This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Add missing STATIC
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52
KW
25 *
26 * All C programs have an underlying locale. Perl generally doesn't pay any
27 * attention to it except within the scope of a 'use locale'. For most
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
166f8a29
DM
35 */
36
98994639
HS
37#include "EXTERN.h"
38#define PERL_IN_LOCALE_C
39#include "perl.h"
40
b310b053
JH
41#ifdef I_LANGINFO
42# include <langinfo.h>
43#endif
44
a4af207c
JH
45#include "reentr.h"
46
2fcc0ca9
KW
47/* If the environment says to, we can output debugging information during
48 * initialization. This is done before option parsing, and before any thread
49 * creation, so can be a file-level static */
50#ifdef DEBUGGING
27cdc72e
DM
51# ifdef PERL_GLOBAL_STRUCT
52 /* no global syms allowed */
53# define debug_initialization 0
54# define DEBUG_INITIALIZATION_set(v)
55# else
2fcc0ca9 56static bool debug_initialization = FALSE;
27cdc72e
DM
57# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
58# endif
2fcc0ca9
KW
59#endif
60
8ef6e574
KW
61#ifdef USE_LOCALE
62
98994639 63/*
0d071d52
KW
64 * Standardize the locale name from a string returned by 'setlocale', possibly
65 * modifying that string.
98994639 66 *
0ef2a2b2 67 * The typical return value of setlocale() is either
98994639
HS
68 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
69 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
70 * (the space-separated values represent the various sublocales,
0ef2a2b2 71 * in some unspecified order). This is not handled by this function.
98994639
HS
72 *
73 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
0ef2a2b2
KW
74 * which is harmful for further use of the string in setlocale(). This
75 * function removes the trailing new line and everything up through the '='
98994639
HS
76 *
77 */
78STATIC char *
79S_stdize_locale(pTHX_ char *locs)
80{
7452cf6a 81 const char * const s = strchr(locs, '=');
98994639
HS
82 bool okay = TRUE;
83
7918f24d
NC
84 PERL_ARGS_ASSERT_STDIZE_LOCALE;
85
8772537c
AL
86 if (s) {
87 const char * const t = strchr(s, '.');
98994639 88 okay = FALSE;
8772537c
AL
89 if (t) {
90 const char * const u = strchr(t, '\n');
91 if (u && (u[1] == 0)) {
92 const STRLEN len = u - s;
93 Move(s + 1, locs, len, char);
94 locs[len] = 0;
95 okay = TRUE;
98994639
HS
96 }
97 }
98 }
99
100 if (!okay)
101 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
102
103 return locs;
104}
105
8ef6e574
KW
106#endif
107
98994639
HS
108void
109Perl_set_numeric_radix(pTHX)
110{
111#ifdef USE_LOCALE_NUMERIC
112# ifdef HAS_LOCALECONV
7452cf6a 113 const struct lconv* const lc = localeconv();
98994639 114
98994639
HS
115 if (lc && lc->decimal_point) {
116 if (lc->decimal_point[0] == '.' && lc->decimal_point[1] == 0) {
117 SvREFCNT_dec(PL_numeric_radix_sv);
a0714e2c 118 PL_numeric_radix_sv = NULL;
98994639
HS
119 }
120 else {
121 if (PL_numeric_radix_sv)
122 sv_setpv(PL_numeric_radix_sv, lc->decimal_point);
123 else
124 PL_numeric_radix_sv = newSVpv(lc->decimal_point, 0);
c5f058df 125 if (! is_utf8_invariant_string((U8 *) lc->decimal_point, 0)
28acfe03 126 && is_utf8_string((U8 *) lc->decimal_point, 0)
c1284011 127 && _is_cur_LC_category_utf8(LC_NUMERIC))
28acfe03
KW
128 {
129 SvUTF8_on(PL_numeric_radix_sv);
130 }
98994639
HS
131 }
132 }
133 else
a0714e2c 134 PL_numeric_radix_sv = NULL;
69014004 135
2fcc0ca9
KW
136#ifdef DEBUGGING
137 if (DEBUG_L_TEST || debug_initialization) {
138 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
69014004 139 (PL_numeric_radix_sv)
37b7e435
KW
140 ? SvPVX(PL_numeric_radix_sv)
141 : "NULL",
142 (PL_numeric_radix_sv)
39eb7305 143 ? cBOOL(SvUTF8(PL_numeric_radix_sv))
2fcc0ca9
KW
144 : 0);
145 }
146#endif
69014004 147
98994639
HS
148# endif /* HAS_LOCALECONV */
149#endif /* USE_LOCALE_NUMERIC */
150}
151
a39edc4c
KW
152/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
153 * return of setlocale(), then this is extremely likely to be the C or POSIX
154 * locale. However, the output of setlocale() is documented to be opaque, but
155 * the odds are extremely small that it would return these two strings for some
156 * other locale. Note that VMS in these two locales includes many non-ASCII
157 * characters as controls and punctuation (below are hex bytes):
158 * cntrl: 00-1F 7F 84-97 9B-9F
159 * 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
160 * Oddly, none there are listed as alphas, though some represent alphabetics
161 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
98b630b3
KW
162#define isNAME_C_OR_POSIX(name) ((name) != NULL \
163 && ((*(name) == 'C' && (*(name + 1)) == '\0') \
a39edc4c
KW
164 || strEQ((name), "POSIX")))
165
98994639 166void
8772537c 167Perl_new_numeric(pTHX_ const char *newnum)
98994639
HS
168{
169#ifdef USE_LOCALE_NUMERIC
0d071d52
KW
170
171 /* Called after all libc setlocale() calls affecting LC_NUMERIC, to tell
172 * core Perl this and that 'newnum' is the name of the new locale.
173 * It installs this locale as the current underlying default.
174 *
175 * The default locale and the C locale can be toggled between by use of the
176 * set_numeric_local() and set_numeric_standard() functions, which should
177 * probably not be called directly, but only via macros like
178 * SET_NUMERIC_STANDARD() in perl.h.
179 *
180 * The toggling is necessary mainly so that a non-dot radix decimal point
181 * character can be output, while allowing internal calculations to use a
182 * dot.
183 *
184 * This sets several interpreter-level variables:
bb304765 185 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
0d071d52 186 * PL_numeric_local A boolean indicating if the toggled state is such
7738054c
KW
187 * that the current locale is the program's underlying
188 * locale
189 * PL_numeric_standard An int indicating if the toggled state is such
190 * that the current locale is the C locale. If non-zero,
191 * it is in C; if > 1, it means it may not be toggled away
192 * from C.
0d071d52
KW
193 * Note that both of the last two variables can be true at the same time,
194 * if the underlying locale is C. (Toggling is a no-op under these
195 * circumstances.)
196 *
197 * Any code changing the locale (outside this file) should use
198 * POSIX::setlocale, which calls this function. Therefore this function
199 * should be called directly only from this file and from
200 * POSIX::setlocale() */
201
b03f34cf 202 char *save_newnum;
98994639
HS
203
204 if (! newnum) {
43c5f42d
NC
205 Safefree(PL_numeric_name);
206 PL_numeric_name = NULL;
98994639
HS
207 PL_numeric_standard = TRUE;
208 PL_numeric_local = TRUE;
209 return;
210 }
211
b03f34cf 212 save_newnum = stdize_locale(savepv(newnum));
abe1abcf
KW
213
214 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
215 PL_numeric_local = TRUE;
216
b03f34cf 217 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 218 Safefree(PL_numeric_name);
b03f34cf 219 PL_numeric_name = save_newnum;
b03f34cf 220 }
abe1abcf
KW
221 else {
222 Safefree(save_newnum);
223 }
4c28b29c
KW
224
225 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
226 * have to worry about the radix being a non-dot. (Core operations that
227 * need the underlying locale change to it temporarily). */
228 set_numeric_standard();
229
e19f01cb 230 set_numeric_radix();
6959d69d 231
f2ce9e1c
JH
232#else
233 PERL_UNUSED_ARG(newnum);
98994639
HS
234#endif /* USE_LOCALE_NUMERIC */
235}
236
237void
238Perl_set_numeric_standard(pTHX)
239{
240#ifdef USE_LOCALE_NUMERIC
28c1bf33
KW
241 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
242 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
243 * macro avoids calling this routine if toggling isn't necessary according
244 * to our records (which could be wrong if some XS code has changed the
245 * locale behind our back) */
0d071d52 246
a9b8c0d8
KW
247 setlocale(LC_NUMERIC, "C");
248 PL_numeric_standard = TRUE;
249 PL_numeric_local = isNAME_C_OR_POSIX(PL_numeric_name);
250 set_numeric_radix();
2fcc0ca9
KW
251#ifdef DEBUGGING
252 if (DEBUG_L_TEST || debug_initialization) {
253 PerlIO_printf(Perl_debug_log,
254 "Underlying LC_NUMERIC locale now is C\n");
255 }
256#endif
98994639
HS
257
258#endif /* USE_LOCALE_NUMERIC */
259}
260
261void
262Perl_set_numeric_local(pTHX)
263{
264#ifdef USE_LOCALE_NUMERIC
28c1bf33
KW
265 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
266 * code should use the macros like SET_NUMERIC_LOCAL() in perl.h instead of
267 * calling this directly. The macro avoids calling this routine if
268 * toggling isn't necessary according to our records (which could be wrong
269 * if some XS code has changed the locale behind our back) */
a9b8c0d8
KW
270
271 setlocale(LC_NUMERIC, PL_numeric_name);
272 PL_numeric_standard = isNAME_C_OR_POSIX(PL_numeric_name);
273 PL_numeric_local = TRUE;
274 set_numeric_radix();
2fcc0ca9
KW
275#ifdef DEBUGGING
276 if (DEBUG_L_TEST || debug_initialization) {
277 PerlIO_printf(Perl_debug_log,
69014004 278 "Underlying LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
279 PL_numeric_name);
280 }
281#endif
98994639
HS
282
283#endif /* USE_LOCALE_NUMERIC */
284}
285
286/*
287 * Set up for a new ctype locale.
288 */
289void
8772537c 290Perl_new_ctype(pTHX_ const char *newctype)
98994639
HS
291{
292#ifdef USE_LOCALE_CTYPE
0d071d52
KW
293
294 /* Called after all libc setlocale() calls affecting LC_CTYPE, to tell
295 * core Perl this and that 'newctype' is the name of the new locale.
296 *
297 * This function sets up the folding arrays for all 256 bytes, assuming
298 * that tofold() is tolc() since fold case is not a concept in POSIX,
299 *
300 * Any code changing the locale (outside this file) should use
301 * POSIX::setlocale, which calls this function. Therefore this function
302 * should be called directly only from this file and from
303 * POSIX::setlocale() */
304
27da23d5 305 dVAR;
68067e4e 306 UV i;
98994639 307
7918f24d
NC
308 PERL_ARGS_ASSERT_NEW_CTYPE;
309
215c5139
KW
310 /* We will replace any bad locale warning with 1) nothing if the new one is
311 * ok; or 2) a new warning for the bad new locale */
312 if (PL_warn_locale) {
313 SvREFCNT_dec_NN(PL_warn_locale);
314 PL_warn_locale = NULL;
315 }
316
c1284011 317 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
318
319 /* A UTF-8 locale gets standard rules. But note that code still has to
320 * handle this specially because of the three problematic code points */
321 if (PL_in_utf8_CTYPE_locale) {
322 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
323 }
324 else {
8c6180a9
KW
325 /* Assume enough space for every character being bad. 4 spaces each
326 * for the 94 printable characters that are output like "'x' "; and 5
327 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
328 * NUL */
329 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
330
cc9eaeb0
KW
331 /* Don't check for problems if we are suppressing the warnings */
332 bool check_for_problems = ckWARN_d(WARN_LOCALE)
333 || UNLIKELY(DEBUG_L_TEST);
8c6180a9
KW
334 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
335 to start */
336 unsigned int bad_count = 0; /* Count of bad characters */
337
baa60164
KW
338 for (i = 0; i < 256; i++) {
339 if (isUPPER_LC((U8) i))
340 PL_fold_locale[i] = (U8) toLOWER_LC((U8) i);
341 else if (isLOWER_LC((U8) i))
342 PL_fold_locale[i] = (U8) toUPPER_LC((U8) i);
343 else
344 PL_fold_locale[i] = (U8) i;
8c6180a9
KW
345
346 /* If checking for locale problems, see if the native ASCII-range
347 * printables plus \n and \t are in their expected categories in
348 * the new locale. If not, this could mean big trouble, upending
349 * Perl's and most programs' assumptions, like having a
350 * metacharacter with special meaning become a \w. Fortunately,
351 * it's very rare to find locales that aren't supersets of ASCII
352 * nowadays. It isn't a problem for most controls to be changed
353 * into something else; we check only \n and \t, though perhaps \r
354 * could be an issue as well. */
355 if (check_for_problems
356 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
357 {
358 if ((isALPHANUMERIC_A(i) && ! isALPHANUMERIC_LC(i))
359 || (isPUNCT_A(i) && ! isPUNCT_LC(i))
360 || (isBLANK_A(i) && ! isBLANK_LC(i))
361 || (i == '\n' && ! isCNTRL_LC(i)))
362 {
363 if (bad_count) { /* Separate multiple entries with a
364 blank */
365 bad_chars_list[bad_count++] = ' ';
366 }
367 bad_chars_list[bad_count++] = '\'';
368 if (isPRINT_A(i)) {
369 bad_chars_list[bad_count++] = (char) i;
370 }
371 else {
372 bad_chars_list[bad_count++] = '\\';
373 if (i == '\n') {
374 bad_chars_list[bad_count++] = 'n';
375 }
376 else {
377 assert(i == '\t');
378 bad_chars_list[bad_count++] = 't';
379 }
380 }
381 bad_chars_list[bad_count++] = '\'';
382 bad_chars_list[bad_count] = '\0';
383 }
384 }
385 }
386
387#ifdef MB_CUR_MAX
388 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 389 * this locale requires more than one byte, there are going to be
8c6180a9 390 * problems. */
ba1a4362
KW
391 if (check_for_problems && MB_CUR_MAX > 1
392
393 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
394 * locale. Just assume that the implementation for them (plus
395 * for POSIX) is correct and the > 1 value is spurious. (Since
396 * these are specially handled to never be considered UTF-8
397 * locales, as long as this is the only problem, everything
398 * should work fine */
399 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
400 {
8c6180a9
KW
401 multi_byte_locale = TRUE;
402 }
403#endif
404
405 if (bad_count || multi_byte_locale) {
780fcc9f 406 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 407 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 408 newctype,
8c6180a9
KW
409 (multi_byte_locale)
410 ? " Some characters in it are not recognized by"
411 " Perl."
412 : "",
413 (bad_count)
414 ? "\nThe following characters (and maybe others)"
415 " may not have the same meaning as the Perl"
416 " program expects:\n"
417 : "",
418 (bad_count)
419 ? bad_chars_list
420 : ""
421 );
cc9eaeb0
KW
422 /* If we are actually in the scope of the locale or are debugging,
423 * output the message now. Otherwise we save it to be output at
424 * the first operation using this locale, if that actually happens.
425 * Most programs don't use locales, so they are immune to bad ones.
426 * */
427 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f
KW
428
429 /* We have to save 'newctype' because the setlocale() just
430 * below may destroy it. The next setlocale() further down
431 * should restore it properly so that the intermediate change
432 * here is transparent to this function's caller */
433 const char * const badlocale = savepv(newctype);
434
435 setlocale(LC_CTYPE, "C");
436
437 /* The '0' below suppresses a bogus gcc compiler warning */
438 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
439 setlocale(LC_CTYPE, badlocale);
c0f3a893 440 Safefree(badlocale);
780fcc9f
KW
441 SvREFCNT_dec_NN(PL_warn_locale);
442 PL_warn_locale = NULL;
443 }
baa60164 444 }
31f05a37 445 }
98994639
HS
446
447#endif /* USE_LOCALE_CTYPE */
7918f24d 448 PERL_ARGS_ASSERT_NEW_CTYPE;
8772537c 449 PERL_UNUSED_ARG(newctype);
96a5add6 450 PERL_UNUSED_CONTEXT;
98994639
HS
451}
452
98994639 453void
2726666d
KW
454Perl__warn_problematic_locale()
455{
2726666d
KW
456
457#ifdef USE_LOCALE_CTYPE
458
5f04a188
KW
459 dTHX;
460
461 /* Internal-to-core function that outputs the message in PL_warn_locale,
462 * and then NULLS it. Should be called only through the macro
463 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
464
2726666d
KW
465 if (PL_warn_locale) {
466 /*GCC_DIAG_IGNORE(-Wformat-security); Didn't work */
467 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
468 SvPVX(PL_warn_locale),
469 0 /* dummy to avoid compiler warning */ );
470 /* GCC_DIAG_RESTORE; */
471 SvREFCNT_dec_NN(PL_warn_locale);
472 PL_warn_locale = NULL;
473 }
474
475#endif
476
477}
478
479void
8772537c 480Perl_new_collate(pTHX_ const char *newcoll)
98994639
HS
481{
482#ifdef USE_LOCALE_COLLATE
0d071d52
KW
483
484 /* Called after all libc setlocale() calls affecting LC_COLLATE, to tell
485 * core Perl this and that 'newcoll' is the name of the new locale.
486 *
487 * Any code changing the locale (outside this file) should use
488 * POSIX::setlocale, which calls this function. Therefore this function
489 * should be called directly only from this file and from
d35fca5f
KW
490 * POSIX::setlocale().
491 *
492 * The design of locale collation is that every locale change is given an
493 * index 'PL_collation_ix'. The first time a string particpates in an
494 * operation that requires collation while locale collation is active, it
495 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
496 * magic includes the collation index, and the transformation of the string
497 * by strxfrm(), q.v. That transformation is used when doing comparisons,
498 * instead of the string itself. If a string changes, the magic is
499 * cleared. The next time the locale changes, the index is incremented,
500 * and so we know during a comparison that the transformation is not
501 * necessarily still valid, and so is recomputed. Note that if the locale
502 * changes enough times, the index could wrap (a U32), and it is possible
503 * that a transformation would improperly be considered valid, leading to
504 * an unlikely bug */
0d071d52 505
98994639
HS
506 if (! newcoll) {
507 if (PL_collation_name) {
508 ++PL_collation_ix;
509 Safefree(PL_collation_name);
510 PL_collation_name = NULL;
511 }
512 PL_collation_standard = TRUE;
00bf60ca 513 is_standard_collation:
98994639
HS
514 PL_collxfrm_base = 0;
515 PL_collxfrm_mult = 2;
165a1c52 516 PL_in_utf8_COLLATE_locale = FALSE;
6696cfa7 517 *PL_strxfrm_min_char = '\0';
a4a439fb 518 PL_strxfrm_max_cp = 0;
98994639
HS
519 return;
520 }
521
d35fca5f 522 /* If this is not the same locale as currently, set the new one up */
98994639
HS
523 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
524 ++PL_collation_ix;
525 Safefree(PL_collation_name);
526 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 527 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
528 if (PL_collation_standard) {
529 goto is_standard_collation;
530 }
98994639 531
165a1c52 532 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
6696cfa7 533 *PL_strxfrm_min_char = '\0';
a4a439fb 534 PL_strxfrm_max_cp = 0;
165a1c52 535
59c018b9
KW
536 /* A locale collation definition includes primary, secondary, tertiary,
537 * etc. weights for each character. To sort, the primary weights are
538 * used, and only if they compare equal, then the secondary weights are
539 * used, and only if they compare equal, then the tertiary, etc.
540 *
541 * strxfrm() works by taking the input string, say ABC, and creating an
542 * output transformed string consisting of first the primary weights,
543 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
544 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
545 * may not have weights at every level. In our example, let's say B
546 * doesn't have a tertiary weight, and A doesn't have a secondary
547 * weight. The constructed string is then going to be
548 * A¹B¹C¹ B²C² A³C³ ....
549 * This has the desired effect that strcmp() will look at the secondary
550 * or tertiary weights only if the strings compare equal at all higher
551 * priority weights. The spaces shown here, like in
552 * "A¹B¹C¹ * A²B²C² "
553 * are not just for readability. In the general case, these must
554 * actually be bytes, which we will call here 'separator weights'; and
555 * they must be smaller than any other weight value, but since these
556 * are C strings, only the terminating one can be a NUL (some
557 * implementations may include a non-NUL separator weight just before
558 * the NUL). Implementations tend to reserve 01 for the separator
559 * weights. They are needed so that a shorter string's secondary
560 * weights won't be misconstrued as primary weights of a longer string,
561 * etc. By making them smaller than any other weight, the shorter
562 * string will sort first. (Actually, if all secondary weights are
563 * smaller than all primary ones, there is no need for a separator
564 * weight between those two levels, etc.)
565 *
566 * The length of the transformed string is roughly a linear function of
567 * the input string. It's not exactly linear because some characters
568 * don't have weights at all levels. When we call strxfrm() we have to
569 * allocate some memory to hold the transformed string. The
570 * calculations below try to find coefficients 'm' and 'b' for this
571 * locale so that m*x + b equals how much space we need, given the size
572 * of the input string in 'x'. If we calculate too small, we increase
573 * the size as needed, and call strxfrm() again, but it is better to
574 * get it right the first time to avoid wasted expensive string
575 * transformations. */
576
98994639 577 {
79f120c8
KW
578 /* We use the string below to find how long the tranformation of it
579 * is. Almost all locales are supersets of ASCII, or at least the
580 * ASCII letters. We use all of them, half upper half lower,
581 * because if we used fewer, we might hit just the ones that are
582 * outliers in a particular locale. Most of the strings being
583 * collated will contain a preponderance of letters, and even if
584 * they are above-ASCII, they are likely to have the same number of
585 * weight levels as the ASCII ones. It turns out that digits tend
586 * to have fewer levels, and some punctuation has more, but those
587 * are relatively sparse in text, and khw believes this gives a
588 * reasonable result, but it could be changed if experience so
589 * dictates. */
590 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
591 char * x_longer; /* Transformed 'longer' */
592 Size_t x_len_longer; /* Length of 'x_longer' */
593
594 char * x_shorter; /* We also transform a substring of 'longer' */
595 Size_t x_len_shorter;
596
a4a439fb 597 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
598 * are interested only in its length). It is used because it has
599 * the intelligence to handle all cases, but to work, it needs some
600 * values of 'm' and 'b' to get it started. For the purposes of
601 * this calculation we use a very conservative estimate of 'm' and
602 * 'b'. This assumes a weight can be multiple bytes, enough to
603 * hold any UV on the platform, and there are 5 levels, 4 weight
604 * bytes, and a trailing NUL. */
605 PL_collxfrm_base = 5;
606 PL_collxfrm_mult = 5 * sizeof(UV);
607
608 /* Find out how long the transformation really is */
a4a439fb
KW
609 x_longer = _mem_collxfrm(longer,
610 sizeof(longer) - 1,
611 &x_len_longer,
612
613 /* We avoid converting to UTF-8 in the
614 * called function by telling it the
615 * string is in UTF-8 if the locale is a
616 * UTF-8 one. Since the string passed
617 * here is invariant under UTF-8, we can
618 * claim it's UTF-8 even though it isn't.
619 * */
620 PL_in_utf8_COLLATE_locale);
79f120c8
KW
621 Safefree(x_longer);
622
623 /* Find out how long the transformation of a substring of 'longer'
624 * is. Together the lengths of these transformations are
625 * sufficient to calculate 'm' and 'b'. The substring is all of
626 * 'longer' except the first character. This minimizes the chances
627 * of being swayed by outliers */
a4a439fb 628 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 629 sizeof(longer) - 2,
a4a439fb
KW
630 &x_len_shorter,
631 PL_in_utf8_COLLATE_locale);
79f120c8
KW
632 Safefree(x_shorter);
633
634 /* If the results are nonsensical for this simple test, the whole
635 * locale definition is suspect. Mark it so that locale collation
636 * is not active at all for it. XXX Should we warn? */
637 if ( x_len_shorter == 0
638 || x_len_longer == 0
639 || x_len_shorter >= x_len_longer)
640 {
641 PL_collxfrm_mult = 0;
642 PL_collxfrm_base = 0;
643 }
644 else {
645 SSize_t base; /* Temporary */
646
647 /* We have both: m * strlen(longer) + b = x_len_longer
648 * m * strlen(shorter) + b = x_len_shorter;
649 * subtracting yields:
650 * m * (strlen(longer) - strlen(shorter))
651 * = x_len_longer - x_len_shorter
652 * But we have set things up so that 'shorter' is 1 byte smaller
653 * than 'longer'. Hence:
654 * m = x_len_longer - x_len_shorter
655 *
656 * But if something went wrong, make sure the multiplier is at
657 * least 1.
658 */
659 if (x_len_longer > x_len_shorter) {
660 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
661 }
662 else {
663 PL_collxfrm_mult = 1;
664 }
665
666 /* mx + b = len
667 * so: b = len - mx
668 * but in case something has gone wrong, make sure it is
669 * non-negative */
670 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
671 if (base < 0) {
672 base = 0;
673 }
674
675 /* Add 1 for the trailing NUL */
676 PL_collxfrm_base = base + 1;
677 }
58eebef2
KW
678
679#ifdef DEBUGGING
680 if (DEBUG_L_TEST || debug_initialization) {
681 PerlIO_printf(Perl_debug_log,
b07929e4
KW
682 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
683 "x_len_longer=%zu,"
684 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
685 __FILE__, __LINE__,
686 PL_in_utf8_COLLATE_locale,
687 x_len_shorter, x_len_longer,
688 PL_collxfrm_mult, PL_collxfrm_base);
689 }
690#endif
98994639
HS
691 }
692 }
693
f2ce9e1c
JH
694#else
695 PERL_UNUSED_ARG(newcoll);
98994639
HS
696#endif /* USE_LOCALE_COLLATE */
697}
698
b385bb4d
KW
699#ifdef WIN32
700
701char *
702Perl_my_setlocale(pTHX_ int category, const char* locale)
703{
704 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
705 * difference unless the input locale is "", which means on Windows to get
706 * the machine default, which is set via the computer's "Regional and
707 * Language Options" (or its current equivalent). In POSIX, it instead
708 * means to find the locale from the user's environment. This routine
709 * looks in the environment, and, if anything is found, uses that instead
710 * of going to the machine default. If there is no environment override,
711 * the machine default is used, as normal, by calling the real setlocale()
712 * with "". The POSIX behavior is to use the LC_ALL variable if set;
713 * otherwise to use the particular category's variable if set; otherwise to
714 * use the LANG variable. */
715
175c4cf9 716 bool override_LC_ALL = FALSE;
89f7b9aa
KW
717 char * result;
718
b385bb4d
KW
719 if (locale && strEQ(locale, "")) {
720# ifdef LC_ALL
721 locale = PerlEnv_getenv("LC_ALL");
722 if (! locale) {
723#endif
724 switch (category) {
725# ifdef LC_ALL
726 case LC_ALL:
481465ea 727 override_LC_ALL = TRUE;
b385bb4d
KW
728 break; /* We already know its variable isn't set */
729# endif
730# ifdef USE_LOCALE_TIME
731 case LC_TIME:
732 locale = PerlEnv_getenv("LC_TIME");
733 break;
734# endif
735# ifdef USE_LOCALE_CTYPE
736 case LC_CTYPE:
737 locale = PerlEnv_getenv("LC_CTYPE");
738 break;
739# endif
740# ifdef USE_LOCALE_COLLATE
741 case LC_COLLATE:
742 locale = PerlEnv_getenv("LC_COLLATE");
743 break;
744# endif
745# ifdef USE_LOCALE_MONETARY
746 case LC_MONETARY:
747 locale = PerlEnv_getenv("LC_MONETARY");
748 break;
749# endif
750# ifdef USE_LOCALE_NUMERIC
751 case LC_NUMERIC:
752 locale = PerlEnv_getenv("LC_NUMERIC");
753 break;
754# endif
755# ifdef USE_LOCALE_MESSAGES
756 case LC_MESSAGES:
757 locale = PerlEnv_getenv("LC_MESSAGES");
758 break;
759# endif
760 default:
761 /* This is a category, like PAPER_SIZE that we don't
762 * know about; and so can't provide a wrapper. */
763 break;
764 }
765 if (! locale) {
766 locale = PerlEnv_getenv("LANG");
481465ea 767 if (! locale) {
b385bb4d
KW
768 locale = "";
769 }
770 }
771# ifdef LC_ALL
772 }
773# endif
774 }
775
89f7b9aa 776 result = setlocale(category, locale);
bbc98134
KW
777 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
778 _setlocale_debug_string(category, locale, result)));
89f7b9aa 779
481465ea 780 if (! override_LC_ALL) {
89f7b9aa
KW
781 return result;
782 }
783
dfd77d7a 784 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
785 * LANG variable or the system default if there is no LANG. But these have
786 * lower priority than the other LC_foo variables, so override it for each
787 * one that is set. (If they are set to "", it means to use the same thing
788 * we just set LC_ALL to, so can skip) */
89f7b9aa
KW
789# ifdef USE_LOCALE_TIME
790 result = PerlEnv_getenv("LC_TIME");
730252b2 791 if (result && strNE(result, "")) {
89f7b9aa 792 setlocale(LC_TIME, result);
bbc98134
KW
793 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
794 __FILE__, __LINE__,
795 _setlocale_debug_string(LC_TIME, result, "not captured")));
89f7b9aa
KW
796 }
797# endif
798# ifdef USE_LOCALE_CTYPE
799 result = PerlEnv_getenv("LC_CTYPE");
730252b2 800 if (result && strNE(result, "")) {
89f7b9aa 801 setlocale(LC_CTYPE, result);
bbc98134
KW
802 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
803 __FILE__, __LINE__,
804 _setlocale_debug_string(LC_CTYPE, result, "not captured")));
89f7b9aa
KW
805 }
806# endif
807# ifdef USE_LOCALE_COLLATE
808 result = PerlEnv_getenv("LC_COLLATE");
730252b2 809 if (result && strNE(result, "")) {
89f7b9aa 810 setlocale(LC_COLLATE, result);
bbc98134
KW
811 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
812 __FILE__, __LINE__,
813 _setlocale_debug_string(LC_COLLATE, result, "not captured")));
89f7b9aa
KW
814 }
815# endif
816# ifdef USE_LOCALE_MONETARY
817 result = PerlEnv_getenv("LC_MONETARY");
730252b2 818 if (result && strNE(result, "")) {
89f7b9aa 819 setlocale(LC_MONETARY, result);
bbc98134
KW
820 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
821 __FILE__, __LINE__,
822 _setlocale_debug_string(LC_MONETARY, result, "not captured")));
89f7b9aa
KW
823 }
824# endif
825# ifdef USE_LOCALE_NUMERIC
826 result = PerlEnv_getenv("LC_NUMERIC");
730252b2 827 if (result && strNE(result, "")) {
89f7b9aa 828 setlocale(LC_NUMERIC, result);
bbc98134
KW
829 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
830 __FILE__, __LINE__,
831 _setlocale_debug_string(LC_NUMERIC, result, "not captured")));
89f7b9aa
KW
832 }
833# endif
834# ifdef USE_LOCALE_MESSAGES
835 result = PerlEnv_getenv("LC_MESSAGES");
730252b2 836 if (result && strNE(result, "")) {
89f7b9aa 837 setlocale(LC_MESSAGES, result);
bbc98134
KW
838 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
839 __FILE__, __LINE__,
840 _setlocale_debug_string(LC_MESSAGES, result, "not captured")));
89f7b9aa
KW
841 }
842# endif
843
bbc98134
KW
844 result = setlocale(LC_ALL, NULL);
845 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
846 __FILE__, __LINE__,
847 _setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 848
bbc98134 849 return result;
b385bb4d
KW
850}
851
852#endif
853
854
98994639
HS
855/*
856 * Initialize locale awareness.
857 */
858int
859Perl_init_i18nl10n(pTHX_ int printwarn)
860{
0e92a118
KW
861 /* printwarn is
862 *
863 * 0 if not to output warning when setup locale is bad
864 * 1 if to output warning based on value of PERL_BADLANG
865 * >1 if to output regardless of PERL_BADLANG
866 *
867 * returns
98994639 868 * 1 = set ok or not applicable,
0e92a118
KW
869 * 0 = fallback to a locale of lower priority
870 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
871 *
872 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
873 * set, debugging information is output.
874 *
875 * This looks more complicated than it is, mainly due to the #ifdefs.
876 *
877 * We try to set LC_ALL to the value determined by the environment. If
878 * there is no LC_ALL on this platform, we try the individual categories we
879 * know about. If this works, we are done.
880 *
881 * But if it doesn't work, we have to do something else. We search the
882 * environment variables ourselves instead of relying on the system to do
883 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
884 * think there is one), and the ultimate fallback "C". This is all done in
885 * the same loop as above to avoid duplicating code, but it makes things
886 * more complex. After the original failure, we add the fallback
887 * possibilities to the list of locales to try, and iterate the loop
888 * through them all until one succeeds.
889 *
890 * On Ultrix, the locale MUST come from the environment, so there is
891 * preliminary code to set it. I (khw) am not sure that it is necessary,
892 * and that this couldn't be folded into the loop, but barring any real
893 * platforms to test on, it's staying as-is
894 *
895 * A slight complication is that in embedded Perls, the locale may already
896 * be set-up, and we don't want to get it from the normal environment
897 * variables. This is handled by having a special environment variable
898 * indicate we're in this situation. We simply set setlocale's 2nd
899 * parameter to be a NULL instead of "". That indicates to setlocale that
900 * it is not to change anything, but to return the current value,
901 * effectively initializing perl's db to what the locale already is.
902 *
903 * We play the same trick with NULL if a LC_ALL succeeds. We call
904 * setlocale() on the individual categores with NULL to get their existing
905 * values for our db, instead of trying to change them.
906 * */
98994639 907
0e92a118
KW
908 int ok = 1;
909
98994639 910#if defined(USE_LOCALE)
98994639
HS
911#ifdef USE_LOCALE_CTYPE
912 char *curctype = NULL;
913#endif /* USE_LOCALE_CTYPE */
914#ifdef USE_LOCALE_COLLATE
915 char *curcoll = NULL;
916#endif /* USE_LOCALE_COLLATE */
917#ifdef USE_LOCALE_NUMERIC
918 char *curnum = NULL;
919#endif /* USE_LOCALE_NUMERIC */
920#ifdef __GLIBC__
175c4cf9 921 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
98994639 922#endif
65ebb059 923
ccd65d51
KW
924 /* NULL uses the existing already set up locale */
925 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
926 ? NULL
927 : "";
c3fcd832
KW
928 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
929 unsigned int trial_locales_count;
175c4cf9
KW
930 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
931 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 932 bool setlocale_failure = FALSE;
65ebb059
KW
933 unsigned int i;
934 char *p;
175c4cf9
KW
935
936 /* A later getenv() could zap this, so only use here */
937 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
938
939 const bool locwarn = (printwarn > 1
940 || (printwarn
941 && (! bad_lang_use_once
22ff3130
HS
942 || (
943 /* disallow with "" or "0" */
944 *bad_lang_use_once
945 && strNE("0", bad_lang_use_once)))));
0e92a118 946 bool done = FALSE;
5d1187d1
KW
947 char * sl_result; /* return from setlocale() */
948 char * locale_param;
6bce99ee
JH
949#ifdef WIN32
950 /* In some systems you can find out the system default locale
951 * and use that as the fallback locale. */
952# define SYSTEM_DEFAULT_LOCALE
953#endif
954#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059 955 const char *system_default_locale = NULL;
6bce99ee 956#endif
98994639 957
2fcc0ca9 958#ifdef DEBUGGING
27cdc72e 959 DEBUG_INITIALIZATION_set((PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT"))
2fcc0ca9 960 ? TRUE
27cdc72e 961 : FALSE);
2fcc0ca9
KW
962# define DEBUG_LOCALE_INIT(category, locale, result) \
963 STMT_START { \
964 if (debug_initialization) { \
965 PerlIO_printf(Perl_debug_log, \
966 "%s:%d: %s\n", \
967 __FILE__, __LINE__, \
968 _setlocale_debug_string(category, \
969 locale, \
970 result)); \
971 } \
972 } STMT_END
973#else
974# define DEBUG_LOCALE_INIT(a,b,c)
975#endif
976
0e92a118
KW
977#ifndef LOCALE_ENVIRON_REQUIRED
978 PERL_UNUSED_VAR(done);
5d1187d1 979 PERL_UNUSED_VAR(locale_param);
0e92a118 980#else
98994639
HS
981
982 /*
983 * Ultrix setlocale(..., "") fails if there are no environment
984 * variables from which to get a locale name.
985 */
986
b3e384bf 987# ifdef LC_ALL
98994639 988 if (lang) {
5d1187d1
KW
989 sl_result = my_setlocale(LC_ALL, setlocale_init);
990 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result);
991 if (sl_result)
98994639
HS
992 done = TRUE;
993 else
994 setlocale_failure = TRUE;
995 }
5d1187d1 996 if (! setlocale_failure) {
b3e384bf 997# ifdef USE_LOCALE_CTYPE
5d1187d1
KW
998 locale_param = (! done && (lang || PerlEnv_getenv("LC_CTYPE")))
999 ? setlocale_init
1000 : NULL;
1001 curctype = my_setlocale(LC_CTYPE, locale_param);
1002 DEBUG_LOCALE_INIT(LC_CTYPE, locale_param, sl_result);
1003 if (! curctype)
98994639
HS
1004 setlocale_failure = TRUE;
1005 else
1006 curctype = savepv(curctype);
b3e384bf
KW
1007# endif /* USE_LOCALE_CTYPE */
1008# ifdef USE_LOCALE_COLLATE
5d1187d1
KW
1009 locale_param = (! done && (lang || PerlEnv_getenv("LC_COLLATE")))
1010 ? setlocale_init
1011 : NULL;
1012 curcoll = my_setlocale(LC_COLLATE, locale_param);
1013 DEBUG_LOCALE_INIT(LC_COLLATE, locale_param, sl_result);
1014 if (! curcoll)
98994639
HS
1015 setlocale_failure = TRUE;
1016 else
1017 curcoll = savepv(curcoll);
b3e384bf
KW
1018# endif /* USE_LOCALE_COLLATE */
1019# ifdef USE_LOCALE_NUMERIC
5d1187d1
KW
1020 locale_param = (! done && (lang || PerlEnv_getenv("LC_NUMERIC")))
1021 ? setlocale_init
1022 : NULL;
1023 curnum = my_setlocale(LC_NUMERIC, locale_param);
1024 DEBUG_LOCALE_INIT(LC_NUMERIC, locale_param, sl_result);
1025 if (! curnum)
98994639
HS
1026 setlocale_failure = TRUE;
1027 else
1028 curnum = savepv(curnum);
b3e384bf 1029# endif /* USE_LOCALE_NUMERIC */
a782673d 1030# ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
1031 locale_param = (! done && (lang || PerlEnv_getenv("LC_MESSAGES")))
1032 ? setlocale_init
1033 : NULL;
1034 sl_result = my_setlocale(LC_MESSAGES, locale_param);
1035 DEBUG_LOCALE_INIT(LC_MESSAGES, locale_param, sl_result);
9f42613c 1036 if (! sl_result) {
a782673d
KW
1037 setlocale_failure = TRUE;
1038 }
1039# endif /* USE_LOCALE_MESSAGES */
c835d6be 1040# ifdef USE_LOCALE_MONETARY
5d1187d1
KW
1041 locale_param = (! done && (lang || PerlEnv_getenv("LC_MONETARY")))
1042 ? setlocale_init
1043 : NULL;
1044 sl_result = my_setlocale(LC_MONETARY, locale_param);
1045 DEBUG_LOCALE_INIT(LC_MONETARY, locale_param, sl_result);
1046 if (! sl_result) {
c835d6be
KW
1047 setlocale_failure = TRUE;
1048 }
1049# endif /* USE_LOCALE_MONETARY */
98994639
HS
1050 }
1051
b3e384bf 1052# endif /* LC_ALL */
98994639
HS
1053
1054#endif /* !LOCALE_ENVIRON_REQUIRED */
1055
65ebb059 1056 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
1057 * the list. Normally the loop is executed just once. But if setting the
1058 * locale fails, inside the loop we add fallback trials to the array and so
1059 * will execute the loop multiple times */
c3fcd832
KW
1060 trial_locales[0] = setlocale_init;
1061 trial_locales_count = 1;
65ebb059
KW
1062 for (i= 0; i < trial_locales_count; i++) {
1063 const char * trial_locale = trial_locales[i];
1064
1065 if (i > 0) {
1066
1067 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
1068 * when i==0, but I (khw) don't think that behavior makes much
1069 * sense */
1070 setlocale_failure = FALSE;
1071
6bce99ee
JH
1072#ifdef SYSTEM_DEFAULT_LOCALE
1073# ifdef WIN32
65ebb059
KW
1074 /* On Windows machines, an entry of "" after the 0th means to use
1075 * the system default locale, which we now proceed to get. */
1076 if (strEQ(trial_locale, "")) {
1077 unsigned int j;
1078
1079 /* Note that this may change the locale, but we are going to do
1080 * that anyway just below */
1081 system_default_locale = setlocale(LC_ALL, "");
5d1187d1 1082 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059
KW
1083
1084 /* Skip if invalid or it's already on the list of locales to
1085 * try */
1086 if (! system_default_locale) {
1087 goto next_iteration;
1088 }
1089 for (j = 0; j < trial_locales_count; j++) {
1090 if (strEQ(system_default_locale, trial_locales[j])) {
1091 goto next_iteration;
1092 }
1093 }
1094
1095 trial_locale = system_default_locale;
1096 }
6bce99ee
JH
1097# endif /* WIN32 */
1098#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
1099 }
1100
98994639 1101#ifdef LC_ALL
5d1187d1
KW
1102 sl_result = my_setlocale(LC_ALL, trial_locale);
1103 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result);
1104 if (! sl_result) {
49c85077 1105 setlocale_failure = TRUE;
7cd8b568
KW
1106 }
1107 else {
1108 /* Since LC_ALL succeeded, it should have changed all the other
1109 * categories it can to its value; so we massage things so that the
1110 * setlocales below just return their category's current values.
1111 * This adequately handles the case in NetBSD where LC_COLLATE may
1112 * not be defined for a locale, and setting it individually will
1113 * fail, whereas setting LC_ALL suceeds, leaving LC_COLLATE set to
1114 * the POSIX locale. */
1115 trial_locale = NULL;
1116 }
98994639
HS
1117#endif /* LC_ALL */
1118
49c85077 1119 if (!setlocale_failure) {
98994639 1120#ifdef USE_LOCALE_CTYPE
49c85077 1121 Safefree(curctype);
5d1187d1
KW
1122 curctype = my_setlocale(LC_CTYPE, trial_locale);
1123 DEBUG_LOCALE_INIT(LC_CTYPE, trial_locale, curctype);
1124 if (! curctype)
49c85077
KW
1125 setlocale_failure = TRUE;
1126 else
1127 curctype = savepv(curctype);
98994639
HS
1128#endif /* USE_LOCALE_CTYPE */
1129#ifdef USE_LOCALE_COLLATE
49c85077 1130 Safefree(curcoll);
5d1187d1
KW
1131 curcoll = my_setlocale(LC_COLLATE, trial_locale);
1132 DEBUG_LOCALE_INIT(LC_COLLATE, trial_locale, curcoll);
1133 if (! curcoll)
49c85077
KW
1134 setlocale_failure = TRUE;
1135 else
1136 curcoll = savepv(curcoll);
98994639
HS
1137#endif /* USE_LOCALE_COLLATE */
1138#ifdef USE_LOCALE_NUMERIC
49c85077 1139 Safefree(curnum);
5d1187d1
KW
1140 curnum = my_setlocale(LC_NUMERIC, trial_locale);
1141 DEBUG_LOCALE_INIT(LC_NUMERIC, trial_locale, curnum);
1142 if (! curnum)
49c85077
KW
1143 setlocale_failure = TRUE;
1144 else
1145 curnum = savepv(curnum);
98994639 1146#endif /* USE_LOCALE_NUMERIC */
a782673d 1147#ifdef USE_LOCALE_MESSAGES
5d1187d1
KW
1148 sl_result = my_setlocale(LC_MESSAGES, trial_locale);
1149 DEBUG_LOCALE_INIT(LC_MESSAGES, trial_locale, sl_result);
1150 if (! (sl_result))
a782673d
KW
1151 setlocale_failure = TRUE;
1152#endif /* USE_LOCALE_MESSAGES */
c835d6be 1153#ifdef USE_LOCALE_MONETARY
5d1187d1
KW
1154 sl_result = my_setlocale(LC_MONETARY, trial_locale);
1155 DEBUG_LOCALE_INIT(LC_MONETARY, trial_locale, sl_result);
1156 if (! (sl_result))
c835d6be
KW
1157 setlocale_failure = TRUE;
1158#endif /* USE_LOCALE_MONETARY */
1159
49c85077
KW
1160 if (! setlocale_failure) { /* Success */
1161 break;
1162 }
65ebb059 1163 }
98994639 1164
49c85077
KW
1165 /* Here, something failed; will need to try a fallback. */
1166 ok = 0;
65ebb059 1167
49c85077
KW
1168 if (i == 0) {
1169 unsigned int j;
98994639 1170
65ebb059 1171 if (locwarn) { /* Output failure info only on the first one */
98994639
HS
1172#ifdef LC_ALL
1173
49c85077
KW
1174 PerlIO_printf(Perl_error_log,
1175 "perl: warning: Setting locale failed.\n");
98994639
HS
1176
1177#else /* !LC_ALL */
1178
49c85077
KW
1179 PerlIO_printf(Perl_error_log,
1180 "perl: warning: Setting locale failed for the categories:\n\t");
20a240df 1181# ifdef USE_LOCALE_CTYPE
49c85077
KW
1182 if (! curctype)
1183 PerlIO_printf(Perl_error_log, "LC_CTYPE ");
20a240df
KW
1184# endif /* USE_LOCALE_CTYPE */
1185# ifdef USE_LOCALE_COLLATE
49c85077
KW
1186 if (! curcoll)
1187 PerlIO_printf(Perl_error_log, "LC_COLLATE ");
20a240df
KW
1188# endif /* USE_LOCALE_COLLATE */
1189# ifdef USE_LOCALE_NUMERIC
49c85077
KW
1190 if (! curnum)
1191 PerlIO_printf(Perl_error_log, "LC_NUMERIC ");
20a240df 1192# endif /* USE_LOCALE_NUMERIC */
a782673d 1193 PerlIO_printf(Perl_error_log, "and possibly others\n");
98994639
HS
1194
1195#endif /* LC_ALL */
1196
49c85077
KW
1197 PerlIO_printf(Perl_error_log,
1198 "perl: warning: Please check that your locale settings:\n");
98994639
HS
1199
1200#ifdef __GLIBC__
49c85077
KW
1201 PerlIO_printf(Perl_error_log,
1202 "\tLANGUAGE = %c%s%c,\n",
1203 language ? '"' : '(',
1204 language ? language : "unset",
1205 language ? '"' : ')');
98994639
HS
1206#endif
1207
49c85077
KW
1208 PerlIO_printf(Perl_error_log,
1209 "\tLC_ALL = %c%s%c,\n",
1210 lc_all ? '"' : '(',
1211 lc_all ? lc_all : "unset",
1212 lc_all ? '"' : ')');
98994639
HS
1213
1214#if defined(USE_ENVIRON_ARRAY)
49c85077
KW
1215 {
1216 char **e;
1217 for (e = environ; *e; e++) {
0eb7f56a
YO
1218 if (strEQs(*e, "LC_")
1219 && strNEs(*e, "LC_ALL=")
49c85077
KW
1220 && (p = strchr(*e, '=')))
1221 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
1222 (int)(p - *e), *e, p + 1);
1223 }
1224 }
98994639 1225#else
49c85077
KW
1226 PerlIO_printf(Perl_error_log,
1227 "\t(possibly more locale environment variables)\n");
98994639
HS
1228#endif
1229
49c85077
KW
1230 PerlIO_printf(Perl_error_log,
1231 "\tLANG = %c%s%c\n",
1232 lang ? '"' : '(',
1233 lang ? lang : "unset",
1234 lang ? '"' : ')');
98994639 1235
49c85077
KW
1236 PerlIO_printf(Perl_error_log,
1237 " are supported and installed on your system.\n");
1238 }
98994639 1239
65ebb059 1240 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 1241 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
1242 * usually change the upper bound of the loop we are in.
1243 *
1244 * Since the system's default way of setting the locale has not
1245 * found one that works, We use Perl's defined ordering: LC_ALL,
1246 * LANG, and the C locale. We don't try the same locale twice, so
1247 * don't add to the list if already there. (On POSIX systems, the
1248 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
1249 * but there's no harm done by doing it explicitly.
1250 *
1251 * Note that this tries the LC_ALL environment variable even on
1252 * systems which have no LC_ALL locale setting. This may or may
1253 * not have been originally intentional, but there's no real need
1254 * to change the behavior. */
65ebb059
KW
1255 if (lc_all) {
1256 for (j = 0; j < trial_locales_count; j++) {
1257 if (strEQ(lc_all, trial_locales[j])) {
1258 goto done_lc_all;
1259 }
1260 }
1261 trial_locales[trial_locales_count++] = lc_all;
1262 }
1263 done_lc_all:
98994639 1264
65ebb059
KW
1265 if (lang) {
1266 for (j = 0; j < trial_locales_count; j++) {
1267 if (strEQ(lang, trial_locales[j])) {
1268 goto done_lang;
1269 }
1270 }
1271 trial_locales[trial_locales_count++] = lang;
1272 }
1273 done_lang:
1274
1275#if defined(WIN32) && defined(LC_ALL)
1276 /* For Windows, we also try the system default locale before "C".
1277 * (If there exists a Windows without LC_ALL we skip this because
1278 * it gets too complicated. For those, the "C" is the next
1279 * fallback possibility). The "" is the same as the 0th element of
1280 * the array, but the code at the loop above knows to treat it
1281 * differently when not the 0th */
1282 trial_locales[trial_locales_count++] = "";
1283#endif
1284
1285 for (j = 0; j < trial_locales_count; j++) {
1286 if (strEQ("C", trial_locales[j])) {
1287 goto done_C;
1288 }
1289 }
1290 trial_locales[trial_locales_count++] = "C";
98994639 1291
65ebb059
KW
1292 done_C: ;
1293 } /* end of first time through the loop */
98994639 1294
65ebb059
KW
1295#ifdef WIN32
1296 next_iteration: ;
1297#endif
1298
1299 } /* end of looping through the trial locales */
1300
1301 if (ok < 1) { /* If we tried to fallback */
1302 const char* msg;
1303 if (! setlocale_failure) { /* fallback succeeded */
1304 msg = "Falling back to";
1305 }
1306 else { /* fallback failed */
98994639 1307
65ebb059
KW
1308 /* We dropped off the end of the loop, so have to decrement i to
1309 * get back to the value the last time through */
1310 i--;
98994639 1311
65ebb059
KW
1312 ok = -1;
1313 msg = "Failed to fall back to";
1314
1315 /* To continue, we should use whatever values we've got */
98994639 1316#ifdef USE_LOCALE_CTYPE
49c85077
KW
1317 Safefree(curctype);
1318 curctype = savepv(setlocale(LC_CTYPE, NULL));
5d1187d1 1319 DEBUG_LOCALE_INIT(LC_CTYPE, NULL, curctype);
98994639
HS
1320#endif /* USE_LOCALE_CTYPE */
1321#ifdef USE_LOCALE_COLLATE
49c85077
KW
1322 Safefree(curcoll);
1323 curcoll = savepv(setlocale(LC_COLLATE, NULL));
5d1187d1 1324 DEBUG_LOCALE_INIT(LC_COLLATE, NULL, curcoll);
98994639
HS
1325#endif /* USE_LOCALE_COLLATE */
1326#ifdef USE_LOCALE_NUMERIC
49c85077
KW
1327 Safefree(curnum);
1328 curnum = savepv(setlocale(LC_NUMERIC, NULL));
5d1187d1 1329 DEBUG_LOCALE_INIT(LC_NUMERIC, NULL, curnum);
98994639 1330#endif /* USE_LOCALE_NUMERIC */
65ebb059
KW
1331 }
1332
1333 if (locwarn) {
1334 const char * description;
1335 const char * name = "";
1336 if (strEQ(trial_locales[i], "C")) {
1337 description = "the standard locale";
1338 name = "C";
1339 }
6bce99ee 1340#ifdef SYSTEM_DEFAULT_LOCALE
65ebb059
KW
1341 else if (strEQ(trial_locales[i], "")) {
1342 description = "the system default locale";
1343 if (system_default_locale) {
1344 name = system_default_locale;
1345 }
1346 }
6bce99ee 1347#endif /* SYSTEM_DEFAULT_LOCALE */
65ebb059
KW
1348 else {
1349 description = "a fallback locale";
1350 name = trial_locales[i];
1351 }
1352 if (name && strNE(name, "")) {
1353 PerlIO_printf(Perl_error_log,
1354 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
1355 }
1356 else {
1357 PerlIO_printf(Perl_error_log,
1358 "perl: warning: %s %s.\n", msg, description);
1359 }
1360 }
1361 } /* End of tried to fallback */
98994639
HS
1362
1363#ifdef USE_LOCALE_CTYPE
1364 new_ctype(curctype);
1365#endif /* USE_LOCALE_CTYPE */
1366
1367#ifdef USE_LOCALE_COLLATE
1368 new_collate(curcoll);
1369#endif /* USE_LOCALE_COLLATE */
1370
1371#ifdef USE_LOCALE_NUMERIC
1372 new_numeric(curnum);
1373#endif /* USE_LOCALE_NUMERIC */
b310b053 1374
8ef6e574 1375#if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
49c85077
KW
1376 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
1377 * locale is UTF-8. If PL_utf8locale and PL_unicode (set by -C or by
1378 * $ENV{PERL_UNICODE}) are true, perl.c:S_parse_body() will turn on the
1379 * PerlIO :utf8 layer on STDIN, STDOUT, STDERR, _and_ the default open
1380 * discipline. */
c1284011 1381 PL_utf8locale = _is_cur_LC_category_utf8(LC_CTYPE);
49c85077 1382
a05d7ebb 1383 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
1384 This is an alternative to using the -C command line switch
1385 (the -C if present will override this). */
1386 {
dd374669 1387 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 1388 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
1389 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
1390 PL_utf8cache = -1;
b310b053 1391 }
ec71e770 1392#endif
b310b053 1393
98994639 1394#ifdef USE_LOCALE_CTYPE
43c5f42d 1395 Safefree(curctype);
98994639
HS
1396#endif /* USE_LOCALE_CTYPE */
1397#ifdef USE_LOCALE_COLLATE
43c5f42d 1398 Safefree(curcoll);
98994639
HS
1399#endif /* USE_LOCALE_COLLATE */
1400#ifdef USE_LOCALE_NUMERIC
43c5f42d 1401 Safefree(curnum);
98994639 1402#endif /* USE_LOCALE_NUMERIC */
8ef6e574 1403
175c4cf9
KW
1404#ifdef __GLIBC__
1405 Safefree(language);
1406#endif
1407
1408 Safefree(lc_all);
1409 Safefree(lang);
1410
e3305790
KW
1411#else /* !USE_LOCALE */
1412 PERL_UNUSED_ARG(printwarn);
1413#endif /* USE_LOCALE */
1414
2fcc0ca9
KW
1415#ifdef DEBUGGING
1416 /* So won't continue to output stuff */
27cdc72e 1417 DEBUG_INITIALIZATION_set(FALSE);
2fcc0ca9
KW
1418#endif
1419
98994639
HS
1420 return ok;
1421}
1422
98994639
HS
1423#ifdef USE_LOCALE_COLLATE
1424
a4a439fb 1425char *
a4a439fb
KW
1426Perl__mem_collxfrm(pTHX_ const char *input_string,
1427 STRLEN len, /* Length of 'input_string' */
1428 STRLEN *xlen, /* Set to length of returned string
1429 (not including the collation index
1430 prefix) */
1431 bool utf8 /* Is the input in UTF-8? */
6696cfa7 1432 )
98994639 1433{
a4a439fb
KW
1434
1435 /* _mem_collxfrm() is a bit like strxfrm() but with two important
1436 * differences. First, it handles embedded NULs. Second, it allocates a bit
1437 * more memory than needed for the transformed data itself. The real
55e5378d 1438 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
1439 * the length of that, and doesn't include the collation index size.
1440 * Please see sv_collxfrm() to see how this is used. */
1441
55e5378d
KW
1442#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
1443
6696cfa7
KW
1444 char * s = (char *) input_string;
1445 STRLEN s_strlen = strlen(input_string);
79f120c8 1446 char *xbuf = NULL;
55e5378d 1447 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 1448 STRLEN length_in_chars;
c664130f 1449 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 1450
a4a439fb
KW
1451 PERL_ARGS_ASSERT__MEM_COLLXFRM;
1452
1453 /* Must be NUL-terminated */
1454 assert(*(input_string + len) == '\0');
7918f24d 1455
79f120c8
KW
1456 /* If this locale has defective collation, skip */
1457 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
1458 DEBUG_L(PerlIO_printf(Perl_debug_log,
1459 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
1460 goto bad;
1461 }
1462
6696cfa7
KW
1463 /* Replace any embedded NULs with the control that sorts before any others.
1464 * This will give as good as possible results on strings that don't
1465 * otherwise contain that character, but otherwise there may be
1466 * less-than-perfect results with that character and NUL. This is
fdc080f3 1467 * unavoidable unless we replace strxfrm with our own implementation. */
6696cfa7
KW
1468 if (s_strlen < len) {
1469 char * e = s + len;
1470 char * sans_nuls;
1471 STRLEN cur_min_char_len;
fdc080f3
KW
1472 STRLEN sans_nuls_len;
1473 STRLEN sans_nuls_pos;
94762aa0 1474 int try_non_controls;
6696cfa7
KW
1475
1476 /* If we don't know what control character sorts lowest for this
1477 * locale, find it */
1478 if (*PL_strxfrm_min_char == '\0') {
1479 int j;
58eebef2 1480#ifdef DEBUGGING
d4ff9586 1481 U8 cur_min_cp = 1; /* The code point that sorts lowest, so far */
58eebef2
KW
1482#endif
1483 char * cur_min_x = NULL; /* And its xfrm, (except it also
6696cfa7
KW
1484 includes the collation index
1485 prefixed. */
1486
91c0e2e0 1487 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
1488
1489 /* Unlikely, but it may be that no control will work to replace
1490 * NUL, in which case we instead look for any character */
1491 for (try_non_controls = 0;
1492 try_non_controls < 2;
1493 try_non_controls++)
1494 {
d4ff9586
KW
1495 /* Look through all legal code points (NUL isn't) */
1496 for (j = 1; j < 256; j++) {
1497 char * x; /* j's xfrm plus collation index */
1498 STRLEN x_len; /* length of 'x' */
1499 STRLEN trial_len = 1;
1500
1501 /* Create a 1 byte string of the current code point, but
1502 * with room to be 2 bytes */
1503 char cur_source[] = { (char) j, '\0' , '\0' };
1504
1505 if (PL_in_utf8_COLLATE_locale) {
1506 if (! try_non_controls && ! isCNTRL_L1(j)) {
1507 continue;
1508 }
6696cfa7 1509
d4ff9586
KW
1510 /* If needs to be 2 bytes, find them */
1511 if (! UVCHR_IS_INVARIANT(j)) {
1512 char * d = cur_source;
1513 append_utf8_from_native_byte((U8) j, (U8 **) &d);
1514 trial_len = 2;
1515 }
1516 }
1517 else if (! try_non_controls && ! isCNTRL_LC(j)) {
1518 continue;
6696cfa7 1519 }
6696cfa7 1520
d4ff9586
KW
1521 /* Then transform it */
1522 x = _mem_collxfrm(cur_source, trial_len, &x_len,
1523 PL_in_utf8_COLLATE_locale);
6696cfa7 1524
d4ff9586
KW
1525 /* Ignore any character that didn't successfully transform
1526 * */
1527 if (! x) {
1528 continue;
1529 }
6696cfa7 1530
d4ff9586
KW
1531 /* If this character's transformation is lower than
1532 * the current lowest, this one becomes the lowest */
1533 if ( cur_min_x == NULL
1534 || strLT(x + COLLXFRM_HDR_LEN,
1535 cur_min_x + COLLXFRM_HDR_LEN))
1536 {
1537 PL_strxfrm_min_char[0] = cur_source[0];
1538 PL_strxfrm_min_char[1] = cur_source[1];
1539 PL_strxfrm_min_char[2] = cur_source[2];
1540 cur_min_x = x;
58eebef2 1541#ifdef DEBUGGING
d4ff9586 1542 cur_min_cp = j;
58eebef2 1543#endif
d4ff9586
KW
1544 }
1545 else {
1546 Safefree(x);
1547 }
1548 } /* end of loop through all bytes */
6696cfa7 1549
94762aa0
KW
1550 if (cur_min_x) {
1551 break;
1552 }
1553
1554 /* Unlikely, but possible, if there aren't any controls that
1555 * work in the locale, repeat the loop, looking for any
1556 * character that works */
1557 DEBUG_L(PerlIO_printf(Perl_debug_log,
1558 "_mem_collxfrm: No control worked. Trying non-controls\n"));
6696cfa7
KW
1559 }
1560
94762aa0
KW
1561 if (! cur_min_x) {
1562 DEBUG_L(PerlIO_printf(Perl_debug_log,
1563 "_mem_collxfrm: Couldn't find any character to replace"
1564 " embedded NULs in locale %s with", PL_collation_name));
1565 goto bad;
58eebef2
KW
1566 }
1567
94762aa0
KW
1568 DEBUG_L(PerlIO_printf(Perl_debug_log,
1569 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
1570 "0x%02X\n", PL_collation_name, cur_min_cp));
1571
6696cfa7
KW
1572 Safefree(cur_min_x);
1573 }
1574
1575 /* The worst case length for the replaced string would be if every
1576 * character in it is NUL. Multiply that by the length of each
1577 * replacement, and allow for a trailing NUL */
1578 cur_min_char_len = strlen(PL_strxfrm_min_char);
fdc080f3
KW
1579 sans_nuls_len = (len * cur_min_char_len) + 1;
1580 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7 1581 *sans_nuls = '\0';
fdc080f3 1582 sans_nuls_pos = 0;
6696cfa7 1583
6696cfa7
KW
1584 /* Replace each NUL with the lowest collating control. Loop until have
1585 * exhausted all the NULs */
1586 while (s + s_strlen < e) {
fdc080f3
KW
1587 sans_nuls_pos = my_strlcat(sans_nuls + sans_nuls_pos,
1588 s,
1589 sans_nuls_len);
6696cfa7
KW
1590
1591 /* Do the actual replacement */
fdc080f3
KW
1592 sans_nuls_pos = my_strlcat(sans_nuls + sans_nuls_pos,
1593 PL_strxfrm_min_char,
1594 sans_nuls_len);
6696cfa7
KW
1595
1596 /* Move past the input NUL */
1597 s += s_strlen + 1;
1598 s_strlen = strlen(s);
1599 }
1600
1601 /* And add anything that trails the final NUL */
fdc080f3 1602 my_strlcat(sans_nuls + sans_nuls_pos, s, sans_nuls_len);
6696cfa7
KW
1603
1604 /* Switch so below we transform this modified string */
1605 s = sans_nuls;
1606 len = strlen(s);
1607 }
1608
a4a439fb
KW
1609 /* Make sure the UTF8ness of the string and locale match */
1610 if (utf8 != PL_in_utf8_COLLATE_locale) {
1611 const char * const t = s; /* Temporary so we can later find where the
1612 input was */
1613
1614 /* Here they don't match. Change the string's to be what the locale is
1615 * expecting */
1616
1617 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
1618 s = (char *) bytes_to_utf8((const U8 *) s, &len);
1619 utf8 = TRUE;
1620 }
1621 else { /* locale is not UTF-8; but input is; downgrade the input */
1622
1623 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
1624
1625 /* If the downgrade was successful we are done, but if the input
1626 * contains things that require UTF-8 to represent, have to do
1627 * damage control ... */
1628 if (UNLIKELY(utf8)) {
1629
1630 /* What we do is construct a non-UTF-8 string with
1631 * 1) the characters representable by a single byte converted
1632 * to be so (if necessary);
1633 * 2) and the rest converted to collate the same as the
1634 * highest collating representable character. That makes
1635 * them collate at the end. This is similar to how we
1636 * handle embedded NULs, but we use the highest collating
1637 * code point instead of the smallest. Like the NUL case,
1638 * this isn't perfect, but is the best we can reasonably
1639 * do. Every above-255 code point will sort the same as
1640 * the highest-sorting 0-255 code point. If that code
1641 * point can combine in a sequence with some other code
1642 * points for weight calculations, us changing something to
1643 * be it can adversely affect the results. But in most
1644 * cases, it should work reasonably. And note that this is
1645 * really an illegal situation: using code points above 255
1646 * on a locale where only 0-255 are valid. If two strings
1647 * sort entirely equal, then the sort order for the
1648 * above-255 code points will be in code point order. */
1649
1650 utf8 = FALSE;
1651
1652 /* If we haven't calculated the code point with the maximum
1653 * collating order for this locale, do so now */
1654 if (! PL_strxfrm_max_cp) {
1655 int j;
1656
1657 /* The current transformed string that collates the
1658 * highest (except it also includes the prefixed collation
1659 * index. */
1660 char * cur_max_x = NULL;
1661
1662 /* Look through all legal code points (NUL isn't) */
1663 for (j = 1; j < 256; j++) {
1664 char * x;
1665 STRLEN x_len;
1666
1667 /* Create a 1-char string of the current code point. */
1668 char cur_source[] = { (char) j, '\0' };
1669
1670 /* Then transform it */
1671 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
1672
1673 /* If something went wrong (which it shouldn't), just
1674 * ignore this code point */
94762aa0 1675 if (! x) {
a4a439fb
KW
1676 continue;
1677 }
1678
1679 /* If this character's transformation is higher than
1680 * the current highest, this one becomes the highest */
1681 if ( cur_max_x == NULL
55e5378d
KW
1682 || strGT(x + COLLXFRM_HDR_LEN,
1683 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
1684 {
1685 PL_strxfrm_max_cp = j;
1686 cur_max_x = x;
1687 }
1688 else {
1689 Safefree(x);
1690 }
1691 }
1692
94762aa0
KW
1693 if (! cur_max_x) {
1694 DEBUG_L(PerlIO_printf(Perl_debug_log,
1695 "_mem_collxfrm: Couldn't find any character to"
1696 " replace above-Latin1 chars in locale %s with",
1697 PL_collation_name));
1698 goto bad;
1699 }
1700
58eebef2
KW
1701 DEBUG_L(PerlIO_printf(Perl_debug_log,
1702 "_mem_collxfrm: highest 1-byte collating character"
1703 " in locale %s is 0x%02X\n",
1704 PL_collation_name,
1705 PL_strxfrm_max_cp));
58eebef2 1706
a4a439fb
KW
1707 Safefree(cur_max_x);
1708 }
1709
1710 /* Here we know which legal code point collates the highest.
1711 * We are ready to construct the non-UTF-8 string. The length
1712 * will be at least 1 byte smaller than the input string
1713 * (because we changed at least one 2-byte character into a
1714 * single byte), but that is eaten up by the trailing NUL */
1715 Newx(s, len, char);
1716
1717 {
1718 STRLEN i;
1719 STRLEN d= 0;
1720
1721 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
1722 U8 cur_char = t[i];
1723 if (UTF8_IS_INVARIANT(cur_char)) {
1724 s[d++] = cur_char;
1725 }
1726 else if (UTF8_IS_DOWNGRADEABLE_START(cur_char)) {
1727 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
1728 }
1729 else { /* Replace illegal cp with highest collating
1730 one */
1731 s[d++] = PL_strxfrm_max_cp;
1732 }
1733 }
1734 s[d++] = '\0';
1735 Renew(s, d, char); /* Free up unused space */
1736 }
1737 }
1738 }
1739
1740 /* Here, we have constructed a modified version of the input. It could
1741 * be that we already had a modified copy before we did this version.
1742 * If so, that copy is no longer needed */
1743 if (t != input_string) {
1744 Safefree(t);
1745 }
1746 }
1747
17f41037
KW
1748 length_in_chars = (utf8)
1749 ? utf8_length((U8 *) s, (U8 *) s + len)
1750 : len;
1751
59c018b9
KW
1752 /* The first element in the output is the collation id, used by
1753 * sv_collxfrm(); then comes the space for the transformed string. The
1754 * equation should give us a good estimate as to how much is needed */
55e5378d 1755 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 1756 + PL_collxfrm_base
17f41037 1757 + (PL_collxfrm_mult * length_in_chars);
a02a5408 1758 Newx(xbuf, xAlloc, char);
c7202dee
KW
1759 if (UNLIKELY(! xbuf)) {
1760 DEBUG_L(PerlIO_printf(Perl_debug_log,
1761 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 1762 goto bad;
c7202dee 1763 }
98994639 1764
d35fca5f 1765 /* Store the collation id */
98994639 1766 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
1767
1768 /* Then the transformation of the input. We loop until successful, or we
1769 * give up */
4ebeff16 1770 for (;;) {
1adab0a7 1771
55e5378d 1772 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
1773
1774 /* If the transformed string occupies less space than we told strxfrm()
1775 * was available, it means it successfully transformed the whole
1776 * string. */
55e5378d 1777 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 1778
1adab0a7
KW
1779 /* Some systems include a trailing NUL in the returned length.
1780 * Ignore it, using a loop in case multiple trailing NULs are
1781 * returned. */
1782 while ( (*xlen) > 0
1783 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
1784 {
1785 (*xlen)--;
1786 }
1787
17f41037
KW
1788 /* If the first try didn't get it, it means our prediction was low.
1789 * Modify the coefficients so that we predict a larger value in any
1790 * future transformations */
1791 if (! first_time) {
1792 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
1793 STRLEN computed_guess = PL_collxfrm_base
1794 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
1795
1796 /* On zero-length input, just keep current slope instead of
1797 * dividing by 0 */
1798 const STRLEN new_m = (length_in_chars != 0)
1799 ? needed / length_in_chars
1800 : PL_collxfrm_mult;
17f41037
KW
1801
1802 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
1803 "%s: %d: initial size of %zu bytes for a length "
1804 "%zu string was insufficient, %zu needed\n",
17f41037 1805 __FILE__, __LINE__,
b07929e4 1806 computed_guess, length_in_chars, needed));
17f41037
KW
1807
1808 /* If slope increased, use it, but discard this result for
1809 * length 1 strings, as we can't be sure that it's a real slope
1810 * change */
1811 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
1812#ifdef DEBUGGING
1813 STRLEN old_m = PL_collxfrm_mult;
1814 STRLEN old_b = PL_collxfrm_base;
1815#endif
1816 PL_collxfrm_mult = new_m;
1817 PL_collxfrm_base = 1; /* +1 For trailing NUL */
1818 computed_guess = PL_collxfrm_base
1819 + (PL_collxfrm_mult * length_in_chars);
1820 if (computed_guess < needed) {
1821 PL_collxfrm_base += needed - computed_guess;
1822 }
1823
1824 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
1825 "%s: %d: slope is now %zu; was %zu, base "
1826 "is now %zu; was %zu\n",
17f41037 1827 __FILE__, __LINE__,
b07929e4
KW
1828 PL_collxfrm_mult, old_m,
1829 PL_collxfrm_base, old_b));
17f41037
KW
1830 }
1831 else { /* Slope didn't change, but 'b' did */
1832 const STRLEN new_b = needed
1833 - computed_guess
1834 + PL_collxfrm_base;
1835 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 1836 "%s: %d: base is now %zu; was %zu\n",
17f41037 1837 __FILE__, __LINE__,
b07929e4 1838 new_b, PL_collxfrm_base));
17f41037
KW
1839 PL_collxfrm_base = new_b;
1840 }
1841 }
1842
4ebeff16
KW
1843 break;
1844 }
bb0f664e 1845
c7202dee
KW
1846 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
1847 DEBUG_L(PerlIO_printf(Perl_debug_log,
1848 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
1849 *xlen, PERL_INT_MAX));
4ebeff16 1850 goto bad;
c7202dee 1851 }
d35fca5f 1852
c664130f 1853 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
1854 * (usually not including the trailing NUL) when it fails due to not
1855 * enough space being provided. Assume that this is the case unless
1856 * it's been proven otherwise */
c664130f 1857 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 1858 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
1859 }
1860 else { /* Here, either:
1861 * 1) The strxfrm() has previously shown bad behavior; or
1862 * 2) It isn't the first time through the loop, which means
1863 * that the strxfrm() is now showing bad behavior, because
1864 * we gave it what it said was needed in the previous
1865 * iteration, and it came back saying it needed still more.
1866 * (Many versions of cygwin fit this. When the buffer size
1867 * isn't sufficient, they return the input size instead of
1868 * how much is needed.)
d4ff9586
KW
1869 * Increase the buffer size by a fixed percentage and try again.
1870 * */
6ddd902c 1871 xAlloc += (xAlloc / 4) + 1;
c664130f 1872 PL_strxfrm_is_behaved = FALSE;
c664130f 1873
58eebef2
KW
1874#ifdef DEBUGGING
1875 if (DEBUG_Lv_TEST || debug_initialization) {
1876 PerlIO_printf(Perl_debug_log,
1877 "_mem_collxfrm required more space than previously calculated"
b07929e4 1878 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 1879 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 1880 xAlloc - COLLXFRM_HDR_LEN);
58eebef2
KW
1881 }
1882#endif
1883 }
c664130f 1884
4ebeff16 1885 Renew(xbuf, xAlloc, char);
c7202dee
KW
1886 if (UNLIKELY(! xbuf)) {
1887 DEBUG_L(PerlIO_printf(Perl_debug_log,
1888 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 1889 goto bad;
c7202dee 1890 }
c664130f
KW
1891
1892 first_time = FALSE;
4ebeff16 1893 }
98994639 1894
6696cfa7 1895
58eebef2
KW
1896#ifdef DEBUGGING
1897 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
1898 Size_t i;
1899
1900 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
1901 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
58eebef2
KW
1902 for (i = COLLXFRM_HDR_LEN; i < *xlen + COLLXFRM_HDR_LEN; i++) {
1903 PerlIO_printf(Perl_debug_log, " %02x", (U8) xbuf[i]);
1904 }
1905 PerlIO_printf(Perl_debug_log, "\n");
1906 }
1907#endif
1908
3c5f993e 1909 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 1910 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 1911
6696cfa7
KW
1912 if (s != input_string) {
1913 Safefree(s);
98994639
HS
1914 }
1915
98994639
HS
1916 return xbuf;
1917
1918 bad:
1919 Safefree(xbuf);
6696cfa7
KW
1920 if (s != input_string) {
1921 Safefree(s);
1922 }
98994639 1923 *xlen = 0;
58eebef2
KW
1924#ifdef DEBUGGING
1925 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 1926 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2
KW
1927 }
1928#endif
98994639
HS
1929 return NULL;
1930}
1931
c7202dee
KW
1932#ifdef DEBUGGING
1933
4cbaac56 1934STATIC void
c7202dee
KW
1935S_print_collxfrm_input_and_return(pTHX_
1936 const char * const s,
1937 const char * const e,
1938 const STRLEN * const xlen,
1939 const bool is_utf8)
1940{
1941 const char * t = s;
1942 bool prev_was_printable = TRUE;
1943 bool first_time = TRUE;
1944
1945 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
1946
e5d8cfc5 1947 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%u]: returning ",
c7202dee
KW
1948 PL_collation_ix);
1949 if (xlen) {
1950 PerlIO_printf(Perl_debug_log, "%"UVuf"", (UV) *xlen);
1951 }
1952 else {
1953 PerlIO_printf(Perl_debug_log, "NULL");
1954 }
1955 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
1956 PL_collation_name);
1957
1958 while (t < e) {
1959 UV cp = (is_utf8)
1960 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
1961 : * (U8 *) t;
1962 if (isPRINT(cp)) {
1963 if (! prev_was_printable) {
1964 PerlIO_printf(Perl_debug_log, " ");
1965 }
1966 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
1967 prev_was_printable = TRUE;
1968 }
1969 else {
1970 if (! first_time) {
1971 PerlIO_printf(Perl_debug_log, " ");
1972 }
1973 PerlIO_printf(Perl_debug_log, "%02"UVXf"", cp);
1974 prev_was_printable = FALSE;
1975 }
1976 t += (is_utf8) ? UTF8SKIP(t) : 1;
1977 first_time = FALSE;
1978 }
1979
1980 PerlIO_printf(Perl_debug_log, "'\n");
1981}
1982
1983#endif /* #ifdef DEBUGGING */
1984
98994639 1985#endif /* USE_LOCALE_COLLATE */
58eebef2 1986
8ef6e574
KW
1987#ifdef USE_LOCALE
1988
c1284011
KW
1989bool
1990Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
1991{
1992 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
1993 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 1994 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
1995 * could give the wrong result. The result will very likely be correct for
1996 * languages that have commonly used non-ASCII characters, but for notably
1997 * English, it comes down to if the locale's name ends in something like
1998 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61
KW
1999
2000 char *save_input_locale = NULL;
7d74bb61
KW
2001 STRLEN final_pos;
2002
8ef6e574 2003#ifdef LC_ALL
7d74bb61 2004 assert(category != LC_ALL);
8ef6e574 2005#endif
7d74bb61
KW
2006
2007 /* First dispose of the trivial cases */
b07fffd1 2008 save_input_locale = setlocale(category, NULL);
7d74bb61 2009 if (! save_input_locale) {
69014004
KW
2010 DEBUG_L(PerlIO_printf(Perl_debug_log,
2011 "Could not find current locale for category %d\n",
2012 category));
7d74bb61
KW
2013 return FALSE; /* XXX maybe should croak */
2014 }
b07fffd1 2015 save_input_locale = stdize_locale(savepv(save_input_locale));
a39edc4c 2016 if (isNAME_C_OR_POSIX(save_input_locale)) {
69014004
KW
2017 DEBUG_L(PerlIO_printf(Perl_debug_log,
2018 "Current locale for category %d is %s\n",
2019 category, save_input_locale));
b07fffd1 2020 Safefree(save_input_locale);
7d74bb61
KW
2021 return FALSE;
2022 }
2023
1d958db2
KW
2024#if defined(USE_LOCALE_CTYPE) \
2025 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 2026
1d958db2 2027 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
2028
2029 char *save_ctype_locale = NULL;
119ee68b 2030 bool is_utf8;
7d74bb61 2031
119ee68b 2032 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
2033
2034 /* Get the current LC_CTYPE locale */
4f72bb37 2035 save_ctype_locale = setlocale(LC_CTYPE, NULL);
7d74bb61 2036 if (! save_ctype_locale) {
69014004
KW
2037 DEBUG_L(PerlIO_printf(Perl_debug_log,
2038 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
2039 goto cant_use_nllanginfo;
2040 }
4f72bb37 2041 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
2042
2043 /* If LC_CTYPE and the desired category use the same locale, this
2044 * means that finding the value for LC_CTYPE is the same as finding
2045 * the value for the desired category. Otherwise, switch LC_CTYPE
2046 * to the desired category's locale */
2047 if (strEQ(save_ctype_locale, save_input_locale)) {
2048 Safefree(save_ctype_locale);
2049 save_ctype_locale = NULL;
2050 }
2051 else if (! setlocale(LC_CTYPE, save_input_locale)) {
69014004
KW
2052 DEBUG_L(PerlIO_printf(Perl_debug_log,
2053 "Could not change LC_CTYPE locale to %s\n",
2054 save_input_locale));
7d74bb61
KW
2055 Safefree(save_ctype_locale);
2056 goto cant_use_nllanginfo;
2057 }
2058 }
2059
69014004
KW
2060 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
2061 save_input_locale));
2062
7d74bb61 2063 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
2064 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
2065 * should give the correct results */
119ee68b 2066
1d958db2
KW
2067# if defined(HAS_NL_LANGINFO) && defined(CODESET)
2068 {
4f72bb37 2069 char *codeset = nl_langinfo(CODESET);
1d958db2 2070 if (codeset && strNE(codeset, "")) {
4f72bb37 2071 codeset = savepv(codeset);
119ee68b 2072
1d958db2
KW
2073 /* If we switched LC_CTYPE, switch back */
2074 if (save_ctype_locale) {
2075 setlocale(LC_CTYPE, save_ctype_locale);
2076 Safefree(save_ctype_locale);
2077 }
2078
2079 is_utf8 = foldEQ(codeset, STR_WITH_LEN("UTF-8"))
2080 || foldEQ(codeset, STR_WITH_LEN("UTF8"));
2081
69014004
KW
2082 DEBUG_L(PerlIO_printf(Perl_debug_log,
2083 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
2084 codeset, is_utf8));
1d958db2
KW
2085 Safefree(codeset);
2086 Safefree(save_input_locale);
2087 return is_utf8;
2088 }
119ee68b
KW
2089 }
2090
1d958db2
KW
2091# endif
2092# ifdef MB_CUR_MAX
2093
2094 /* Here, either we don't have nl_langinfo, or it didn't return a
2095 * codeset. Try MB_CUR_MAX */
2096
119ee68b
KW
2097 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
2098 * Unicode code point. Since UTF-8 is the only non-single byte
2099 * encoding we handle, we just say any such encoding is UTF-8, and if
2100 * turns out to be wrong, other things will fail */
2101 is_utf8 = MB_CUR_MAX >= 4;
2102
69014004
KW
2103 DEBUG_L(PerlIO_printf(Perl_debug_log,
2104 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
2105 (int) MB_CUR_MAX, is_utf8));
2106
119ee68b
KW
2107 Safefree(save_input_locale);
2108
2109# ifdef HAS_MBTOWC
2110
2111 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
2112 * since they are both in the C99 standard. We can feed a known byte
2113 * string to the latter function, and check that it gives the expected
2114 * result */
2115 if (is_utf8) {
2116 wchar_t wc;
856b881c 2117 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 2118 errno = 0;
f019f68f 2119 if ((size_t)mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8))
119ee68b
KW
2120 != strlen(HYPHEN_UTF8)
2121 || wc != (wchar_t) 0x2010)
2122 {
2123 is_utf8 = FALSE;
abdcbdb8 2124 DEBUG_L(PerlIO_printf(Perl_debug_log, "\thyphen=U+%x\n", (unsigned int)wc));
69014004
KW
2125 DEBUG_L(PerlIO_printf(Perl_debug_log,
2126 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
2127 mbtowc(&wc, HYPHEN_UTF8, strlen(HYPHEN_UTF8)), errno));
119ee68b
KW
2128 }
2129 }
119ee68b
KW
2130# endif
2131
1d958db2
KW
2132 /* If we switched LC_CTYPE, switch back */
2133 if (save_ctype_locale) {
2134 setlocale(LC_CTYPE, save_ctype_locale);
2135 Safefree(save_ctype_locale);
119ee68b 2136 }
7d74bb61 2137
1d958db2 2138 return is_utf8;
119ee68b 2139# endif
7d74bb61 2140 }
119ee68b 2141
7d74bb61
KW
2142 cant_use_nllanginfo:
2143
0080c90a
KW
2144#else /* nl_langinfo should work if available, so don't bother compiling this
2145 fallback code. The final fallback of looking at the name is
2146 compiled, and will be executed if nl_langinfo fails */
7d74bb61 2147
97f4de96
KW
2148 /* nl_langinfo not available or failed somehow. Next try looking at the
2149 * currency symbol to see if it disambiguates things. Often that will be
2150 * in the native script, and if the symbol isn't in UTF-8, we know that the
2151 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
2152 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
2153 * */
fa9b773e
KW
2154
2155#ifdef HAS_LOCALECONV
fa9b773e 2156# ifdef USE_LOCALE_MONETARY
fa9b773e
KW
2157 {
2158 char *save_monetary_locale = NULL;
fa9b773e 2159 bool only_ascii = FALSE;
13542a67
KW
2160 bool is_utf8 = FALSE;
2161 struct lconv* lc;
fa9b773e 2162
97f4de96
KW
2163 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
2164 * the desired category, if it isn't that locale already */
2165
fa9b773e
KW
2166 if (category != LC_MONETARY) {
2167
4f72bb37 2168 save_monetary_locale = setlocale(LC_MONETARY, NULL);
fa9b773e 2169 if (! save_monetary_locale) {
69014004
KW
2170 DEBUG_L(PerlIO_printf(Perl_debug_log,
2171 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
2172 goto cant_use_monetary;
2173 }
4f72bb37 2174 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 2175
13542a67
KW
2176 if (strEQ(save_monetary_locale, save_input_locale)) {
2177 Safefree(save_monetary_locale);
2178 save_monetary_locale = NULL;
2179 }
2180 else if (! setlocale(LC_MONETARY, save_input_locale)) {
59c234b4
KW
2181 DEBUG_L(PerlIO_printf(Perl_debug_log,
2182 "Could not change LC_MONETARY locale to %s\n",
2183 save_input_locale));
2184 Safefree(save_monetary_locale);
2185 goto cant_use_monetary;
fa9b773e
KW
2186 }
2187 }
2188
2189 /* Here the current LC_MONETARY is set to the locale of the category
2190 * whose information is desired. */
2191
13542a67
KW
2192 lc = localeconv();
2193 if (! lc
2194 || ! lc->currency_symbol
c5f058df 2195 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
2196 {
2197 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));
2198 only_ascii = TRUE;
2199 }
2200 else {
2201 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
2202 }
2203
2204 /* If we changed it, restore LC_MONETARY to its original locale */
2205 if (save_monetary_locale) {
2206 setlocale(LC_MONETARY, save_monetary_locale);
2207 Safefree(save_monetary_locale);
2208 }
2209
13542a67 2210 if (! only_ascii) {
fa9b773e 2211
59c234b4
KW
2212 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
2213 * otherwise assume the locale is UTF-8 if and only if the symbol
2214 * is non-ascii UTF-8. */
2215 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
2216 save_input_locale, is_utf8));
2217 Safefree(save_input_locale);
2218 return is_utf8;
13542a67 2219 }
fa9b773e
KW
2220 }
2221 cant_use_monetary:
2222
2223# endif /* USE_LOCALE_MONETARY */
2224#endif /* HAS_LOCALECONV */
2225
15f7e74e
KW
2226#if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
2227
2228/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
2229 * the names of the months and weekdays, timezone, and am/pm indicator */
2230 {
2231 char *save_time_locale = NULL;
2232 int hour = 10;
2233 bool is_dst = FALSE;
2234 int dom = 1;
2235 int month = 0;
2236 int i;
2237 char * formatted_time;
2238
2239
2240 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
2241 * desired category, if it isn't that locale already */
2242
2243 if (category != LC_TIME) {
2244
2245 save_time_locale = setlocale(LC_TIME, NULL);
2246 if (! save_time_locale) {
2247 DEBUG_L(PerlIO_printf(Perl_debug_log,
2248 "Could not find current locale for LC_TIME\n"));
2249 goto cant_use_time;
2250 }
2251 save_time_locale = stdize_locale(savepv(save_time_locale));
2252
2253 if (strEQ(save_time_locale, save_input_locale)) {
2254 Safefree(save_time_locale);
2255 save_time_locale = NULL;
2256 }
2257 else if (! setlocale(LC_TIME, save_input_locale)) {
2258 DEBUG_L(PerlIO_printf(Perl_debug_log,
2259 "Could not change LC_TIME locale to %s\n",
2260 save_input_locale));
2261 Safefree(save_time_locale);
2262 goto cant_use_time;
2263 }
2264 }
2265
2266 /* Here the current LC_TIME is set to the locale of the category
2267 * whose information is desired. Look at all the days of the week and
9f10db87 2268 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
2269 * characters. The first such a one found will tell us if the locale
2270 * is UTF-8 or not */
2271
2272 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
2273 formatted_time = my_strftime("%A %B %Z %p",
2274 0, 0, hour, dom, month, 112, 0, 0, is_dst);
c5f058df
KW
2275 if ( ! formatted_time
2276 || is_utf8_invariant_string((U8 *) formatted_time, 0))
2277 {
15f7e74e
KW
2278
2279 /* Here, we didn't find a non-ASCII. Try the next time through
2280 * with the complemented dst and am/pm, and try with the next
2281 * weekday. After we have gotten all weekdays, try the next
2282 * month */
2283 is_dst = ! is_dst;
2284 hour = (hour + 12) % 24;
2285 dom++;
2286 if (i > 6) {
2287 month++;
2288 }
2289 continue;
2290 }
2291
2292 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
2293 * false otherwise. But first, restore LC_TIME to its original
2294 * locale if we changed it */
2295 if (save_time_locale) {
2296 setlocale(LC_TIME, save_time_locale);
2297 Safefree(save_time_locale);
2298 }
2299
2300 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
2301 save_input_locale,
2302 is_utf8_string((U8 *) formatted_time, 0)));
2303 Safefree(save_input_locale);
2304 return is_utf8_string((U8 *) formatted_time, 0);
2305 }
2306
2307 /* Falling off the end of the loop indicates all the names were just
2308 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
2309 * to its original locale */
2310 if (save_time_locale) {
2311 setlocale(LC_TIME, save_time_locale);
2312 Safefree(save_time_locale);
2313 }
2314 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));
2315 }
2316 cant_use_time:
2317
2318#endif
2319
2320#if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
2321
2322/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
2323 * on our dromedary test machine, which has over 700 locales. There, this
2324 * added no value to looking at the currency symbol and the time strings. I
2325 * left it in so as to avoid rewriting it if real-world experience indicates
2326 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
2327 * haven't found illegal utf8, we continue on and examine all the strerror()
2328 * messages on the platform for utf8ness. If all are ASCII, we still don't
2329 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
2330 * utf8ness. The reason this doesn't help much is that the messages may not
2331 * have been translated into the locale. The currency symbol and time strings
2332 * are much more likely to have been translated. */
2333 {
855aeb93 2334 int e;
5857e934
KW
2335 bool is_utf8 = FALSE;
2336 bool non_ascii = FALSE;
855aeb93 2337 char *save_messages_locale = NULL;
5857e934 2338 const char * errmsg = NULL;
855aeb93 2339
5857e934
KW
2340 /* Like above, we set LC_MESSAGES to the locale of the desired
2341 * category, if it isn't that locale already */
855aeb93
JH
2342
2343 if (category != LC_MESSAGES) {
2344
5857e934 2345 save_messages_locale = setlocale(LC_MESSAGES, NULL);
855aeb93 2346 if (! save_messages_locale) {
5857e934
KW
2347 DEBUG_L(PerlIO_printf(Perl_debug_log,
2348 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
2349 goto cant_use_messages;
2350 }
5857e934 2351 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
2352
2353 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
2354 Safefree(save_messages_locale);
2355 save_messages_locale = NULL;
855aeb93
JH
2356 }
2357 else if (! setlocale(LC_MESSAGES, save_input_locale)) {
5857e934
KW
2358 DEBUG_L(PerlIO_printf(Perl_debug_log,
2359 "Could not change LC_MESSAGES locale to %s\n",
2360 save_input_locale));
855aeb93
JH
2361 Safefree(save_messages_locale);
2362 goto cant_use_messages;
2363 }
2364 }
2365
2366 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
2367 * whose information is desired. Look through all the messages. We
2368 * can't use Strerror() here because it may expand to code that
2369 * segfaults in miniperl */
855aeb93 2370
5857e934
KW
2371 for (e = 0; e <= sys_nerr; e++) {
2372 errno = 0;
2373 errmsg = sys_errlist[e];
2374 if (errno || !errmsg) {
855aeb93
JH
2375 break;
2376 }
5857e934 2377 errmsg = savepv(errmsg);
c5f058df 2378 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
2379 non_ascii = TRUE;
2380 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
2381 break;
855aeb93
JH
2382 }
2383 }
5857e934 2384 Safefree(errmsg);
855aeb93
JH
2385
2386 /* And, if we changed it, restore LC_MESSAGES to its original locale */
2387 if (save_messages_locale) {
2388 setlocale(LC_MESSAGES, save_messages_locale);
2389 Safefree(save_messages_locale);
2390 }
2391
5857e934
KW
2392 if (non_ascii) {
2393
2394 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
2395 * any non-ascii means it is one; otherwise we assume it isn't */
2396 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
2397 save_input_locale,
2398 is_utf8));
2399 Safefree(save_input_locale);
2400 return is_utf8;
2401 }
855aeb93 2402
5857e934 2403 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
2404 }
2405 cant_use_messages:
2406
2407#endif
fa9b773e 2408
0080c90a
KW
2409#endif /* the code that is compiled when no nl_langinfo */
2410
92c0a900
KW
2411#ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
2412 UTF-8 locale */
97f4de96
KW
2413 /* As a last resort, look at the locale name to see if it matches
2414 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
2415 * return of setlocale(), is actually defined to be opaque, so we can't
2416 * really rely on the absence of various substrings in the name to indicate
2417 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
2418 * be a UTF-8 locale. Similarly for the other common names */
2419
2420 final_pos = strlen(save_input_locale) - 1;
2421 if (final_pos >= 3) {
2422 char *name = save_input_locale;
2423
2424 /* Find next 'U' or 'u' and look from there */
2425 while ((name += strcspn(name, "Uu") + 1)
2426 <= save_input_locale + final_pos - 2)
2427 {
305b8651
KW
2428 if (!isALPHA_FOLD_NE(*name, 't')
2429 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
2430 {
2431 continue;
2432 }
2433 name += 2;
2434 if (*(name) == '-') {
2435 if ((name > save_input_locale + final_pos - 1)) {
2436 break;
2437 }
2438 name++;
2439 }
2440 if (*(name) == '8') {
97f4de96
KW
2441 DEBUG_L(PerlIO_printf(Perl_debug_log,
2442 "Locale %s ends with UTF-8 in name\n",
2443 save_input_locale));
00c54b9c 2444 Safefree(save_input_locale);
97f4de96
KW
2445 return TRUE;
2446 }
2447 }
2448 DEBUG_L(PerlIO_printf(Perl_debug_log,
2449 "Locale %s doesn't end with UTF-8 in name\n",
2450 save_input_locale));
2451 }
92c0a900 2452#endif
97f4de96
KW
2453
2454#ifdef WIN32
2455 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
2456 if (final_pos >= 4
2457 && *(save_input_locale + final_pos - 0) == '1'
2458 && *(save_input_locale + final_pos - 1) == '0'
2459 && *(save_input_locale + final_pos - 2) == '0'
2460 && *(save_input_locale + final_pos - 3) == '5'
2461 && *(save_input_locale + final_pos - 4) == '6')
2462 {
2463 DEBUG_L(PerlIO_printf(Perl_debug_log,
2464 "Locale %s ends with 10056 in name, is UTF-8 locale\n",
2465 save_input_locale));
2466 Safefree(save_input_locale);
2467 return TRUE;
2468 }
2469#endif
2470
2471 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
2472 * since we are about to return FALSE anyway, there is no point in doing
2473 * this extra work */
2474#if 0
2475 if (instr(save_input_locale, "8859")) {
2476 DEBUG_L(PerlIO_printf(Perl_debug_log,
2477 "Locale %s has 8859 in name, not UTF-8 locale\n",
2478 save_input_locale));
2479 Safefree(save_input_locale);
2480 return FALSE;
2481 }
2482#endif
2483
69014004
KW
2484 DEBUG_L(PerlIO_printf(Perl_debug_log,
2485 "Assuming locale %s is not a UTF-8 locale\n",
2486 save_input_locale));
fa9b773e 2487 Safefree(save_input_locale);
7d74bb61
KW
2488 return FALSE;
2489}
2490
8ef6e574 2491#endif
7d74bb61 2492
d6ded950
KW
2493
2494bool
2495Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
2496{
1a4f13e1 2497 dVAR;
d6ded950
KW
2498 /* Internal function which returns if we are in the scope of a pragma that
2499 * enables the locale category 'category'. 'compiling' should indicate if
2500 * this is during the compilation phase (TRUE) or not (FALSE). */
2501
2502 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
2503
2504 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
2505 if (! categories || categories == &PL_sv_placeholder) {
2506 return FALSE;
2507 }
2508
2509 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
2510 * a valid unsigned */
2511 assert(category >= -1);
2512 return cBOOL(SvUV(categories) & (1U << (category + 1)));
2513}
2514
2c6ee1a7 2515char *
6ebbc862
KW
2516Perl_my_strerror(pTHX_ const int errnum)
2517{
2518 /* Returns a mortalized copy of the text of the error message associated
2519 * with 'errnum'. It uses the current locale's text unless the platform
2520 * doesn't have the LC_MESSAGES category or we are not being called from
2521 * within the scope of 'use locale'. In the former case, it uses whatever
2522 * strerror returns; in the latter case it uses the text from the C locale.
2523 *
2524 * The function just calls strerror(), but temporarily switches, if needed,
2525 * to the C locale */
2526
2527 char *errstr;
2528
2529#ifdef USE_LOCALE_MESSAGES /* If platform doesn't have messages category, we
2530 don't do any switching to the C locale; we just
2531 use whatever strerror() returns */
2532 const bool within_locale_scope = IN_LC(LC_MESSAGES);
2533
a0b53297 2534 dVAR;
2c6ee1a7 2535
6ebbc862
KW
2536# ifdef USE_THREAD_SAFE_LOCALE
2537 locale_t save_locale;
2538# else
2539 char * save_locale;
c9dda6da 2540 bool locale_is_C = FALSE;
2c6ee1a7 2541
6ebbc862
KW
2542 /* We have a critical section to prevent another thread from changing the
2543 * locale out from under us (or zapping the buffer returned from
2544 * setlocale() ) */
2545 LOCALE_LOCK;
2546
2547# endif
2548
2549 if (! within_locale_scope) {
c9dda6da 2550 errno = 0;
a0b53297 2551
6ebbc862
KW
2552# ifdef USE_THREAD_SAFE_LOCALE /* Use the thread-safe locale functions */
2553
2554 save_locale = uselocale(PL_C_locale_obj);
c9dda6da
KW
2555 if (! save_locale) {
2556 DEBUG_L(PerlIO_printf(Perl_debug_log,
2557 "uselocale failed, errno=%d\n", errno));
2558 }
6ebbc862
KW
2559
2560# else /* Not thread-safe build */
a0b53297
KW
2561
2562 save_locale = setlocale(LC_MESSAGES, NULL);
c9dda6da
KW
2563 if (! save_locale) {
2564 DEBUG_L(PerlIO_printf(Perl_debug_log,
2565 "setlocale failed, errno=%d\n", errno));
2566 }
2567 else {
2568 locale_is_C = isNAME_C_OR_POSIX(save_locale);
2c6ee1a7 2569
c9dda6da
KW
2570 /* Switch to the C locale if not already in it */
2571 if (! locale_is_C) {
2c6ee1a7 2572
c9dda6da
KW
2573 /* The setlocale() just below likely will zap 'save_locale', so
2574 * create a copy. */
2575 save_locale = savepv(save_locale);
2576 setlocale(LC_MESSAGES, "C");
2577 }
6ebbc862 2578 }
2c6ee1a7 2579
6ebbc862 2580# endif
2c6ee1a7 2581
6ebbc862 2582 } /* end of ! within_locale_scope */
a0b53297 2583
6ebbc862 2584#endif
a0b53297 2585
6ebbc862
KW
2586 errstr = Strerror(errnum);
2587 if (errstr) {
2588 errstr = savepv(errstr);
2589 SAVEFREEPV(errstr);
2590 }
2591
2592#ifdef USE_LOCALE_MESSAGES
2593
2594 if (! within_locale_scope) {
c9dda6da 2595 errno = 0;
a0b53297 2596
6ebbc862
KW
2597# ifdef USE_THREAD_SAFE_LOCALE
2598
c9dda6da
KW
2599 if (save_locale && ! uselocale(save_locale)) {
2600 DEBUG_L(PerlIO_printf(Perl_debug_log,
2601 "uselocale restore failed, errno=%d\n", errno));
2602 }
2c6ee1a7 2603 }
6ebbc862
KW
2604
2605# else
2606
c9dda6da
KW
2607 if (save_locale && ! locale_is_C) {
2608 if (! setlocale(LC_MESSAGES, save_locale)) {
2609 DEBUG_L(PerlIO_printf(Perl_debug_log,
2610 "setlocale restore failed, errno=%d\n", errno));
2611 }
6ebbc862
KW
2612 Safefree(save_locale);
2613 }
2614 }
2615
2616 LOCALE_UNLOCK;
2617
2618# endif
2c6ee1a7
KW
2619#endif
2620
6ebbc862 2621 return errstr;
2c6ee1a7
KW
2622}
2623
66610fdd 2624/*
747c467a
KW
2625
2626=head1 Locale-related functions and macros
2627
2628=for apidoc sync_locale
2629
2630Changing the program's locale should be avoided by XS code. Nevertheless,
2631certain non-Perl libraries called from XS, such as C<Gtk> do so. When this
2632happens, Perl needs to be told that the locale has changed. Use this function
2633to do so, before returning to Perl.
2634
2635=cut
2636*/
2637
2638void
2639Perl_sync_locale(pTHX)
2640{
2641
2642#ifdef USE_LOCALE_CTYPE
2643 new_ctype(setlocale(LC_CTYPE, NULL));
2644#endif /* USE_LOCALE_CTYPE */
2645
2646#ifdef USE_LOCALE_COLLATE
2647 new_collate(setlocale(LC_COLLATE, NULL));
2648#endif
2649
2650#ifdef USE_LOCALE_NUMERIC
2651 set_numeric_local(); /* Switch from "C" to underlying LC_NUMERIC */
2652 new_numeric(setlocale(LC_NUMERIC, NULL));
2653#endif /* USE_LOCALE_NUMERIC */
2654
2655}
2656
5d1187d1
KW
2657#if defined(DEBUGGING) && defined(USE_LOCALE)
2658
2659char *
2660Perl__setlocale_debug_string(const int category, /* category number,
2661 like LC_ALL */
2662 const char* const locale, /* locale name */
2663
2664 /* return value from setlocale() when attempting to
2665 * set 'category' to 'locale' */
2666 const char* const retval)
2667{
2668 /* Returns a pointer to a NUL-terminated string in static storage with
2669 * added text about the info passed in. This is not thread safe and will
2670 * be overwritten by the next call, so this should be used just to
fa07b8e5 2671 * formulate a string to immediately print or savepv() on. */
5d1187d1 2672
398a990f
DM
2673 /* initialise to a non-null value to keep it out of BSS and so keep
2674 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
60b45a7d
KW
2675 static char ret[128] = "If you can read this, thank your buggy C"
2676 " library strlcpy(), and change your hints file"
2677 " to undef it";
fa07b8e5 2678 my_strlcpy(ret, "setlocale(", sizeof(ret));
5d1187d1
KW
2679
2680 switch (category) {
2681 default:
fa07b8e5 2682 my_snprintf(ret, sizeof(ret), "%s? %d", ret, category);
5d1187d1
KW
2683 break;
2684# ifdef LC_ALL
2685 case LC_ALL:
fa07b8e5 2686 my_strlcat(ret, "LC_ALL", sizeof(ret));
5d1187d1
KW
2687 break;
2688# endif
2689# ifdef LC_CTYPE
2690 case LC_CTYPE:
fa07b8e5 2691 my_strlcat(ret, "LC_CTYPE", sizeof(ret));
5d1187d1
KW
2692 break;
2693# endif
2694# ifdef LC_NUMERIC
2695 case LC_NUMERIC:
fa07b8e5 2696 my_strlcat(ret, "LC_NUMERIC", sizeof(ret));
5d1187d1
KW
2697 break;
2698# endif
2699# ifdef LC_COLLATE
2700 case LC_COLLATE:
fa07b8e5 2701 my_strlcat(ret, "LC_COLLATE", sizeof(ret));
5d1187d1
KW
2702 break;
2703# endif
2704# ifdef LC_TIME
2705 case LC_TIME:
fa07b8e5 2706 my_strlcat(ret, "LC_TIME", sizeof(ret));
5d1187d1
KW
2707 break;
2708# endif
2709# ifdef LC_MONETARY
2710 case LC_MONETARY:
fa07b8e5 2711 my_strlcat(ret, "LC_MONETARY", sizeof(ret));
5d1187d1
KW
2712 break;
2713# endif
2714# ifdef LC_MESSAGES
2715 case LC_MESSAGES:
fa07b8e5 2716 my_strlcat(ret, "LC_MESSAGES", sizeof(ret));
5d1187d1
KW
2717 break;
2718# endif
2719 }
2720
fa07b8e5 2721 my_strlcat(ret, ", ", sizeof(ret));
5d1187d1
KW
2722
2723 if (locale) {
fa07b8e5
KW
2724 my_strlcat(ret, "\"", sizeof(ret));
2725 my_strlcat(ret, locale, sizeof(ret));
2726 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2727 }
2728 else {
fa07b8e5 2729 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2730 }
2731
fa07b8e5 2732 my_strlcat(ret, ") returned ", sizeof(ret));
5d1187d1
KW
2733
2734 if (retval) {
fa07b8e5
KW
2735 my_strlcat(ret, "\"", sizeof(ret));
2736 my_strlcat(ret, retval, sizeof(ret));
2737 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
2738 }
2739 else {
fa07b8e5 2740 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
2741 }
2742
2743 assert(strlen(ret) < sizeof(ret));
2744
2745 return ret;
2746}
2747
2748#endif
747c467a
KW
2749
2750
2751/*
14d04a33 2752 * ex: set ts=8 sts=4 sw=4 et:
37442d52 2753 */