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1245
1246 if (toggle) {
1247 setlocale(LC_NUMERIC, "C");
1248 }
1249
1250 LOCALE_UNLOCK;
1251
1252 return PL_langinfo_buf;
1253
1254# endif
1255#else /* Below, emulate nl_langinfo as best we can */
1256
1257 dTHX;
1258
1259# ifdef HAS_LOCALECONV
1260
1261 const struct lconv* lc;
1262
1263# endif
1264# ifdef HAS_STRFTIME
1265
1266 struct tm tm;
1267 bool return_format = FALSE; /* Return the %format, not the value */
1268 const char * format;
1269
1270# endif
1271
1272 /* We copy the results to a per-thread buffer, even if not multi-threaded.
1273 * This is in part to simplify this code, and partly because we need a
1274 * buffer anyway for strftime(), and partly because a call of localeconv()
1275 * could otherwise wipe out the buffer, and the programmer would not be
1276 * expecting this, as this is a nl_langinfo() substitute after all, so s/he
1277 * might be thinking their localeconv() is safe until another localeconv()
1278 * call. */
1279
1280 switch (item) {
1281 Size_t len;
1282 const char * retval;
1283
1284 /* These 2 are unimplemented */
1285 case PERL_CODESET:
1286 case PERL_ERA: /* For use with strftime() %E modifier */
1287
1288 default:
1289 return "";
1290
1291 /* We use only an English set, since we don't know any more */
1292 case PERL_YESEXPR: return "^[+1yY]";
1293 case PERL_NOEXPR: return "^[-0nN]";
1294
1295# ifdef HAS_LOCALECONV
1296
1297 case PERL_CRNCYSTR:
1298
1299 LOCALE_LOCK;
1300
1301 lc = localeconv();
1302 if (! lc || ! lc->currency_symbol || strEQ("", lc->currency_symbol))
1303 {
1304 LOCALE_UNLOCK;
1305 return "";
1306 }
1307
1308 /* Leave the first spot empty to be filled in below */
1309 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1310 &PL_langinfo_bufsize, 1);
1311 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1312 { /* khw couldn't figure out how the localedef specifications
1313 would show that the $ should replace the radix; this is
1314 just a guess as to how it might work.*/
1315 *PL_langinfo_buf = '.';
1316 }
1317 else if (lc->p_cs_precedes) {
1318 *PL_langinfo_buf = '-';
1319 }
1320 else {
1321 *PL_langinfo_buf = '+';
1322 }
1323
1324 LOCALE_UNLOCK;
1325 break;
1326
1327 case PERL_RADIXCHAR:
1328 case PERL_THOUSEP:
1329
1330 LOCALE_LOCK;
1331
1332 if (toggle) {
1333 setlocale(LC_NUMERIC, PL_numeric_name);
1334 }
1335
1336 lc = localeconv();
1337 if (! lc) {
1338 retval = "";
1339 }
1340 else switch (item) {
1341 case PERL_RADIXCHAR:
1342 if (! lc->decimal_point) {
1343 retval = "";
1344 }
1345 else {
1346 retval = lc->decimal_point;
1347 }
1348 break;
1349
1350 case PERL_THOUSEP:
1351 if (! lc->thousands_sep || strEQ("", lc->thousands_sep)) {
1352 retval = "";
1353 }
1354 else {
1355 retval = lc->thousands_sep;
1356 }
1357 break;
1358
1359 default:
1360 LOCALE_UNLOCK;
1361 Perl_croak(aTHX_ "panic: %s: %d: switch case: %d problem",
1362 __FILE__, __LINE__, item);
1363 }
1364
1365 save_to_buffer(retval, &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1366
1367 if (toggle) {
1368 setlocale(LC_NUMERIC, "C");
1369 }
1370
1371 LOCALE_UNLOCK;
1372
1373 break;
1374
1375# endif
1376# ifdef HAS_STRFTIME
1377
1378 /* These are defined by C89, so we assume that strftime supports them,
1379 * and so are returned unconditionally; they may not be what the locale
1380 * actually says, but should give good enough results for someone using
1381 * them as formats (as opposed to trying to parse them to figure out
1382 * what the locale says). The other format ones are actually tested to
1383 * verify they work on the platform */
1384 case PERL_D_FMT: return "%x";
1385 case PERL_T_FMT: return "%X";
1386 case PERL_D_T_FMT: return "%c";
1387
1388 /* These formats are only available in later strfmtime's */
1389 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1390 case PERL_T_FMT_AMPM:
1391
1392 /* The rest can be gotten from most versions of strftime(). */
1393 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1394 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1395 case PERL_ABDAY_7:
1396 case PERL_ALT_DIGITS:
1397 case PERL_AM_STR: case PERL_PM_STR:
1398 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1399 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1400 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1401 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1402 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1403 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1404 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1405 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1406 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11: case PERL_MON_12:
1407
1408 LOCALE_LOCK;
1409
1410 init_tm(&tm); /* Precaution against core dumps */
1411 tm.tm_sec = 30;
1412 tm.tm_min = 30;
1413 tm.tm_hour = 6;
1414 tm.tm_year = 2017 - 1900;
1415 tm.tm_wday = 0;
1416 tm.tm_mon = 0;
1417 switch (item) {
1418 default:
1419 LOCALE_UNLOCK;
1420 Perl_croak(aTHX_ "panic: %s: %d: switch case: %d problem",
1421 __FILE__, __LINE__, item);
1422 NOT_REACHED; /* NOTREACHED */
1423
1424 case PERL_PM_STR: tm.tm_hour = 18;
1425 case PERL_AM_STR:
1426 format = "%p";
1427 break;
1428
1429 case PERL_ABDAY_7: tm.tm_wday++;
1430 case PERL_ABDAY_6: tm.tm_wday++;
1431 case PERL_ABDAY_5: tm.tm_wday++;
1432 case PERL_ABDAY_4: tm.tm_wday++;
1433 case PERL_ABDAY_3: tm.tm_wday++;
1434 case PERL_ABDAY_2: tm.tm_wday++;
1435 case PERL_ABDAY_1:
1436 format = "%a";
1437 break;
1438
1439 case PERL_DAY_7: tm.tm_wday++;
1440 case PERL_DAY_6: tm.tm_wday++;
1441 case PERL_DAY_5: tm.tm_wday++;
1442 case PERL_DAY_4: tm.tm_wday++;
1443 case PERL_DAY_3: tm.tm_wday++;
1444 case PERL_DAY_2: tm.tm_wday++;
1445 case PERL_DAY_1:
1446 format = "%A";
1447 break;
1448
1449 case PERL_ABMON_12: tm.tm_mon++;
1450 case PERL_ABMON_11: tm.tm_mon++;
1451 case PERL_ABMON_10: tm.tm_mon++;
1452 case PERL_ABMON_9: tm.tm_mon++;
1453 case PERL_ABMON_8: tm.tm_mon++;
1454 case PERL_ABMON_7: tm.tm_mon++;
1455 case PERL_ABMON_6: tm.tm_mon++;
1456 case PERL_ABMON_5: tm.tm_mon++;
1457 case PERL_ABMON_4: tm.tm_mon++;
1458 case PERL_ABMON_3: tm.tm_mon++;
1459 case PERL_ABMON_2: tm.tm_mon++;
1460 case PERL_ABMON_1:
1461 format = "%b";
1462 break;
1463
1464 case PERL_MON_12: tm.tm_mon++;
1465 case PERL_MON_11: tm.tm_mon++;
1466 case PERL_MON_10: tm.tm_mon++;
1467 case PERL_MON_9: tm.tm_mon++;
1468 case PERL_MON_8: tm.tm_mon++;
1469 case PERL_MON_7: tm.tm_mon++;
1470 case PERL_MON_6: tm.tm_mon++;
1471 case PERL_MON_5: tm.tm_mon++;
1472 case PERL_MON_4: tm.tm_mon++;
1473 case PERL_MON_3: tm.tm_mon++;
1474 case PERL_MON_2: tm.tm_mon++;
1475 case PERL_MON_1:
1476 format = "%B";
1477 break;
1478
1479 case PERL_T_FMT_AMPM:
1480 format = "%r";
1481 return_format = TRUE;
1482 break;
1483
1484 case PERL_ERA_D_FMT:
1485 format = "%Ex";
1486 return_format = TRUE;
1487 break;
1488
1489 case PERL_ERA_T_FMT:
1490 format = "%EX";
1491 return_format = TRUE;
1492 break;
1493
1494 case PERL_ERA_D_T_FMT:
1495 format = "%Ec";
1496 return_format = TRUE;
1497 break;
1498
1499 case PERL_ALT_DIGITS:
1500 tm.tm_wday = 0;
1501 format = "%Ow"; /* Find the alternate digit for 0 */
1502 break;
1503 }
1504
1505 /* We can't use my_strftime() because it doesn't look at tm_wday */
1506 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1507 format, &tm))
1508 {
1509 /* A zero return means one of:
1510 * a) there wasn't enough space in PL_langinfo_buf
1511 * b) the format, like a plain %p, returns empty
1512 * c) it was an illegal format, though some implementations of
1513 * strftime will just return the illegal format as a plain
1514 * character sequence.
1515 *
1516 * To quickly test for case 'b)', try again but precede the
1517 * format with a plain character. If that result is still
1518 * empty, the problem is either 'a)' or 'c)' */
1519
1520 Size_t format_size = strlen(format) + 1;
1521 Size_t mod_size = format_size + 1;
1522 char * mod_format;
1523 char * temp_result;
1524
1525 Newx(mod_format, mod_size, char);
1526 Newx(temp_result, PL_langinfo_bufsize, char);
1527 *mod_format = '\a';
1528 my_strlcpy(mod_format + 1, format, mod_size);
1529 len = strftime(temp_result,
1530 PL_langinfo_bufsize,
1531 mod_format, &tm);
1532 Safefree(mod_format);
1533 Safefree(temp_result);
1534
1535 /* If 'len' is non-zero, it means that we had a case like %p
1536 * which means the current locale doesn't use a.m. or p.m., and
1537 * that is valid */
1538 if (len == 0) {
1539
1540 /* Here, still didn't work. If we get well beyond a
1541 * reasonable size, bail out to prevent an infinite loop. */
1542
1543 if (PL_langinfo_bufsize > 100 * format_size) {
1544 *PL_langinfo_buf = '\0';
1545 }
1546 else { /* Double the buffer size to retry; Add 1 in case
1547 original was 0, so we aren't stuck at 0. */
1548 PL_langinfo_bufsize *= 2;
1549 PL_langinfo_bufsize++;
1550 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1551 continue;
1552 }
1553 }
1554
1555 break;
1556 }
1557
1558 /* Here, we got a result.
1559 *
1560 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1561 * alternate format for wday 0. If the value is the same as the
1562 * normal 0, there isn't an alternate, so clear the buffer. */
1563 if ( item == PERL_ALT_DIGITS
1564 && strEQ(PL_langinfo_buf, "0"))
1565 {
1566 *PL_langinfo_buf = '\0';
1567 }
1568
1569 /* ALT_DIGITS is problematic. Experiments on it showed that
1570 * strftime() did not always work properly when going from alt-9 to
1571 * alt-10. Only a few locales have this item defined, and in all
1572 * of them on Linux that khw was able to find, nl_langinfo() merely
1573 * returned the alt-0 character, possibly doubled. Most Unicode
1574 * digits are in blocks of 10 consecutive code points, so that is
1575 * sufficient information for those scripts, as we can infer alt-1,
1576 * alt-2, .... But for a Japanese locale, a CJK ideographic 0 is
1577 * returned, and the CJK digits are not in code point order, so you
1578 * can't really infer anything. The localedef for this locale did
1579 * specify the succeeding digits, so that strftime() works properly
1580 * on them, without needing to infer anything. But the
1581 * nl_langinfo() return did not give sufficient information for the
1582 * caller to understand what's going on. So until there is
1583 * evidence that it should work differently, this returns the alt-0
1584 * string for ALT_DIGITS.
1585 *
1586 * wday was chosen because its range is all a single digit. Things
1587 * like tm_sec have two digits as the minimum: '00' */
1588
1589 LOCALE_UNLOCK;
1590
1591 /* If to return the format, not the value, overwrite the buffer
1592 * with it. But some strftime()s will keep the original format if
1593 * illegal, so change those to "" */
1594 if (return_format) {
1595 if (strEQ(PL_langinfo_buf, format)) {
1596 *PL_langinfo_buf = '\0';
1597 }
1598 else {
1599 save_to_buffer(format, &PL_langinfo_buf,
1600 &PL_langinfo_bufsize, 0);
1601 }
1602 }
1603
1604 break;
1605
1606# endif
1607
1608 }
1609
1610 return PL_langinfo_buf;
1611
1612#endif
1613
a4f00dcc 1614}
b385bb4d 1615
98994639
HS
1616/*
1617 * Initialize locale awareness.
1618 */
1619int
1620Perl_init_i18nl10n(pTHX_ int printwarn)
1621{
0e92a118
KW
1622 /* printwarn is
1623 *
1624 * 0 if not to output warning when setup locale is bad
1625 * 1 if to output warning based on value of PERL_BADLANG
1626 * >1 if to output regardless of PERL_BADLANG
1627 *
1628 * returns
98994639 1629 * 1 = set ok or not applicable,
0e92a118
KW
1630 * 0 = fallback to a locale of lower priority
1631 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1632 *
1633 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1634 * set, debugging information is output.
1635 *
1636 * This looks more complicated than it is, mainly due to the #ifdefs.
1637 *
1638 * We try to set LC_ALL to the value determined by the environment. If
1639 * there is no LC_ALL on this platform, we try the individual categories we
1640 * know about. If this works, we are done.
1641 *
1642 * But if it doesn't work, we have to do something else. We search the
1643 * environment variables ourselves instead of relying on the system to do
1644 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1645 * think there is one), and the ultimate fallback "C". This is all done in
1646 * the same loop as above to avoid duplicating code, but it makes things
1647 * more complex. After the original failure, we add the fallback
1648 * possibilities to the list of locales to try, and iterate the loop
1649 * through them all until one succeeds.
1650 *
1651 * On Ultrix, the locale MUST come from the environment, so there is
1652 * preliminary code to set it. I (khw) am not sure that it is necessary,
1653 * and that this couldn't be folded into the loop, but barring any real
1654 * platforms to test on, it's staying as-is
1655 *
1656 * A slight complication is that in embedded Perls, the locale may already
1657 * be set-up, and we don't want to get it from the normal environment
1658 * variables. This is handled by having a special environment variable
1659 * indicate we're in this situation. We simply set setlocale's 2nd
1660 * parameter to be a NULL instead of "". That indicates to setlocale that
1661 * it is not to change anything, but to return the current value,
1662 * effectively initializing perl's db to what the locale already is.
1663 *
1664 * We play the same trick with NULL if a LC_ALL succeeds. We call
1665 * setlocale() on the individual categores with NULL to get their existing
1666 * values for our db, instead of trying to change them.
1667 * */
98994639 1668
0e92a118
KW
1669 int ok = 1;
1670
98994639 1671#if defined(USE_LOCALE)
98994639
HS
1672#ifdef USE_LOCALE_CTYPE
1673 char *curctype = NULL;
1674#endif /* USE_LOCALE_CTYPE */
1675#ifdef USE_LOCALE_COLLATE
1676 char *curcoll = NULL;
1677#endif /* USE_LOCALE_COLLATE */
1678#ifdef USE_LOCALE_NUMERIC
1679 char *curnum = NULL;
1680#endif /* USE_LOCALE_NUMERIC */
1681#ifdef __GLIBC__
175c4cf9 1682 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
98994639 1683#endif
65ebb059 1684
ccd65d51
KW
1685 /* NULL uses the existing already set up locale */
1686 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
1687 ? NULL
1688 : "";
c3fcd832
KW
1689 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
1690 unsigned int trial_locales_count;
175c4cf9
KW
1691 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
1692 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 1693 bool setlocale_failure = FALSE;
65ebb059
KW
1694 unsigned int i;
1695 char *p;
175c4cf9
KW
1696
1697 /* A later getenv() could zap this, so only use here */
1698 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
1699
1700 const bool locwarn = (printwarn > 1
1701 || (printwarn
1702 && (! bad_lang_use_once
22ff3130
HS
1703 || (
1704 /* disallow with "" or "0" */
1705 *bad_lang_use_once
1706 && strNE("0", bad_lang_use_once)))));
0e92a118 1707 bool done = FALSE;
5d1187d1
KW
1708 char * sl_result; /* return from setlocale() */
1709 char * locale_param;
6bce99ee
JH
1710#ifdef WIN32
1711 /* In some systems you can find out the system default locale
1712 * and use that as the fallback locale. */
1713# define SYSTEM_DEFAULT_LOCALE
1714#endif
1715#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059 1716 const char *system_default_locale = NULL;
6bce99ee 1717#endif
98994639 1718
2fcc0ca9 1719#ifdef DEBUGGING
8298454c 1720 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
2fcc0ca9
KW
1721# define DEBUG_LOCALE_INIT(category, locale, result) \
1722 STMT_START { \
1723 if (debug_initialization) { \
1724 PerlIO_printf(Perl_debug_log, \
1725 "%s:%d: %s\n", \
1726 __FILE__, __LINE__, \
a4f00dcc 1727 setlocale_debug_string(category, \
2fcc0ca9
KW
1728 locale, \
1729 result)); \
1730 } \
1731 } STMT_END
1732#else
1733# define DEBUG_LOCALE_INIT(a,b,c)
1734#endif
1735
0e92a118
KW
1736#ifndef LOCALE_ENVIRON_REQUIRED
1737 PERL_UNUSED_VAR(done);
5d1187d1 1738 PERL_UNUSED_VAR(locale_param);
0e92a118 1739#else
98994639
HS
1740
1741 /*
1742 * Ultrix setlocale(..., "") fails if there are no environment
1743 * variables from which to get a locale name.
1744 */
1745
b3e384bf 1746# ifdef LC_ALL
98994639 1747 if (lang) {
5d1187d1
KW
1748 sl_result = my_setlocale(LC_ALL, setlocale_init);
1749 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result);
1750 if (sl_result)
98994639
HS
1751 done = TRUE;
1752 else
1753 setlocale_failure = TRUE;
1754 }
5d1187d1 1755 if (! setlocale_failure) {
b3e384bf 1756# ifdef USE_LOCALE_CTYPE
5d1187d1
KW
1757 locale_param = (! done && (lang || PerlEnv_getenv("LC_CTYPE")))
1758 ? setlocale_init
1759 : NULL;
1760 curctype = my_setlocale(LC_CTYPE, locale_param);
1761 DEBUG_LOCALE_INIT(LC_CTYPE, locale_param, sl_result);
1762 if (! curctype)
98994639
HS
1763 setlocale_failure = TRUE;
1764 else
1765 curctype = savepv(curctype);
b3e384bf
KW
1766# endif /* USE_LOCALE_CTYPE */
1767# ifdef USE_LOCALE_COLLATE
5d1187d1
KW
1768 locale_param = (! done && (lang || PerlEnv_getenv("LC_COLLATE")))
1769 ? setlocale_init
1770 : NULL;
1771 curcoll = my_setlocale(LC_COLLATE, locale_param);
1772 DEBUG_LOCALE_INIT(LC_COLLATE, locale_param, sl_result);
1773 if (! curcoll)
98994639
HS
1774 setlocale_failure = TRUE;
1775 else
1776 curcoll = savepv(curcoll);
b3e384bf
KW
1777# endif /* USE_LOCALE_COLLATE */
1778# ifdef USE_LOCALE_NUMERIC
5d1187d1
KW
1779 locale_param = (! done && (lang || PerlEnv_getenv("LC_NUMERIC")))
1780 ? setlocale_init
1781 : NULL;
1782 curnum = my_setlocale(LC_NUMERIC, locale_param);
1783 DEBUG_LOCALE_INIT(LC_NUMERIC, locale_param, sl_result);
1784 if (! curnum)
98994639
HS
1785 setlocale_failure = TRUE;
1786 else
1787 curnum = savepv(curnum);
b3e384bf 1788# endif /* USE_LOCALE_NUMERIC */
a782673d 1789# ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
1790 locale_param = (! done && (lang || PerlEnv_getenv("LC_MESSAGES")))
1791 ? setlocale_init
1792 : NULL;
1793 sl_result = my_setlocale(LC_MESSAGES, locale_param);
1794 DEBUG_LOCALE_INIT(LC_MESSAGES, locale_param, sl_result);
9f42613c 1795 if (! sl_result) {
a782673d
KW
1796 setlocale_failure = TRUE;
1797 }
1798# endif /* USE_LOCALE_MESSAGES */
c835d6be 1799# ifdef USE_LOCALE_MONETARY
5d1187d1
KW
1800 locale_param = (! done && (lang || PerlEnv_getenv("LC_MONETARY")))
1801 ? setlocale_init
1802 : NULL;
1803 sl_result = my_setlocale(LC_MONETARY, locale_param);
1804 DEBUG_LOCALE_INIT(LC_MONETARY, locale_param, sl_result);
1805 if (! sl_result) {
c835d6be
KW
1806 setlocale_failure = TRUE;
1807 }
1808# endif /* USE_LOCALE_MONETARY */
98994639
HS
1809 }
1810
b3e384bf 1811# endif /* LC_ALL */
98994639
HS
1812
1813#endif /* !LOCALE_ENVIRON_REQUIRED */
1814
65ebb059 1815 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
1816 * the list. Normally the loop is executed just once. But if setting the
1817 * locale fails, inside the loop we add fallback trials to the array and so
1818 * will execute the loop multiple times */
c3fcd832
KW
1819 trial_locales[0] = setlocale_init;
1820 trial_locales_count = 1;
65ebb059
KW
1821 for (i= 0; i < trial_locales_count; i++) {
1822 const char * trial_locale = trial_locales[i];
1823
1824 if (i > 0) {
1825
1826 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
1827 * when i==0, but I (khw) don't think that behavior makes much
1828 * sense */
1829 setlocale_failure = FALSE;
1830
6bce99ee
JH
1831#ifdef SYSTEM_DEFAULT_LOCALE
1832# ifdef WIN32
65ebb059
KW
1833 /* On Windows machines, an entry of "" after the 0th means to use
1834 * the system default locale, which we now proceed to get. */
1835 if (strEQ(trial_locale, "")) {
1836 unsigned int j;
1837
1838 /* Note that this may change the locale, but we are going to do
1839 * that anyway just below */
1840 system_default_locale = setlocale(LC_ALL, "");
5d1187d1 1841 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059
KW
1842
1843 /* Skip if invalid or it's already on the list of locales to
1844 * try */
1845 if (! system_default_locale) {
1846 goto next_iteration;
1847 }
1848 for (j = 0; j < trial_locales_count; j++) {
1849 if (strEQ(system_default_locale, trial_locales[j])) {
1850 goto next_iteration;
1851 }
1852 }
1853
1854 trial_locale = system_default_locale;
1855 }
6bce99ee
JH
1856# endif /* WIN32 */
1857#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
1858 }
1859
98994639 1860#ifdef LC_ALL
5d1187d1
KW
1861 sl_result = my_setlocale(LC_ALL, trial_locale);
1862 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result);
1863 if (! sl_result) {
49c85077 1864 setlocale_failure = TRUE;
7cd8b568
KW
1865 }
1866 else {
1867 /* Since LC_ALL succeeded, it should have changed all the other
1868 * categories it can to its value; so we massage things so that the
1869 * setlocales below just return their category's current values.
1870 * This adequately handles the case in NetBSD where LC_COLLATE may
1871 * not be defined for a locale, and setting it individually will
1872 * fail, whereas setting LC_ALL suceeds, leaving LC_COLLATE set to
1873 * the POSIX locale. */
1874 trial_locale = NULL;
1875 }
98994639
HS
1876#endif /* LC_ALL */
1877
49c85077 1878 if (!setlocale_failure) {
98994639 1879#ifdef USE_LOCALE_CTYPE
49c85077 1880 Safefree(curctype);
5d1187d1
KW
1881 curctype = my_setlocale(LC_CTYPE, trial_locale);
1882 DEBUG_LOCALE_INIT(LC_CTYPE, trial_locale, curctype);
1883 if (! curctype)
49c85077
KW
1884 setlocale_failure = TRUE;
1885 else
1886 curctype = savepv(curctype);
98994639
HS
1887#endif /* USE_LOCALE_CTYPE */
1888#ifdef USE_LOCALE_COLLATE
49c85077 1889 Safefree(curcoll);
5d1187d1
KW
1890 curcoll = my_setlocale(LC_COLLATE, trial_locale);
1891 DEBUG_LOCALE_INIT(LC_COLLATE, trial_locale, curcoll);
1892 if (! curcoll)
49c85077
KW
1893 setlocale_failure = TRUE;
1894 else
1895 curcoll = savepv(curcoll);
98994639
HS
1896#endif /* USE_LOCALE_COLLATE */
1897#ifdef USE_LOCALE_NUMERIC
49c85077 1898 Safefree(curnum);
5d1187d1
KW
1899 curnum = my_setlocale(LC_NUMERIC, trial_locale);
1900 DEBUG_LOCALE_INIT(LC_NUMERIC, trial_locale, curnum);
1901 if (! curnum)
49c85077
KW
1902 setlocale_failure = TRUE;
1903 else
1904 curnum = savepv(curnum);
98994639 1905#endif /* USE_LOCALE_NUMERIC */
a782673d 1906#ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
1907 sl_result = my_setlocale(LC_MESSAGES, trial_locale);
1908 DEBUG_LOCALE_INIT(LC_MESSAGES, trial_locale, sl_result);
1909 if (! (sl_result))
a782673d
KW
1910 setlocale_failure = TRUE;
1911#endif /* USE_LOCALE_MESSAGES */
c835d6be 1912#ifdef USE_LOCALE_MONETARY
5d1187d1
KW
1913 sl_result = my_setlocale(LC_MONETARY, trial_locale);
1914 DEBUG_LOCALE_INIT(LC_MONETARY, trial_locale, sl_result);
1915 if (! (sl_result))
c835d6be
KW
1916 setlocale_failure = TRUE;
1917#endif /* USE_LOCALE_MONETARY */
1918
49c85077
KW
1919 if (! setlocale_failure) { /* Success */
1920 break;
1921 }
65ebb059 1922 }
98994639 1923
49c85077
KW
1924 /* Here, something failed; will need to try a fallback. */
1925 ok = 0;
65ebb059 1926
49c85077
KW
1927 if (i == 0) {
1928 unsigned int j;
98994639 1929
65ebb059 1930 if (locwarn) { /* Output failure info only on the first one */
98994639
HS
1931#ifdef LC_ALL
1932
49c85077
KW
1933 PerlIO_printf(Perl_error_log,
1934 "perl: warning: Setting locale failed.\n");
98994639
HS
1935
1936#else /* !LC_ALL */
1937
49c85077
KW
1938 PerlIO_printf(Perl_error_log,
1939 "perl: warning: Setting locale failed for the categories:\n\t");
20a240df 1940# ifdef USE_LOCALE_CTYPE
49c85077
KW
1941 if (! curctype)
1942 PerlIO_printf(Perl_error_log, "LC_CTYPE ");
20a240df
KW
1943# endif /* USE_LOCALE_CTYPE */
1944# ifdef USE_LOCALE_COLLATE
49c85077
KW
1945 if (! curcoll)
1946 PerlIO_printf(Perl_error_log, "LC_COLLATE ");
20a240df
KW
1947# endif /* USE_LOCALE_COLLATE */
1948# ifdef USE_LOCALE_NUMERIC
49c85077
KW
1949 if (! curnum)
1950 PerlIO_printf(Perl_error_log, "LC_NUMERIC ");
20a240df 1951# endif /* USE_LOCALE_NUMERIC */
a782673d 1952 PerlIO_printf(Perl_error_log, "and possibly others\n");
98994639
HS
1953
1954#endif /* LC_ALL */
1955
49c85077
KW
1956 PerlIO_printf(Perl_error_log,
1957 "perl: warning: Please check that your locale settings:\n");
98994639
HS
1958
1959#ifdef __GLIBC__
49c85077
KW
1960 PerlIO_printf(Perl_error_log,
1961 "\tLANGUAGE = %c%s%c,\n",
1962 language ? '"' : '(',
1963 language ? language : "unset",
1964 language ? '"' : ')');
98994639
HS
1965#endif
1966
49c85077
KW
1967 PerlIO_printf(Perl_error_log,
1968 "\tLC_ALL = %c%s%c,\n",
1969 lc_all ? '"' : '(',
1970 lc_all ? lc_all : "unset",
1971 lc_all ? '"' : ')');
98994639
HS
1972
1973#if defined(USE_ENVIRON_ARRAY)
49c85077
KW
1974 {
1975 char **e;
1976 for (e = environ; *e; e++) {
0eb7f56a
YO
1977 if (strEQs(*e, "LC_")
1978 && strNEs(*e, "LC_ALL=")
49c85077
KW
1979 && (p = strchr(*e, '=')))
1980 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
1981 (int)(p - *e), *e, p + 1);
1982 }
1983 }
98994639 1984#else
49c85077
KW
1985 PerlIO_printf(Perl_error_log,
1986 "\t(possibly more locale environment variables)\n");
98994639
HS
1987#endif
1988
49c85077
KW
1989 PerlIO_printf(Perl_error_log,
1990 "\tLANG = %c%s%c\n",
1991 lang ? '"' : '(',
1992 lang ? lang : "unset",
1993 lang ? '"' : ')');
98994639 1994
49c85077
KW
1995 PerlIO_printf(Perl_error_log,
1996 " are supported and installed on your system.\n");
1997 }
98994639 1998
65ebb059 1999 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2000 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2001 * usually change the upper bound of the loop we are in.
2002 *
2003 * Since the system's default way of setting the locale has not
2004 * found one that works, We use Perl's defined ordering: LC_ALL,
2005 * LANG, and the C locale. We don't try the same locale twice, so
2006 * don't add to the list if already there. (On POSIX systems, the
2007 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2008 * but there's no harm done by doing it explicitly.
2009 *
2010 * Note that this tries the LC_ALL environment variable even on
2011 * systems which have no LC_ALL locale setting. This may or may
2012 * not have been originally intentional, but there's no real need
2013 * to change the behavior. */
65ebb059
KW
2014 if (lc_all) {
2015 for (j = 0; j < trial_locales_count; j++) {
2016 if (strEQ(lc_all, trial_locales[j])) {
2017 goto done_lc_all;
2018 }
2019 }
2020 trial_locales[trial_locales_count++] = lc_all;
2021 }
2022 done_lc_all:
98994639 2023
65ebb059
KW
2024 if (lang) {
2025 for (j = 0; j < trial_locales_count; j++) {
2026 if (strEQ(lang, trial_locales[j])) {
2027 goto done_lang;
2028 }
2029 }
2030 trial_locales[trial_locales_count++] = lang;
2031 }
2032 done_lang:
2033
2034#if defined(WIN32) && defined(LC_ALL)
2035 /* For Windows, we also try the system default locale before "C".
2036 * (If there exists a Windows without LC_ALL we skip this because
2037 * it gets too complicated. For those, the "C" is the next
2038 * fallback possibility). The "" is the same as the 0th element of
2039 * the array, but the code at the loop above knows to treat it
2040 * differently when not the 0th */
2041 trial_locales[trial_locales_count++] = "";
2042#endif
2043
2044 for (j = 0; j < trial_locales_count; j++) {
2045 if (strEQ("C", trial_locales[j])) {
2046 goto done_C;
2047 }
2048 }
2049 trial_locales[trial_locales_count++] = "C";
98994639 2050
65ebb059
KW
2051 done_C: ;
2052 } /* end of first time through the loop */
98994639 2053
65ebb059
KW
2054#ifdef WIN32
2055 next_iteration: ;
2056#endif
2057
2058 } /* end of looping through the trial locales */
2059
2060 if (ok < 1) { /* If we tried to fallback */
2061 const char* msg;
2062 if (! setlocale_failure) { /* fallback succeeded */
2063 msg = "Falling back to";
2064 }
2065 else { /* fallback failed */
98994639 2066
65ebb059
KW
2067 /* We dropped off the end of the loop, so have to decrement i to
2068 * get back to the value the last time through */
2069 i--;
98994639 2070
65ebb059
KW
2071 ok = -1;
2072 msg = "Failed to fall back to";
2073
2074 /* To continue, we should use whatever values we've got */
98994639 2075#ifdef USE_LOCALE_CTYPE
49c85077
KW
2076 Safefree(curctype);
2077 curctype = savepv(setlocale(LC_CTYPE, NULL));
5d1187d1 2078 DEBUG_LOCALE_INIT(LC_CTYPE, NULL, curctype);
98994639
HS
2079#endif /* USE_LOCALE_CTYPE */
2080#ifdef USE_LOCALE_COLLATE
49c85077
KW
2081 Safefree(curcoll);
2082 curcoll = savepv(setlocale(LC_COLLATE, NULL));
5d1187d1 2083 DEBUG_LOCALE_INIT(LC_COLLATE, NULL, curcoll);
98994639
HS
2084#endif /* USE_LOCALE_COLLATE */
2085#ifdef USE_LOCALE_NUMERIC
49c85077
KW
2086 Safefree(curnum);
2087 curnum = savepv(setlocale(LC_NUMERIC, NULL));
5d1187d1 2088 DEBUG_LOCALE_INIT(LC_NUMERIC, NULL, curnum);
98994639 2089#endif /* USE_LOCALE_NUMERIC */
65ebb059
KW
2090 }
2091
2092 if (locwarn) {
2093 const char * description;
2094 const char * name = "";
2095 if (strEQ(trial_locales[i], "C")) {
2096 description = "the standard locale";
2097 name = "C";
2098 }
6bce99ee 2099#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059
KW
2100 else if (strEQ(trial_locales[i], "")) {
2101 description = "the system default locale";
2102 if (system_default_locale) {
2103 name = system_default_locale;
2104 }
2105 }
6bce99ee 2106#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
2107 else {
2108 description = "a fallback locale";
2109 name = trial_locales[i];
2110 }
2111 if (name && strNE(name, "")) {
2112 PerlIO_printf(Perl_error_log,
2113 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2114 }
2115 else {
2116 PerlIO_printf(Perl_error_log,
2117 "perl: warning: %s %s.\n", msg, description);
2118 }
2119 }
2120 } /* End of tried to fallback */
98994639
HS
2121
2122#ifdef USE_LOCALE_CTYPE
2123 new_ctype(curctype);
2124#endif /* USE_LOCALE_CTYPE */
2125
2126#ifdef USE_LOCALE_COLLATE
2127 new_collate(curcoll);
2128#endif /* USE_LOCALE_COLLATE */
2129
2130#ifdef USE_LOCALE_NUMERIC
2131 new_numeric(curnum);
2132#endif /* USE_LOCALE_NUMERIC */
b310b053 2133
8ef6e574 2134#if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
49c85077
KW
2135 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
2136 * locale is UTF-8. If PL_utf8locale and PL_unicode (set by -C or by
2137 * $ENV{PERL_UNICODE}) are true, perl.c:S_parse_body() will turn on the
2138 * PerlIO :utf8 layer on STDIN, STDOUT, STDERR, _and_ the default open
2139 * discipline. */
c1284011 2140 PL_utf8locale = _is_cur_LC_category_utf8(LC_CTYPE);
49c85077 2141
a05d7ebb 2142 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2143 This is an alternative to using the -C command line switch
2144 (the -C if present will override this). */
2145 {
dd374669 2146 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2147 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2148 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2149 PL_utf8cache = -1;
b310b053 2150 }
ec71e770 2151#endif
b310b053 2152
98994639 2153#ifdef USE_LOCALE_CTYPE
43c5f42d 2154 Safefree(curctype);
98994639
HS
2155#endif /* USE_LOCALE_CTYPE */
2156#ifdef USE_LOCALE_COLLATE
43c5f42d 2157 Safefree(curcoll);
98994639
HS
2158#endif /* USE_LOCALE_COLLATE */
2159#ifdef USE_LOCALE_NUMERIC
43c5f42d 2160 Safefree(curnum);
98994639 2161#endif /* USE_LOCALE_NUMERIC */
8ef6e574 2162
175c4cf9
KW
2163#ifdef __GLIBC__
2164 Safefree(language);
2165#endif
2166
2167 Safefree(lc_all);
2168 Safefree(lang);
2169
e3305790
KW
2170#else /* !USE_LOCALE */
2171 PERL_UNUSED_ARG(printwarn);
2172#endif /* USE_LOCALE */
2173
2fcc0ca9
KW
2174#ifdef DEBUGGING
2175 /* So won't continue to output stuff */
27cdc72e 2176 DEBUG_INITIALIZATION_set(FALSE);
2fcc0ca9
KW
2177#endif
2178
98994639
HS
2179 return ok;
2180}
2181
98994639
HS
2182#ifdef USE_LOCALE_COLLATE
2183
a4a439fb 2184char *
a4a439fb
KW
2185Perl__mem_collxfrm(pTHX_ const char *input_string,
2186 STRLEN len, /* Length of 'input_string' */
2187 STRLEN *xlen, /* Set to length of returned string
2188 (not including the collation index
2189 prefix) */
2190 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2191 )
98994639 2192{
a4a439fb
KW
2193
2194 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2195 * differences. First, it handles embedded NULs. Second, it allocates a bit
2196 * more memory than needed for the transformed data itself. The real
55e5378d 2197 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2198 * the length of that, and doesn't include the collation index size.
2199 * Please see sv_collxfrm() to see how this is used. */
2200
55e5378d
KW
2201#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2202
6696cfa7
KW
2203 char * s = (char *) input_string;
2204 STRLEN s_strlen = strlen(input_string);
79f120c8 2205 char *xbuf = NULL;
55e5378d 2206 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2207 STRLEN length_in_chars;
c664130f 2208 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2209
a4a439fb
KW
2210 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2211
2212 /* Must be NUL-terminated */
2213 assert(*(input_string + len) == '\0');
7918f24d 2214
79f120c8
KW
2215 /* If this locale has defective collation, skip */
2216 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2217 DEBUG_L(PerlIO_printf(Perl_debug_log,
2218 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2219 goto bad;
2220 }
2221
6696cfa7
KW
2222 /* Replace any embedded NULs with the control that sorts before any others.
2223 * This will give as good as possible results on strings that don't
2224 * otherwise contain that character, but otherwise there may be
2225 * less-than-perfect results with that character and NUL. This is
fdc080f3 2226 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2227 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2228 NUL */
6696cfa7
KW
2229 char * e = s + len;
2230 char * sans_nuls;
fdc080f3 2231 STRLEN sans_nuls_len;
94762aa0 2232 int try_non_controls;
afc4976f
KW
2233 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2234 making sure 2nd byte is NUL.
2235 */
2236 STRLEN this_replacement_len;
2237
1e4c9676
KW
2238 /* If we don't know what non-NUL control character sorts lowest for
2239 * this locale, find it */
f28f4d2a 2240 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2241 int j;
afc4976f 2242 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2243 includes the collation index
2244 prefixed. */
2245
91c0e2e0 2246 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2247
2248 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2249 * NUL, in which case we instead look for any character. Controls
2250 * are preferred because collation order is, in general, context
2251 * sensitive, with adjoining characters affecting the order, and
2252 * controls are less likely to have such interactions, allowing the
2253 * NUL-replacement to stand on its own. (Another way to look at it
2254 * is to imagine what would happen if the NUL were replaced by a
2255 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2256 for (try_non_controls = 0;
2257 try_non_controls < 2;
2258 try_non_controls++)
2259 {
d4ff9586
KW
2260 /* Look through all legal code points (NUL isn't) */
2261 for (j = 1; j < 256; j++) {
2262 char * x; /* j's xfrm plus collation index */
2263 STRLEN x_len; /* length of 'x' */
2264 STRLEN trial_len = 1;
736a4fed 2265 char cur_source[] = { '\0', '\0' };
d4ff9586 2266
736a4fed
KW
2267 /* Skip non-controls the first time through the loop. The
2268 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2269 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2270 ? ! isCNTRL_L1(j)
2271 : ! isCNTRL_LC(j))
2272 {
d4ff9586 2273 continue;
6696cfa7 2274 }
6696cfa7 2275
736a4fed
KW
2276 /* Create a 1-char string of the current code point */
2277 cur_source[0] = (char) j;
2278
d4ff9586
KW
2279 /* Then transform it */
2280 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2281 0 /* The string is not in UTF-8 */);
6696cfa7 2282
1e4c9676 2283 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2284 * */
2285 if (! x) {
2286 continue;
2287 }
6696cfa7 2288
d4ff9586
KW
2289 /* If this character's transformation is lower than
2290 * the current lowest, this one becomes the lowest */
2291 if ( cur_min_x == NULL
2292 || strLT(x + COLLXFRM_HDR_LEN,
2293 cur_min_x + COLLXFRM_HDR_LEN))
2294 {
f28f4d2a 2295 PL_strxfrm_NUL_replacement = j;
d4ff9586 2296 cur_min_x = x;
d4ff9586
KW
2297 }
2298 else {
2299 Safefree(x);
2300 }
1e4c9676 2301 } /* end of loop through all 255 characters */
6696cfa7 2302
1e4c9676 2303 /* Stop looking if found */
94762aa0
KW
2304 if (cur_min_x) {
2305 break;
2306 }
2307
2308 /* Unlikely, but possible, if there aren't any controls that
2309 * work in the locale, repeat the loop, looking for any
2310 * character that works */
2311 DEBUG_L(PerlIO_printf(Perl_debug_log,
2312 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2313 } /* End of loop to try first the controls, then any char */
6696cfa7 2314
94762aa0
KW
2315 if (! cur_min_x) {
2316 DEBUG_L(PerlIO_printf(Perl_debug_log,
2317 "_mem_collxfrm: Couldn't find any character to replace"
2318 " embedded NULs in locale %s with", PL_collation_name));
2319 goto bad;
58eebef2
KW
2320 }
2321
94762aa0
KW
2322 DEBUG_L(PerlIO_printf(Perl_debug_log,
2323 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2324 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2325
6696cfa7 2326 Safefree(cur_min_x);
1e4c9676 2327 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2328
2329 /* If the replacement is variant under UTF-8, it must match the
2330 * UTF8-ness as the original */
f28f4d2a
KW
2331 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2332 this_replacement_char[0] =
2333 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2334 this_replacement_char[1] =
2335 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2336 this_replacement_len = 2;
2337 }
2338 else {
f28f4d2a 2339 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2340 /* this_replacement_char[1] = '\0' was done at initialization */
2341 this_replacement_len = 1;
6696cfa7
KW
2342 }
2343
2344 /* The worst case length for the replaced string would be if every
2345 * character in it is NUL. Multiply that by the length of each
2346 * replacement, and allow for a trailing NUL */
afc4976f 2347 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2348 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2349 *sans_nuls = '\0';
2350
6696cfa7
KW
2351 /* Replace each NUL with the lowest collating control. Loop until have
2352 * exhausted all the NULs */
2353 while (s + s_strlen < e) {
6069d6c5 2354 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2355
2356 /* Do the actual replacement */
6069d6c5 2357 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2358
2359 /* Move past the input NUL */
2360 s += s_strlen + 1;
2361 s_strlen = strlen(s);
2362 }
2363
2364 /* And add anything that trails the final NUL */
6069d6c5 2365 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2366
2367 /* Switch so below we transform this modified string */
2368 s = sans_nuls;
2369 len = strlen(s);
1e4c9676 2370 } /* End of replacing NULs */
6696cfa7 2371
a4a439fb
KW
2372 /* Make sure the UTF8ness of the string and locale match */
2373 if (utf8 != PL_in_utf8_COLLATE_locale) {
2374 const char * const t = s; /* Temporary so we can later find where the
2375 input was */
2376
2377 /* Here they don't match. Change the string's to be what the locale is
2378 * expecting */
2379
2380 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2381 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2382 utf8 = TRUE;
2383 }
2384 else { /* locale is not UTF-8; but input is; downgrade the input */
2385
2386 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2387
2388 /* If the downgrade was successful we are done, but if the input
2389 * contains things that require UTF-8 to represent, have to do
2390 * damage control ... */
2391 if (UNLIKELY(utf8)) {
2392
2393 /* What we do is construct a non-UTF-8 string with
2394 * 1) the characters representable by a single byte converted
2395 * to be so (if necessary);
2396 * 2) and the rest converted to collate the same as the
2397 * highest collating representable character. That makes
2398 * them collate at the end. This is similar to how we
2399 * handle embedded NULs, but we use the highest collating
2400 * code point instead of the smallest. Like the NUL case,
2401 * this isn't perfect, but is the best we can reasonably
2402 * do. Every above-255 code point will sort the same as
2403 * the highest-sorting 0-255 code point. If that code
2404 * point can combine in a sequence with some other code
2405 * points for weight calculations, us changing something to
2406 * be it can adversely affect the results. But in most
2407 * cases, it should work reasonably. And note that this is
2408 * really an illegal situation: using code points above 255
2409 * on a locale where only 0-255 are valid. If two strings
2410 * sort entirely equal, then the sort order for the
2411 * above-255 code points will be in code point order. */
2412
2413 utf8 = FALSE;
2414
2415 /* If we haven't calculated the code point with the maximum
2416 * collating order for this locale, do so now */
2417 if (! PL_strxfrm_max_cp) {
2418 int j;
2419
2420 /* The current transformed string that collates the
2421 * highest (except it also includes the prefixed collation
2422 * index. */
2423 char * cur_max_x = NULL;
2424
2425 /* Look through all legal code points (NUL isn't) */
2426 for (j = 1; j < 256; j++) {
2427 char * x;
2428 STRLEN x_len;
736a4fed 2429 char cur_source[] = { '\0', '\0' };
a4a439fb 2430
736a4fed
KW
2431 /* Create a 1-char string of the current code point */
2432 cur_source[0] = (char) j;
a4a439fb
KW
2433
2434 /* Then transform it */
2435 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2436
2437 /* If something went wrong (which it shouldn't), just
2438 * ignore this code point */
94762aa0 2439 if (! x) {
a4a439fb
KW
2440 continue;
2441 }
2442
2443 /* If this character's transformation is higher than
2444 * the current highest, this one becomes the highest */
2445 if ( cur_max_x == NULL
55e5378d
KW
2446 || strGT(x + COLLXFRM_HDR_LEN,
2447 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2448 {
2449 PL_strxfrm_max_cp = j;
2450 cur_max_x = x;
2451 }
2452 else {
2453 Safefree(x);
2454 }
2455 }
2456
94762aa0
KW
2457 if (! cur_max_x) {
2458 DEBUG_L(PerlIO_printf(Perl_debug_log,
2459 "_mem_collxfrm: Couldn't find any character to"
2460 " replace above-Latin1 chars in locale %s with",
2461 PL_collation_name));
2462 goto bad;
2463 }
2464
58eebef2
KW
2465 DEBUG_L(PerlIO_printf(Perl_debug_log,
2466 "_mem_collxfrm: highest 1-byte collating character"
2467 " in locale %s is 0x%02X\n",
2468 PL_collation_name,
2469 PL_strxfrm_max_cp));
58eebef2 2470
a4a439fb
KW
2471 Safefree(cur_max_x);
2472 }
2473
2474 /* Here we know which legal code point collates the highest.
2475 * We are ready to construct the non-UTF-8 string. The length
2476 * will be at least 1 byte smaller than the input string
2477 * (because we changed at least one 2-byte character into a
2478 * single byte), but that is eaten up by the trailing NUL */
2479 Newx(s, len, char);
2480
2481 {
2482 STRLEN i;
2483 STRLEN d= 0;
042d9e50 2484 char * e = (char *) t + len;
a4a439fb
KW
2485
2486 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2487 U8 cur_char = t[i];
2488 if (UTF8_IS_INVARIANT(cur_char)) {
2489 s[d++] = cur_char;
2490 }
042d9e50 2491 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2492 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2493 }
2494 else { /* Replace illegal cp with highest collating
2495 one */
2496 s[d++] = PL_strxfrm_max_cp;
2497 }
2498 }
2499 s[d++] = '\0';
2500 Renew(s, d, char); /* Free up unused space */
2501 }
2502 }
2503 }
2504
2505 /* Here, we have constructed a modified version of the input. It could
2506 * be that we already had a modified copy before we did this version.
2507 * If so, that copy is no longer needed */
2508 if (t != input_string) {
2509 Safefree(t);
2510 }
2511 }
2512
17f41037
KW
2513 length_in_chars = (utf8)
2514 ? utf8_length((U8 *) s, (U8 *) s + len)
2515 : len;
2516
59c018b9
KW
2517 /* The first element in the output is the collation id, used by
2518 * sv_collxfrm(); then comes the space for the transformed string. The
2519 * equation should give us a good estimate as to how much is needed */
55e5378d 2520 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2521 + PL_collxfrm_base
17f41037 2522 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2523 Newx(xbuf, xAlloc, char);
c7202dee
KW
2524 if (UNLIKELY(! xbuf)) {
2525 DEBUG_L(PerlIO_printf(Perl_debug_log,
2526 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2527 goto bad;
c7202dee 2528 }
98994639 2529
d35fca5f 2530 /* Store the collation id */
98994639 2531 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2532
2533 /* Then the transformation of the input. We loop until successful, or we
2534 * give up */
4ebeff16 2535 for (;;) {
1adab0a7 2536
55e5378d 2537 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2538
2539 /* If the transformed string occupies less space than we told strxfrm()
2540 * was available, it means it successfully transformed the whole
2541 * string. */
55e5378d 2542 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2543
1adab0a7
KW
2544 /* Some systems include a trailing NUL in the returned length.
2545 * Ignore it, using a loop in case multiple trailing NULs are
2546 * returned. */
2547 while ( (*xlen) > 0
2548 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2549 {
2550 (*xlen)--;
2551 }
2552
17f41037
KW
2553 /* If the first try didn't get it, it means our prediction was low.
2554 * Modify the coefficients so that we predict a larger value in any
2555 * future transformations */
2556 if (! first_time) {
2557 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2558 STRLEN computed_guess = PL_collxfrm_base
2559 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2560
2561 /* On zero-length input, just keep current slope instead of
2562 * dividing by 0 */
2563 const STRLEN new_m = (length_in_chars != 0)
2564 ? needed / length_in_chars
2565 : PL_collxfrm_mult;
17f41037
KW
2566
2567 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2568 "%s: %d: initial size of %zu bytes for a length "
2569 "%zu string was insufficient, %zu needed\n",
17f41037 2570 __FILE__, __LINE__,
b07929e4 2571 computed_guess, length_in_chars, needed));
17f41037
KW
2572
2573 /* If slope increased, use it, but discard this result for
2574 * length 1 strings, as we can't be sure that it's a real slope
2575 * change */
2576 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
2577#ifdef DEBUGGING
2578 STRLEN old_m = PL_collxfrm_mult;
2579 STRLEN old_b = PL_collxfrm_base;
2580#endif
2581 PL_collxfrm_mult = new_m;
2582 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2583 computed_guess = PL_collxfrm_base
2584 + (PL_collxfrm_mult * length_in_chars);
2585 if (computed_guess < needed) {
2586 PL_collxfrm_base += needed - computed_guess;
2587 }
2588
2589 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2590 "%s: %d: slope is now %zu; was %zu, base "
2591 "is now %zu; was %zu\n",
17f41037 2592 __FILE__, __LINE__,
b07929e4
KW
2593 PL_collxfrm_mult, old_m,
2594 PL_collxfrm_base, old_b));
17f41037
KW
2595 }
2596 else { /* Slope didn't change, but 'b' did */
2597 const STRLEN new_b = needed
2598 - computed_guess
2599 + PL_collxfrm_base;
2600 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2601 "%s: %d: base is now %zu; was %zu\n",
17f41037 2602 __FILE__, __LINE__,
b07929e4 2603 new_b, PL_collxfrm_base));
17f41037
KW
2604 PL_collxfrm_base = new_b;
2605 }
2606 }
2607
4ebeff16
KW
2608 break;
2609 }
bb0f664e 2610
c7202dee
KW
2611 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2612 DEBUG_L(PerlIO_printf(Perl_debug_log,
2613 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2614 *xlen, PERL_INT_MAX));
4ebeff16 2615 goto bad;
c7202dee 2616 }
d35fca5f 2617
c664130f 2618 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
2619 * (usually not including the trailing NUL) when it fails due to not
2620 * enough space being provided. Assume that this is the case unless
2621 * it's been proven otherwise */
c664130f 2622 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 2623 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
2624 }
2625 else { /* Here, either:
2626 * 1) The strxfrm() has previously shown bad behavior; or
2627 * 2) It isn't the first time through the loop, which means
2628 * that the strxfrm() is now showing bad behavior, because
2629 * we gave it what it said was needed in the previous
2630 * iteration, and it came back saying it needed still more.
2631 * (Many versions of cygwin fit this. When the buffer size
2632 * isn't sufficient, they return the input size instead of
2633 * how much is needed.)
d4ff9586
KW
2634 * Increase the buffer size by a fixed percentage and try again.
2635 * */
6ddd902c 2636 xAlloc += (xAlloc / 4) + 1;
c664130f 2637 PL_strxfrm_is_behaved = FALSE;
c664130f 2638
58eebef2
KW
2639#ifdef DEBUGGING
2640 if (DEBUG_Lv_TEST || debug_initialization) {
2641 PerlIO_printf(Perl_debug_log,
2642 "_mem_collxfrm required more space than previously calculated"
b07929e4 2643 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 2644 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 2645 xAlloc - COLLXFRM_HDR_LEN);
58eebef2
KW
2646 }
2647#endif
2648 }
c664130f 2649
4ebeff16 2650 Renew(xbuf, xAlloc, char);
c7202dee
KW
2651 if (UNLIKELY(! xbuf)) {
2652 DEBUG_L(PerlIO_printf(Perl_debug_log,
2653 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 2654 goto bad;
c7202dee 2655 }
c664130f
KW
2656
2657 first_time = FALSE;
4ebeff16 2658 }
98994639 2659
6696cfa7 2660
58eebef2
KW
2661#ifdef DEBUGGING
2662 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
2663
2664 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
2665 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
2666 PerlIO_printf(Perl_debug_log, "%s\n",
2667 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
2668 *xlen, 1));
58eebef2
KW
2669 }
2670#endif
2671
3c5f993e 2672 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 2673 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 2674
6696cfa7
KW
2675 if (s != input_string) {
2676 Safefree(s);
98994639
HS
2677 }
2678
98994639
HS
2679 return xbuf;
2680
2681 bad:
2682 Safefree(xbuf);
6696cfa7
KW
2683 if (s != input_string) {
2684 Safefree(s);
2685 }
98994639 2686 *xlen = 0;
58eebef2
KW
2687#ifdef DEBUGGING
2688 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 2689 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2
KW
2690 }
2691#endif
98994639
HS
2692 return NULL;
2693}
2694
c7202dee
KW
2695#ifdef DEBUGGING
2696
4cbaac56 2697STATIC void
c7202dee
KW
2698S_print_collxfrm_input_and_return(pTHX_
2699 const char * const s,
2700 const char * const e,
2701 const STRLEN * const xlen,
2702 const bool is_utf8)
2703{
c7202dee
KW
2704
2705 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
2706
511e4ff7
DM
2707 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
2708 (UV)PL_collation_ix);
c7202dee 2709 if (xlen) {
08b6dc1d 2710 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
2711 }
2712 else {
2713 PerlIO_printf(Perl_debug_log, "NULL");
2714 }
2715 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
2716 PL_collation_name);
9c8a6dc2
KW
2717 print_bytes_for_locale(s, e, is_utf8);
2718
2719 PerlIO_printf(Perl_debug_log, "'\n");
2720}
2721
2722STATIC void
2723S_print_bytes_for_locale(pTHX_
2724 const char * const s,
2725 const char * const e,
2726 const bool is_utf8)
2727{
2728 const char * t = s;
2729 bool prev_was_printable = TRUE;
2730 bool first_time = TRUE;
2731
2732 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
2733
2734 while (t < e) {
2735 UV cp = (is_utf8)
2736 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
2737 : * (U8 *) t;
2738 if (isPRINT(cp)) {
2739 if (! prev_was_printable) {
2740 PerlIO_printf(Perl_debug_log, " ");
2741 }
2742 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
2743 prev_was_printable = TRUE;
2744 }
2745 else {
2746 if (! first_time) {
2747 PerlIO_printf(Perl_debug_log, " ");
2748 }
147e3846 2749 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
2750 prev_was_printable = FALSE;
2751 }
2752 t += (is_utf8) ? UTF8SKIP(t) : 1;
2753 first_time = FALSE;
2754 }
c7202dee
KW
2755}
2756
2757#endif /* #ifdef DEBUGGING */
2758
98994639 2759#endif /* USE_LOCALE_COLLATE */
58eebef2 2760
8ef6e574
KW
2761#ifdef USE_LOCALE
2762
c1284011
KW
2763bool
2764Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
2765{
2766 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
2767 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 2768 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
2769 * could give the wrong result. The result will very likely be correct for
2770 * languages that have commonly used non-ASCII characters, but for notably
2771 * English, it comes down to if the locale's name ends in something like
2772 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61
KW
2773
2774 char *save_input_locale = NULL;
7d74bb61
KW
2775 STRLEN final_pos;
2776
8ef6e574 2777#ifdef LC_ALL
7d74bb61 2778 assert(category != LC_ALL);
8ef6e574 2779#endif
7d74bb61
KW
2780
2781 /* First dispose of the trivial cases */
b07fffd1 2782 save_input_locale = setlocale(category, NULL);
7d74bb61 2783 if (! save_input_locale) {
69014004
KW
2784 DEBUG_L(PerlIO_printf(Perl_debug_log,
2785 "Could not find current locale for category %d\n",
2786 category));
7d74bb61
KW
2787 return FALSE; /* XXX maybe should croak */
2788 }
b07fffd1 2789 save_input_locale = stdize_locale(savepv(save_input_locale));
a39edc4c 2790 if (isNAME_C_OR_POSIX(save_input_locale)) {
69014004
KW
2791 DEBUG_L(PerlIO_printf(Perl_debug_log,
2792 "Current locale for category %d is %s\n",
2793 category, save_input_locale));
b07fffd1 2794 Safefree(save_input_locale);
7d74bb61
KW
2795 return FALSE;
2796 }
2797
1d958db2
KW
2798#if defined(USE_LOCALE_CTYPE) \
2799 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 2800
1d958db2 2801 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
2802
2803 char *save_ctype_locale = NULL;
119ee68b 2804 bool is_utf8;
7d74bb61 2805
119ee68b 2806 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
2807
2808 /* Get the current LC_CTYPE locale */
4f72bb37 2809 save_ctype_locale = setlocale(LC_CTYPE, NULL);
7d74bb61 2810 if (! save_ctype_locale) {
69014004
KW
2811 DEBUG_L(PerlIO_printf(Perl_debug_log,
2812 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
2813 goto cant_use_nllanginfo;
2814 }
4f72bb37 2815 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
2816
2817 /* If LC_CTYPE and the desired category use the same locale, this
2818 * means that finding the value for LC_CTYPE is the same as finding
2819 * the value for the desired category. Otherwise, switch LC_CTYPE
2820 * to the desired category's locale */
2821 if (strEQ(save_ctype_locale, save_input_locale)) {
2822 Safefree(save_ctype_locale);
2823 save_ctype_locale = NULL;
2824 }
2825 else if (! setlocale(LC_CTYPE, save_input_locale)) {
69014004
KW
2826 DEBUG_L(PerlIO_printf(Perl_debug_log,
2827 "Could not change LC_CTYPE locale to %s\n",
2828 save_input_locale));
7d74bb61
KW
2829 Safefree(save_ctype_locale);
2830 goto cant_use_nllanginfo;
2831 }
2832 }
2833
69014004
KW
2834 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
2835 save_input_locale));
2836
7d74bb61 2837 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
2838 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
2839 * should give the correct results */
119ee68b 2840
1d958db2
KW
2841# if defined(HAS_NL_LANGINFO) && defined(CODESET)
2842 {
4f72bb37 2843 char *codeset = nl_langinfo(CODESET);
1d958db2 2844 if (codeset && strNE(codeset, "")) {
4f72bb37 2845 codeset = savepv(codeset);
119ee68b 2846
1d958db2
KW
2847 /* If we switched LC_CTYPE, switch back */
2848 if (save_ctype_locale) {
2849 setlocale(LC_CTYPE, save_ctype_locale);
2850 Safefree(save_ctype_locale);
2851 }
2852
2853 is_utf8 = foldEQ(codeset, STR_WITH_LEN("UTF-8"))
2854 || foldEQ(codeset, STR_WITH_LEN("UTF8"));
2855
69014004
KW
2856 DEBUG_L(PerlIO_printf(Perl_debug_log,
2857 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
2858 codeset, is_utf8));
1d958db2
KW
2859 Safefree(codeset);
2860 Safefree(save_input_locale);
2861 return is_utf8;
2862 }
119ee68b
KW
2863 }
2864
1d958db2
KW
2865# endif
2866# ifdef MB_CUR_MAX
2867
2868 /* Here, either we don't have nl_langinfo, or it didn't return a
2869 * codeset. Try MB_CUR_MAX */
2870
119ee68b
KW
2871 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
2872 * Unicode code point. Since UTF-8 is the only non-single byte
2873 * encoding we handle, we just say any such encoding is UTF-8, and if
2874 * turns out to be wrong, other things will fail */
2875 is_utf8 = MB_CUR_MAX >= 4;
2876
69014004
KW
2877 DEBUG_L(PerlIO_printf(Perl_debug_log,
2878 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
2879 (int) MB_CUR_MAX, is_utf8));
2880
119ee68b
KW
2881 Safefree(save_input_locale);
2882
2883# ifdef HAS_MBTOWC
2884
2885 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
2886 * since they are both in the C99 standard. We can feed a known byte
2887 * string to the latter function, and check that it gives the expected
2888 * result */
2889 if (is_utf8) {
2890 wchar_t wc;
856b881c 2891 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 2892 errno = 0;
f019f68f 2893 if ((size_t)mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8))
119ee68b
KW
2894 != strlen(HYPHEN_UTF8)
2895 || wc != (wchar_t) 0x2010)
2896 {
2897 is_utf8 = FALSE;
abdcbdb8 2898 DEBUG_L(PerlIO_printf(Perl_debug_log, "\thyphen=U+%x\n", (unsigned int)wc));
69014004
KW
2899 DEBUG_L(PerlIO_printf(Perl_debug_log,
2900 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
2901 mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8)), errno));
119ee68b
KW
2902 }
2903 }
119ee68b
KW
2904# endif
2905
1d958db2
KW
2906 /* If we switched LC_CTYPE, switch back */
2907 if (save_ctype_locale) {
2908 setlocale(LC_CTYPE, save_ctype_locale);
2909 Safefree(save_ctype_locale);
119ee68b 2910 }
7d74bb61 2911
1d958db2 2912 return is_utf8;
119ee68b 2913# endif
7d74bb61 2914 }
119ee68b 2915
7d74bb61
KW
2916 cant_use_nllanginfo:
2917
0080c90a
KW
2918#else /* nl_langinfo should work if available, so don't bother compiling this
2919 fallback code. The final fallback of looking at the name is
2920 compiled, and will be executed if nl_langinfo fails */
7d74bb61 2921
97f4de96
KW
2922 /* nl_langinfo not available or failed somehow. Next try looking at the
2923 * currency symbol to see if it disambiguates things. Often that will be
2924 * in the native script, and if the symbol isn't in UTF-8, we know that the
2925 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
2926 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
2927 * */
fa9b773e
KW
2928
2929#ifdef HAS_LOCALECONV
fa9b773e 2930# ifdef USE_LOCALE_MONETARY
fa9b773e
KW
2931 {
2932 char *save_monetary_locale = NULL;
fa9b773e 2933 bool only_ascii = FALSE;
13542a67
KW
2934 bool is_utf8 = FALSE;
2935 struct lconv* lc;
fa9b773e 2936
97f4de96
KW
2937 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
2938 * the desired category, if it isn't that locale already */
2939
fa9b773e
KW
2940 if (category != LC_MONETARY) {
2941
4f72bb37 2942 save_monetary_locale = setlocale(LC_MONETARY, NULL);
fa9b773e 2943 if (! save_monetary_locale) {
69014004
KW
2944 DEBUG_L(PerlIO_printf(Perl_debug_log,
2945 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
2946 goto cant_use_monetary;
2947 }
4f72bb37 2948 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 2949
13542a67
KW
2950 if (strEQ(save_monetary_locale, save_input_locale)) {
2951 Safefree(save_monetary_locale);
2952 save_monetary_locale = NULL;
2953 }
2954 else if (! setlocale(LC_MONETARY, save_input_locale)) {
59c234b4
KW
2955 DEBUG_L(PerlIO_printf(Perl_debug_log,
2956 "Could not change LC_MONETARY locale to %s\n",
2957 save_input_locale));
2958 Safefree(save_monetary_locale);
2959 goto cant_use_monetary;
fa9b773e
KW
2960 }
2961 }
2962
2963 /* Here the current LC_MONETARY is set to the locale of the category
2964 * whose information is desired. */
2965
13542a67
KW
2966 lc = localeconv();
2967 if (! lc
2968 || ! lc->currency_symbol
c5f058df 2969 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
2970 {
2971 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));
2972 only_ascii = TRUE;
2973 }
2974 else {
2975 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
2976 }
2977
2978 /* If we changed it, restore LC_MONETARY to its original locale */
2979 if (save_monetary_locale) {
2980 setlocale(LC_MONETARY, save_monetary_locale);
2981 Safefree(save_monetary_locale);
2982 }
2983
13542a67 2984 if (! only_ascii) {
fa9b773e 2985
59c234b4
KW
2986 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
2987 * otherwise assume the locale is UTF-8 if and only if the symbol
2988 * is non-ascii UTF-8. */
2989 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
2990 save_input_locale, is_utf8));
2991 Safefree(save_input_locale);
2992 return is_utf8;
13542a67 2993 }
fa9b773e
KW
2994 }
2995 cant_use_monetary:
2996
2997# endif /* USE_LOCALE_MONETARY */
2998#endif /* HAS_LOCALECONV */
2999
15f7e74e
KW
3000#if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
3001
3002/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3003 * the names of the months and weekdays, timezone, and am/pm indicator */
3004 {
3005 char *save_time_locale = NULL;
3006 int hour = 10;
3007 bool is_dst = FALSE;
3008 int dom = 1;
3009 int month = 0;
3010 int i;
3011 char * formatted_time;
3012
3013
3014 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
3015 * desired category, if it isn't that locale already */
3016
3017 if (category != LC_TIME) {
3018
3019 save_time_locale = setlocale(LC_TIME, NULL);
3020 if (! save_time_locale) {
3021 DEBUG_L(PerlIO_printf(Perl_debug_log,
3022 "Could not find current locale for LC_TIME\n"));
3023 goto cant_use_time;
3024 }
3025 save_time_locale = stdize_locale(savepv(save_time_locale));
3026
3027 if (strEQ(save_time_locale, save_input_locale)) {
3028 Safefree(save_time_locale);
3029 save_time_locale = NULL;
3030 }
3031 else if (! setlocale(LC_TIME, save_input_locale)) {
3032 DEBUG_L(PerlIO_printf(Perl_debug_log,
3033 "Could not change LC_TIME locale to %s\n",
3034 save_input_locale));
3035 Safefree(save_time_locale);
3036 goto cant_use_time;
3037 }
3038 }
3039
3040 /* Here the current LC_TIME is set to the locale of the category
3041 * whose information is desired. Look at all the days of the week and
9f10db87 3042 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
3043 * characters. The first such a one found will tell us if the locale
3044 * is UTF-8 or not */
3045
3046 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3047 formatted_time = my_strftime("%A %B %Z %p",
3ae5cd07 3048 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
c5f058df
KW
3049 if ( ! formatted_time
3050 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3051 {
15f7e74e
KW
3052
3053 /* Here, we didn't find a non-ASCII. Try the next time through
3054 * with the complemented dst and am/pm, and try with the next
3055 * weekday. After we have gotten all weekdays, try the next
3056 * month */
3057 is_dst = ! is_dst;
3058 hour = (hour + 12) % 24;
3059 dom++;
3060 if (i > 6) {
3061 month++;
3062 }
3063 continue;
3064 }
3065
3066 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3067 * false otherwise. But first, restore LC_TIME to its original
3068 * locale if we changed it */
3069 if (save_time_locale) {
3070 setlocale(LC_TIME, save_time_locale);
3071 Safefree(save_time_locale);
3072 }
3073
3074 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3075 save_input_locale,
3076 is_utf8_string((U8 *) formatted_time, 0)));
3077 Safefree(save_input_locale);
3078 return is_utf8_string((U8 *) formatted_time, 0);
3079 }
3080
3081 /* Falling off the end of the loop indicates all the names were just
3082 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3083 * to its original locale */
3084 if (save_time_locale) {
3085 setlocale(LC_TIME, save_time_locale);
3086 Safefree(save_time_locale);
3087 }
3088 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));
3089 }
3090 cant_use_time:
3091
3092#endif
3093
3094#if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
3095
3096/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
3097 * on our dromedary test machine, which has over 700 locales. There, this
3098 * added no value to looking at the currency symbol and the time strings. I
3099 * left it in so as to avoid rewriting it if real-world experience indicates
3100 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
3101 * haven't found illegal utf8, we continue on and examine all the strerror()
3102 * messages on the platform for utf8ness. If all are ASCII, we still don't
3103 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
3104 * utf8ness. The reason this doesn't help much is that the messages may not
3105 * have been translated into the locale. The currency symbol and time strings
3106 * are much more likely to have been translated. */
3107 {
855aeb93 3108 int e;
5857e934
KW
3109 bool is_utf8 = FALSE;
3110 bool non_ascii = FALSE;
855aeb93 3111 char *save_messages_locale = NULL;
5857e934 3112 const char * errmsg = NULL;
855aeb93 3113
5857e934
KW
3114 /* Like above, we set LC_MESSAGES to the locale of the desired
3115 * category, if it isn't that locale already */
855aeb93
JH
3116
3117 if (category != LC_MESSAGES) {
3118
5857e934 3119 save_messages_locale = setlocale(LC_MESSAGES, NULL);
855aeb93 3120 if (! save_messages_locale) {
5857e934
KW
3121 DEBUG_L(PerlIO_printf(Perl_debug_log,
3122 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
3123 goto cant_use_messages;
3124 }
5857e934 3125 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
3126
3127 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
3128 Safefree(save_messages_locale);
3129 save_messages_locale = NULL;
855aeb93
JH
3130 }
3131 else if (! setlocale(LC_MESSAGES, save_input_locale)) {
5857e934
KW
3132 DEBUG_L(PerlIO_printf(Perl_debug_log,
3133 "Could not change LC_MESSAGES locale to %s\n",
3134 save_input_locale));
855aeb93
JH
3135 Safefree(save_messages_locale);
3136 goto cant_use_messages;
3137 }
3138 }
3139
3140 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
3141 * whose information is desired. Look through all the messages. We
3142 * can't use Strerror() here because it may expand to code that
3143 * segfaults in miniperl */
855aeb93 3144
5857e934
KW
3145 for (e = 0; e <= sys_nerr; e++) {
3146 errno = 0;
3147 errmsg = sys_errlist[e];
3148 if (errno || !errmsg) {
855aeb93
JH
3149 break;
3150 }
5857e934 3151 errmsg = savepv(errmsg);
c5f058df 3152 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
3153 non_ascii = TRUE;
3154 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3155 break;
855aeb93
JH
3156 }
3157 }
5857e934 3158 Safefree(errmsg);
855aeb93
JH
3159
3160 /* And, if we changed it, restore LC_MESSAGES to its original locale */
3161 if (save_messages_locale) {
3162 setlocale(LC_MESSAGES, save_messages_locale);
3163 Safefree(save_messages_locale);
3164 }
3165
5857e934
KW
3166 if (non_ascii) {
3167
3168 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3169 * any non-ascii means it is one; otherwise we assume it isn't */
3170 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3171 save_input_locale,
3172 is_utf8));
3173 Safefree(save_input_locale);
3174 return is_utf8;
3175 }
855aeb93 3176
5857e934 3177 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
3178 }
3179 cant_use_messages:
3180
3181#endif
fa9b773e 3182
0080c90a
KW
3183#endif /* the code that is compiled when no nl_langinfo */
3184
92c0a900
KW
3185#ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
3186 UTF-8 locale */
97f4de96
KW
3187 /* As a last resort, look at the locale name to see if it matches
3188 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3189 * return of setlocale(), is actually defined to be opaque, so we can't
3190 * really rely on the absence of various substrings in the name to indicate
3191 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3192 * be a UTF-8 locale. Similarly for the other common names */
3193
3194 final_pos = strlen(save_input_locale) - 1;
3195 if (final_pos >= 3) {
3196 char *name = save_input_locale;
3197
3198 /* Find next 'U' or 'u' and look from there */
3199 while ((name += strcspn(name, "Uu") + 1)
3200 <= save_input_locale + final_pos - 2)
3201 {
305b8651
KW
3202 if (!isALPHA_FOLD_NE(*name, 't')
3203 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
3204 {
3205 continue;
3206 }
3207 name += 2;
3208 if (*(name) == '-') {
3209 if ((name > save_input_locale + final_pos - 1)) {
3210 break;
3211 }
3212 name++;
3213 }
3214 if (*(name) == '8') {
97f4de96
KW
3215 DEBUG_L(PerlIO_printf(Perl_debug_log,
3216 "Locale %s ends with UTF-8 in name\n",
3217 save_input_locale));
00c54b9c 3218 Safefree(save_input_locale);
97f4de96
KW
3219 return TRUE;
3220 }
3221 }
3222 DEBUG_L(PerlIO_printf(Perl_debug_log,
3223 "Locale %s doesn't end with UTF-8 in name\n",
3224 save_input_locale));
3225 }
92c0a900 3226#endif
97f4de96
KW
3227
3228#ifdef WIN32
3229 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
3230 if (final_pos >= 4
3231 && *(save_input_locale + final_pos - 0) == '1'
3232 && *(save_input_locale + final_pos - 1) == '0'
3233 && *(save_input_locale + final_pos - 2) == '0'
3234 && *(save_input_locale + final_pos - 3) == '5'
3235 && *(save_input_locale + final_pos - 4) == '6')
3236 {
3237 DEBUG_L(PerlIO_printf(Perl_debug_log,
3238 "Locale %s ends with 10056 in name, is UTF-8 locale\n",
3239 save_input_locale));
3240 Safefree(save_input_locale);
3241 return TRUE;
3242 }
3243#endif
3244
3245 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3246 * since we are about to return FALSE anyway, there is no point in doing
3247 * this extra work */
3248#if 0
3249 if (instr(save_input_locale, "8859")) {
3250 DEBUG_L(PerlIO_printf(Perl_debug_log,
3251 "Locale %s has 8859 in name, not UTF-8 locale\n",
3252 save_input_locale));
3253 Safefree(save_input_locale);
3254 return FALSE;
3255 }
3256#endif
3257
69014004
KW
3258 DEBUG_L(PerlIO_printf(Perl_debug_log,
3259 "Assuming locale %s is not a UTF-8 locale\n",
3260 save_input_locale));
fa9b773e 3261 Safefree(save_input_locale);
7d74bb61
KW
3262 return FALSE;
3263}
3264
8ef6e574 3265#endif
7d74bb61 3266
d6ded950
KW
3267
3268bool
3269Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3270{
1a4f13e1 3271 dVAR;
d6ded950
KW
3272 /* Internal function which returns if we are in the scope of a pragma that
3273 * enables the locale category 'category'. 'compiling' should indicate if
3274 * this is during the compilation phase (TRUE) or not (FALSE). */
3275
3276 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
3277
3278 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
3279 if (! categories || categories == &PL_sv_placeholder) {
3280 return FALSE;
3281 }
3282
3283 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
3284 * a valid unsigned */
3285 assert(category >= -1);
3286 return cBOOL(SvUV(categories) & (1U << (category + 1)));
3287}
3288
2c6ee1a7 3289char *
6ebbc862
KW
3290Perl_my_strerror(pTHX_ const int errnum)
3291{
3292 /* Returns a mortalized copy of the text of the error message associated
3293 * with 'errnum'. It uses the current locale's text unless the platform
3294 * doesn't have the LC_MESSAGES category or we are not being called from
3295 * within the scope of 'use locale'. In the former case, it uses whatever
3296 * strerror returns; in the latter case it uses the text from the C locale.
3297 *
3298 * The function just calls strerror(), but temporarily switches, if needed,
3299 * to the C locale */
3300
3301 char *errstr;
52770946 3302 dVAR;
6ebbc862 3303
52770946 3304#ifndef USE_LOCALE_MESSAGES
6ebbc862 3305
52770946
KW
3306 /* If platform doesn't have messages category, we don't do any switching to
3307 * the C locale; we just use whatever strerror() returns */
3308
3309 errstr = savepv(Strerror(errnum));
3310
3311#else /* Has locale messages */
3312
3313 const bool within_locale_scope = IN_LC(LC_MESSAGES);
2c6ee1a7 3314
7aaa36b1
KW
3315# if defined(HAS_POSIX_2008_LOCALE) && defined(HAS_STRERROR_L)
3316
3317 /* This function is trivial if we have strerror_l() */
3318
3319 if (within_locale_scope) {
3320 errstr = strerror(errnum);
3321 }
3322 else {
3323 errstr = strerror_l(errnum, PL_C_locale_obj);
3324 }
3325
3326 errstr = savepv(errstr);
3327
3328# else /* Doesn't have strerror_l(). */
3329
3330# ifdef USE_POSIX_2008_LOCALE
3331
fcd0e682 3332 locale_t save_locale = NULL;
7aaa36b1
KW
3333
3334# else
3335
fcd0e682 3336 char * save_locale = NULL;
c9dda6da 3337 bool locale_is_C = FALSE;
2c6ee1a7 3338
6ebbc862
KW
3339 /* We have a critical section to prevent another thread from changing the
3340 * locale out from under us (or zapping the buffer returned from
3341 * setlocale() ) */
3342 LOCALE_LOCK;
3343
7aaa36b1 3344# endif
6ebbc862 3345
9c8a6dc2
KW
3346 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3347 "my_strerror called with errnum %d\n", errnum));
6ebbc862 3348 if (! within_locale_scope) {
c9dda6da 3349 errno = 0;
a0b53297 3350
f1d2176b 3351# ifdef USE_POSIX_2008_LOCALE /* Use the thread-safe locale functions */
6ebbc862 3352
9c8a6dc2
KW
3353 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3354 "Not within locale scope, about to call"
3355 " uselocale(0x%p)\n", PL_C_locale_obj));
6ebbc862 3356 save_locale = uselocale(PL_C_locale_obj);
c9dda6da
KW
3357 if (! save_locale) {
3358 DEBUG_L(PerlIO_printf(Perl_debug_log,
9c8a6dc2
KW
3359 "uselocale failed, errno=%d\n", errno));
3360 }
3361 else {
3362 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3363 "uselocale returned 0x%p\n", save_locale));
c9dda6da 3364 }
6ebbc862 3365
7aaa36b1 3366# else /* Not thread-safe build */
a0b53297
KW
3367
3368 save_locale = setlocale(LC_MESSAGES, NULL);
c9dda6da
KW
3369 if (! save_locale) {
3370 DEBUG_L(PerlIO_printf(Perl_debug_log,
3371 "setlocale failed, errno=%d\n", errno));
3372 }
3373 else {
3374 locale_is_C = isNAME_C_OR_POSIX(save_locale);
2c6ee1a7 3375
c9dda6da
KW
3376 /* Switch to the C locale if not already in it */
3377 if (! locale_is_C) {
2c6ee1a7 3378
c9dda6da
KW
3379 /* The setlocale() just below likely will zap 'save_locale', so
3380 * create a copy. */
3381 save_locale = savepv(save_locale);
3382 setlocale(LC_MESSAGES, "C");
3383 }
6ebbc862 3384 }
2c6ee1a7 3385
7aaa36b1 3386# endif
2c6ee1a7 3387
6ebbc862 3388 } /* end of ! within_locale_scope */
9c8a6dc2
KW
3389 else {
3390 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
3391 __FILE__, __LINE__));
3392 }
a0b53297 3393
9c8a6dc2
KW
3394 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3395 "Any locale change has been done; about to call Strerror\n"));
52770946 3396 errstr = savepv(Strerror(errnum));
6ebbc862
KW
3397
3398 if (! within_locale_scope) {
c9dda6da 3399 errno = 0;
a0b53297 3400
f1d2176b 3401# ifdef USE_POSIX_2008_LOCALE
6ebbc862 3402
9c8a6dc2
KW
3403 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3404 "%s: %d: not within locale scope, restoring the locale\n",
3405 __FILE__, __LINE__));
c9dda6da
KW
3406 if (save_locale && ! uselocale(save_locale)) {
3407 DEBUG_L(PerlIO_printf(Perl_debug_log,
3408 "uselocale restore failed, errno=%d\n", errno));
3409 }
2c6ee1a7 3410 }
6ebbc862 3411
7aaa36b1 3412# else
6ebbc862 3413
c9dda6da
KW
3414 if (save_locale && ! locale_is_C) {
3415 if (! setlocale(LC_MESSAGES, save_locale)) {
3416 DEBUG_L(PerlIO_printf(Perl_debug_log,
3417 "setlocale restore failed, errno=%d\n", errno));
3418 }
6ebbc862
KW
3419 Safefree(save_locale);
3420 }
3421 }
3422
3423 LOCALE_UNLOCK;
3424
7aaa36b1
KW
3425# endif
3426# endif /* End of doesn't have strerror_l */
52770946 3427#endif /* End of does have locale messages */
6affbbf0
KW
3428
3429#ifdef DEBUGGING
3430
3431 if (DEBUG_Lv_TEST) {
3432 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
3433 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
3434 PerlIO_printf(Perl_debug_log, "'\n");
3435 }
3436
2c6ee1a7
KW
3437#endif
3438
52770946 3439 SAVEFREEPV(errstr);
6ebbc862 3440 return errstr;
2c6ee1a7
KW
3441}
3442
66610fdd 3443/*
747c467a 3444
747c467a
KW
3445=for apidoc sync_locale
3446
3447Changing the program's locale should be avoided by XS code. Nevertheless,
3448certain non-Perl libraries called from XS, such as C<Gtk> do so. When this
3449happens, Perl needs to be told that the locale has changed. Use this function
3450to do so, before returning to Perl.
3451
3452=cut
3453*/
3454
3455void
3456Perl_sync_locale(pTHX)
3457{
3458
3459#ifdef USE_LOCALE_CTYPE
3460 new_ctype(setlocale(LC_CTYPE, NULL));
3461#endif /* USE_LOCALE_CTYPE */
3462
3463#ifdef USE_LOCALE_COLLATE
3464 new_collate(setlocale(LC_COLLATE, NULL));
3465#endif
3466
3467#ifdef USE_LOCALE_NUMERIC
3468 set_numeric_local(); /* Switch from "C" to underlying LC_NUMERIC */
3469 new_numeric(setlocale(LC_NUMERIC, NULL));
3470#endif /* USE_LOCALE_NUMERIC */
3471
3472}
3473
5d1187d1
KW
3474#if defined(DEBUGGING) && defined(USE_LOCALE)
3475
a4f00dcc
KW
3476STATIC char *
3477S_setlocale_debug_string(const int category, /* category number,
5d1187d1
KW
3478 like LC_ALL */
3479 const char* const locale, /* locale name */
3480
3481 /* return value from setlocale() when attempting to
3482 * set 'category' to 'locale' */
3483 const char* const retval)
3484{
3485 /* Returns a pointer to a NUL-terminated string in static storage with
3486 * added text about the info passed in. This is not thread safe and will
3487 * be overwritten by the next call, so this should be used just to
fa07b8e5 3488 * formulate a string to immediately print or savepv() on. */
5d1187d1 3489
398a990f
DM
3490 /* initialise to a non-null value to keep it out of BSS and so keep
3491 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
60b45a7d
KW
3492 static char ret[128] = "If you can read this, thank your buggy C"
3493 " library strlcpy(), and change your hints file"
3494 " to undef it";
fa07b8e5 3495 my_strlcpy(ret, "setlocale(", sizeof(ret));
5d1187d1
KW
3496
3497 switch (category) {
3498 default:
fa07b8e5 3499 my_snprintf(ret, sizeof(ret), "%s? %d", ret, category);
5d1187d1
KW
3500 break;
3501# ifdef LC_ALL
3502 case LC_ALL:
fa07b8e5 3503 my_strlcat(ret, "LC_ALL", sizeof(ret));
5d1187d1
KW
3504 break;
3505# endif
3506# ifdef LC_CTYPE
3507 case LC_CTYPE:
fa07b8e5 3508 my_strlcat(ret, "LC_CTYPE", sizeof(ret));
5d1187d1
KW
3509 break;
3510# endif
3511# ifdef LC_NUMERIC
3512 case LC_NUMERIC:
fa07b8e5 3513 my_strlcat(ret, "LC_NUMERIC", sizeof(ret));
5d1187d1
KW
3514 break;
3515# endif
3516# ifdef LC_COLLATE
3517 case LC_COLLATE:
fa07b8e5 3518 my_strlcat(ret, "LC_COLLATE", sizeof(ret));
5d1187d1
KW
3519 break;
3520# endif
3521# ifdef LC_TIME
3522 case LC_TIME:
fa07b8e5 3523 my_strlcat(ret, "LC_TIME", sizeof(ret));
5d1187d1
KW
3524 break;
3525# endif
3526# ifdef LC_MONETARY
3527 case LC_MONETARY:
fa07b8e5 3528 my_strlcat(ret, "LC_MONETARY", sizeof(ret));
5d1187d1
KW
3529 break;
3530# endif
3531# ifdef LC_MESSAGES
3532 case LC_MESSAGES:
fa07b8e5 3533 my_strlcat(ret, "LC_MESSAGES", sizeof(ret));
5d1187d1
KW
3534 break;
3535# endif
3536 }
3537
fa07b8e5 3538 my_strlcat(ret, ", ", sizeof(ret));
5d1187d1
KW
3539
3540 if (locale) {
fa07b8e5
KW
3541 my_strlcat(ret, "\"", sizeof(ret));
3542 my_strlcat(ret, locale, sizeof(ret));
3543 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
3544 }
3545 else {
fa07b8e5 3546 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
3547 }
3548
fa07b8e5 3549 my_strlcat(ret, ") returned ", sizeof(ret));
5d1187d1
KW
3550
3551 if (retval) {
fa07b8e5
KW
3552 my_strlcat(ret, "\"", sizeof(ret));
3553 my_strlcat(ret, retval, sizeof(ret));
3554 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
3555 }
3556 else {
fa07b8e5 3557 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
3558 }
3559
3560 assert(strlen(ret) < sizeof(ret));
3561
3562 return ret;
3563}
3564
3565#endif
747c467a
KW
3566
3567
3568/*
14d04a33 3569 * ex: set ts=8 sts=4 sw=4 et:
37442d52 3570 */