This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Use fcns instead of macros
[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 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
166f8a29
DM
35 */
36
98994639
HS
37#include "EXTERN.h"
38#define PERL_IN_LOCALE_C
f7416781 39#include "perl_langinfo.h"
98994639
HS
40#include "perl.h"
41
a4af207c
JH
42#include "reentr.h"
43
2fcc0ca9
KW
44/* If the environment says to, we can output debugging information during
45 * initialization. This is done before option parsing, and before any thread
46 * creation, so can be a file-level static */
5a4b0634
KW
47#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
48# define debug_initialization 0
49# define DEBUG_INITIALIZATION_set(v)
50#else
2fcc0ca9 51static bool debug_initialization = FALSE;
5a4b0634 52# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
53#endif
54
291a84fb
KW
55/* strlen() of a literal string constant. We might want this more general,
56 * but using it in just this file for now. A problem with more generality is
57 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
58#define STRLENs(s) (sizeof("" s "") - 1)
59
63e5b0d7
KW
60/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
61 * return of setlocale(), then this is extremely likely to be the C or POSIX
62 * locale. However, the output of setlocale() is documented to be opaque, but
63 * the odds are extremely small that it would return these two strings for some
64 * other locale. Note that VMS in these two locales includes many non-ASCII
65 * characters as controls and punctuation (below are hex bytes):
66 * cntrl: 84-97 9B-9F
67 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
68 * Oddly, none there are listed as alphas, though some represent alphabetics
69 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
70#define isNAME_C_OR_POSIX(name) \
71 ( (name) != NULL \
72 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
73 || strEQ((name), "POSIX")))
74
8ef6e574
KW
75#ifdef USE_LOCALE
76
98994639 77/*
0d071d52
KW
78 * Standardize the locale name from a string returned by 'setlocale', possibly
79 * modifying that string.
98994639 80 *
0ef2a2b2 81 * The typical return value of setlocale() is either
98994639
HS
82 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
83 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
84 * (the space-separated values represent the various sublocales,
0ef2a2b2 85 * in some unspecified order). This is not handled by this function.
98994639
HS
86 *
87 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
0ef2a2b2
KW
88 * which is harmful for further use of the string in setlocale(). This
89 * function removes the trailing new line and everything up through the '='
98994639
HS
90 *
91 */
92STATIC char *
93S_stdize_locale(pTHX_ char *locs)
94{
7452cf6a 95 const char * const s = strchr(locs, '=');
98994639
HS
96 bool okay = TRUE;
97
7918f24d
NC
98 PERL_ARGS_ASSERT_STDIZE_LOCALE;
99
8772537c
AL
100 if (s) {
101 const char * const t = strchr(s, '.');
98994639 102 okay = FALSE;
8772537c
AL
103 if (t) {
104 const char * const u = strchr(t, '\n');
105 if (u && (u[1] == 0)) {
106 const STRLEN len = u - s;
107 Move(s + 1, locs, len, char);
108 locs[len] = 0;
109 okay = TRUE;
98994639
HS
110 }
111 }
112 }
113
114 if (!okay)
115 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
116
117 return locs;
118}
119
e5f10d49
KW
120/* Two parallel arrays; first the locale categories Perl uses on this system;
121 * the second array is their names. These arrays are in mostly arbitrary
122 * order. */
123
124const int categories[] = {
125
126# ifdef USE_LOCALE_NUMERIC
127 LC_NUMERIC,
128# endif
129# ifdef USE_LOCALE_CTYPE
130 LC_CTYPE,
131# endif
132# ifdef USE_LOCALE_COLLATE
133 LC_COLLATE,
134# endif
135# ifdef USE_LOCALE_TIME
136 LC_TIME,
137# endif
138# ifdef USE_LOCALE_MESSAGES
139 LC_MESSAGES,
140# endif
141# ifdef USE_LOCALE_MONETARY
142 LC_MONETARY,
143# endif
144# ifdef LC_ALL
145 LC_ALL,
146# endif
147 -1 /* Placeholder because C doesn't allow a
148 trailing comma, and it would get complicated
149 with all the #ifdef's */
150};
151
152/* The top-most real element is LC_ALL */
153
154const char * category_names[] = {
155
156# ifdef USE_LOCALE_NUMERIC
157 "LC_NUMERIC",
158# endif
159# ifdef USE_LOCALE_CTYPE
160 "LC_CTYPE",
161# endif
162# ifdef USE_LOCALE_COLLATE
163 "LC_COLLATE",
164# endif
165# ifdef USE_LOCALE_TIME
166 "LC_TIME",
167# endif
168# ifdef USE_LOCALE_MESSAGES
169 "LC_MESSAGES",
170# endif
171# ifdef USE_LOCALE_MONETARY
172 "LC_MONETARY",
173# endif
174# ifdef LC_ALL
175 "LC_ALL",
176# endif
177 NULL /* Placeholder */
178 };
179
180# ifdef LC_ALL
181
182 /* On systems with LC_ALL, it is kept in the highest index position. (-2
183 * to account for the final unused placeholder element.) */
184# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
185
186# else
187
188 /* On systems without LC_ALL, we pretend it is there, one beyond the real
189 * top element, hence in the unused placeholder element. */
190# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
191
192# endif
193
194/* Pretending there is an LC_ALL element just above allows us to avoid most
195 * special cases. Most loops through these arrays in the code below are
196 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
197 * on either type of system. But the code must be written to not access the
948523db
KW
198 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
199 * checked for at compile time by using the #define LC_ALL_INDEX which is only
200 * defined if we do have LC_ALL. */
e5f10d49 201
b09aaf40
KW
202STATIC const char *
203S_category_name(const int category)
204{
205 unsigned int i;
206
207#ifdef LC_ALL
208
209 if (category == LC_ALL) {
210 return "LC_ALL";
211 }
212
213#endif
214
215 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
216 if (category == categories[i]) {
217 return category_names[i];
218 }
219 }
220
221 {
222 const char suffix[] = " (unknown)";
223 int temp = category;
224 Size_t length = sizeof(suffix) + 1;
225 char * unknown;
226 dTHX;
227
228 if (temp < 0) {
229 length++;
230 temp = - temp;
231 }
232
233 /* Calculate the number of digits */
234 while (temp >= 10) {
235 temp /= 10;
236 length++;
237 }
238
239 Newx(unknown, length, char);
240 my_snprintf(unknown, length, "%d%s", category, suffix);
241 SAVEFREEPV(unknown);
242 return unknown;
243 }
244}
245
948523db
KW
246/* Now create LC_foo_INDEX #defines for just those categories on this system */
247# ifdef USE_LOCALE_NUMERIC
248# define LC_NUMERIC_INDEX 0
249# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
250# else
251# define _DUMMY_NUMERIC -1
252# endif
253# ifdef USE_LOCALE_CTYPE
254# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
255# define _DUMMY_CTYPE LC_CTYPE_INDEX
256# else
257# define _DUMMY_CTYPE _DUMMY_NUMERIC
258# endif
259# ifdef USE_LOCALE_COLLATE
260# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
261# define _DUMMY_COLLATE LC_COLLATE_INDEX
262# else
263# define _DUMMY_COLLATE _DUMMY_COLLATE
264# endif
265# ifdef USE_LOCALE_TIME
266# define LC_TIME_INDEX _DUMMY_COLLATE + 1
267# define _DUMMY_TIME LC_TIME_INDEX
268# else
269# define _DUMMY_TIME _DUMMY_COLLATE
270# endif
271# ifdef USE_LOCALE_MESSAGES
272# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
273# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
274# else
275# define _DUMMY_MESSAGES _DUMMY_TIME
276# endif
277# ifdef USE_LOCALE_MONETARY
278# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
279# define _DUMMY_MONETARY LC_MONETARY_INDEX
280# else
281# define _DUMMY_MONETARY _DUMMY_MESSAGES
282# endif
283# ifdef LC_ALL
284# define LC_ALL_INDEX _DUMMY_MONETARY + 1
285# endif
286#endif /* ifdef USE_LOCALE */
8ef6e574 287
d2b24094
KW
288/* Windows requres a customized base-level setlocale() */
289# ifdef WIN32
290# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
291# else
292# define my_setlocale(cat, locale) setlocale(cat, locale)
293# endif
294
837ce802
KW
295/* Just placeholders for now. "_c" is intended to be called when the category
296 * is a constant known at compile time; "_r", not known until run time */
d2b24094
KW
297# define do_setlocale_c(category, locale) my_setlocale(category, locale)
298# define do_setlocale_r(category, locale) my_setlocale(category, locale)
837ce802 299
a4f00dcc 300STATIC void
86799d2d 301S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 302{
86799d2d
KW
303 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
304 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 305
86799d2d
KW
306#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
307 || defined(HAS_NL_LANGINFO))
98994639 308
86799d2d
KW
309 /* We only set up the radix SV if we are to use a locale radix ... */
310 if (use_locale) {
311 const char * radix = my_nl_langinfo(PERL_RADIXCHAR, FALSE);
312 /* FALSE => already in dest locale */
313
314 /* ... and the character being used isn't a dot */
315 if (strNE(radix, ".")) {
316 if (PL_numeric_radix_sv) {
317 sv_setpv(PL_numeric_radix_sv, radix);
318 }
319 else {
320 PL_numeric_radix_sv = newSVpv(radix, 0);
321 }
322
323 if ( ! is_utf8_invariant_string(
324 (U8 *) SvPVX(PL_numeric_radix_sv), SvCUR(PL_numeric_radix_sv))
325 && is_utf8_string(
326 (U8 *) SvPVX(PL_numeric_radix_sv), SvCUR(PL_numeric_radix_sv))
c1284011 327 && _is_cur_LC_category_utf8(LC_NUMERIC))
28acfe03 328 {
86799d2d 329 SvUTF8_on(PL_numeric_radix_sv);
28acfe03 330 }
86799d2d
KW
331 goto done;
332 }
98994639 333 }
69014004 334
86799d2d
KW
335 SvREFCNT_dec(PL_numeric_radix_sv);
336 PL_numeric_radix_sv = NULL;
337
338 done: ;
339
340# ifdef DEBUGGING
7d4bcc4a 341
2fcc0ca9
KW
342 if (DEBUG_L_TEST || debug_initialization) {
343 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
69014004 344 (PL_numeric_radix_sv)
37b7e435
KW
345 ? SvPVX(PL_numeric_radix_sv)
346 : "NULL",
347 (PL_numeric_radix_sv)
39eb7305 348 ? cBOOL(SvUTF8(PL_numeric_radix_sv))
2fcc0ca9
KW
349 : 0);
350 }
69014004 351
86799d2d
KW
352# endif
353#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 354
98994639
HS
355}
356
a39edc4c 357
98994639 358void
8772537c 359Perl_new_numeric(pTHX_ const char *newnum)
98994639 360{
7d4bcc4a
KW
361
362#ifndef USE_LOCALE_NUMERIC
363
364 PERL_UNUSED_ARG(newnum);
365
366#else
0d071d52 367
291a84fb 368 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
369 * core Perl this and that 'newnum' is the name of the new locale.
370 * It installs this locale as the current underlying default.
371 *
372 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
373 * set_numeric_underlying() and set_numeric_standard() functions, which
374 * should probably not be called directly, but only via macros like
0d071d52
KW
375 * SET_NUMERIC_STANDARD() in perl.h.
376 *
377 * The toggling is necessary mainly so that a non-dot radix decimal point
378 * character can be output, while allowing internal calculations to use a
379 * dot.
380 *
381 * This sets several interpreter-level variables:
bb304765 382 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 383 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
384 * that the current locale is the program's underlying
385 * locale
386 * PL_numeric_standard An int indicating if the toggled state is such
387 * that the current locale is the C locale. If non-zero,
388 * it is in C; if > 1, it means it may not be toggled away
389 * from C.
0d071d52
KW
390 * Note that both of the last two variables can be true at the same time,
391 * if the underlying locale is C. (Toggling is a no-op under these
392 * circumstances.)
393 *
394 * Any code changing the locale (outside this file) should use
395 * POSIX::setlocale, which calls this function. Therefore this function
396 * should be called directly only from this file and from
397 * POSIX::setlocale() */
398
b03f34cf 399 char *save_newnum;
98994639
HS
400
401 if (! newnum) {
43c5f42d
NC
402 Safefree(PL_numeric_name);
403 PL_numeric_name = NULL;
98994639 404 PL_numeric_standard = TRUE;
892e6465 405 PL_numeric_underlying = TRUE;
98994639
HS
406 return;
407 }
408
b03f34cf 409 save_newnum = stdize_locale(savepv(newnum));
abe1abcf
KW
410
411 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
892e6465 412 PL_numeric_underlying = TRUE;
abe1abcf 413
b03f34cf 414 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 415 Safefree(PL_numeric_name);
b03f34cf 416 PL_numeric_name = save_newnum;
b03f34cf 417 }
abe1abcf
KW
418 else {
419 Safefree(save_newnum);
420 }
4c28b29c
KW
421
422 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
423 * have to worry about the radix being a non-dot. (Core operations that
424 * need the underlying locale change to it temporarily). */
425 set_numeric_standard();
426
98994639 427#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 428
98994639
HS
429}
430
431void
432Perl_set_numeric_standard(pTHX)
433{
7d4bcc4a 434
98994639 435#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 436
28c1bf33
KW
437 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
438 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
439 * macro avoids calling this routine if toggling isn't necessary according
440 * to our records (which could be wrong if some XS code has changed the
441 * locale behind our back) */
0d071d52 442
837ce802 443 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 444 PL_numeric_standard = TRUE;
892e6465 445 PL_numeric_underlying = isNAME_C_OR_POSIX(PL_numeric_name);
86799d2d 446 set_numeric_radix(0);
7d4bcc4a
KW
447
448# ifdef DEBUGGING
449
2fcc0ca9
KW
450 if (DEBUG_L_TEST || debug_initialization) {
451 PerlIO_printf(Perl_debug_log,
58e4a467 452 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 453 }
98994639 454
7d4bcc4a 455# endif
98994639 456#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 457
98994639
HS
458}
459
460void
5792c642 461Perl_set_numeric_underlying(pTHX)
98994639 462{
7d4bcc4a 463
98994639 464#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 465
28c1bf33 466 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
467 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
468 * instead of calling this directly. The macro avoids calling this routine
469 * if toggling isn't necessary according to our records (which could be
470 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 471
837ce802 472 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
a9b8c0d8 473 PL_numeric_standard = isNAME_C_OR_POSIX(PL_numeric_name);
892e6465 474 PL_numeric_underlying = TRUE;
86799d2d 475 set_numeric_radix(1);
7d4bcc4a
KW
476
477# ifdef DEBUGGING
478
2fcc0ca9
KW
479 if (DEBUG_L_TEST || debug_initialization) {
480 PerlIO_printf(Perl_debug_log,
58e4a467 481 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
482 PL_numeric_name);
483 }
98994639 484
7d4bcc4a 485# endif
98994639 486#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 487
98994639
HS
488}
489
490/*
491 * Set up for a new ctype locale.
492 */
a4f00dcc
KW
493STATIC void
494S_new_ctype(pTHX_ const char *newctype)
98994639 495{
7d4bcc4a
KW
496
497#ifndef USE_LOCALE_CTYPE
498
499 PERL_ARGS_ASSERT_NEW_CTYPE;
500 PERL_UNUSED_ARG(newctype);
501 PERL_UNUSED_CONTEXT;
502
503#else
0d071d52 504
291a84fb 505 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
506 * core Perl this and that 'newctype' is the name of the new locale.
507 *
508 * This function sets up the folding arrays for all 256 bytes, assuming
509 * that tofold() is tolc() since fold case is not a concept in POSIX,
510 *
511 * Any code changing the locale (outside this file) should use
512 * POSIX::setlocale, which calls this function. Therefore this function
513 * should be called directly only from this file and from
514 * POSIX::setlocale() */
515
27da23d5 516 dVAR;
68067e4e 517 UV i;
98994639 518
7918f24d
NC
519 PERL_ARGS_ASSERT_NEW_CTYPE;
520
215c5139
KW
521 /* We will replace any bad locale warning with 1) nothing if the new one is
522 * ok; or 2) a new warning for the bad new locale */
523 if (PL_warn_locale) {
524 SvREFCNT_dec_NN(PL_warn_locale);
525 PL_warn_locale = NULL;
526 }
527
c1284011 528 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
529
530 /* A UTF-8 locale gets standard rules. But note that code still has to
531 * handle this specially because of the three problematic code points */
532 if (PL_in_utf8_CTYPE_locale) {
533 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
534 }
535 else {
8c6180a9
KW
536 /* Assume enough space for every character being bad. 4 spaces each
537 * for the 94 printable characters that are output like "'x' "; and 5
538 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
539 * NUL */
540 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
541
cc9eaeb0
KW
542 /* Don't check for problems if we are suppressing the warnings */
543 bool check_for_problems = ckWARN_d(WARN_LOCALE)
544 || UNLIKELY(DEBUG_L_TEST);
8c6180a9
KW
545 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
546 to start */
547 unsigned int bad_count = 0; /* Count of bad characters */
548
baa60164 549 for (i = 0; i < 256; i++) {
5a8a8f21
KW
550 if (isupper(i))
551 PL_fold_locale[i] = (U8) tolower(i);
552 else if (islower(i))
553 PL_fold_locale[i] = (U8) toupper(i);
baa60164
KW
554 else
555 PL_fold_locale[i] = (U8) i;
8c6180a9
KW
556
557 /* If checking for locale problems, see if the native ASCII-range
558 * printables plus \n and \t are in their expected categories in
559 * the new locale. If not, this could mean big trouble, upending
560 * Perl's and most programs' assumptions, like having a
561 * metacharacter with special meaning become a \w. Fortunately,
562 * it's very rare to find locales that aren't supersets of ASCII
563 * nowadays. It isn't a problem for most controls to be changed
564 * into something else; we check only \n and \t, though perhaps \r
565 * could be an issue as well. */
7d4bcc4a 566 if ( check_for_problems
8c6180a9
KW
567 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
568 {
adf947ba
KW
569 if ( cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC(i))
570 || cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i))
571 || cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i))
572 || cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i))
573 || cBOOL(islower(i)) != cBOOL(isLOWER_A(i))
574 || cBOOL(isprint(i)) != cBOOL(isPRINT_A(i))
575 || cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i))
576 || cBOOL(isspace(i)) != cBOOL(isSPACE_A(i))
577 || cBOOL(isupper(i)) != cBOOL(isUPPER_A(i))
578 || cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i))
579 || tolower(i) != (int) toLOWER_A(i)
580 || toupper(i) != (int) toUPPER_A(i)
581 || (i == '\n' && ! isCNTRL_LC(i)))
8c6180a9
KW
582 {
583 if (bad_count) { /* Separate multiple entries with a
584 blank */
585 bad_chars_list[bad_count++] = ' ';
586 }
587 bad_chars_list[bad_count++] = '\'';
588 if (isPRINT_A(i)) {
589 bad_chars_list[bad_count++] = (char) i;
590 }
591 else {
592 bad_chars_list[bad_count++] = '\\';
593 if (i == '\n') {
594 bad_chars_list[bad_count++] = 'n';
595 }
596 else {
597 assert(i == '\t');
598 bad_chars_list[bad_count++] = 't';
599 }
600 }
601 bad_chars_list[bad_count++] = '\'';
602 bad_chars_list[bad_count] = '\0';
603 }
604 }
605 }
606
7d4bcc4a
KW
607# ifdef MB_CUR_MAX
608
8c6180a9 609 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 610 * this locale requires more than one byte, there are going to be
8c6180a9 611 * problems. */
9c8a6dc2
KW
612 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
613 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
614 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
615
ba1a4362
KW
616 if (check_for_problems && MB_CUR_MAX > 1
617
618 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
619 * locale. Just assume that the implementation for them (plus
620 * for POSIX) is correct and the > 1 value is spurious. (Since
621 * these are specially handled to never be considered UTF-8
622 * locales, as long as this is the only problem, everything
623 * should work fine */
624 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
625 {
8c6180a9
KW
626 multi_byte_locale = TRUE;
627 }
7d4bcc4a
KW
628
629# endif
8c6180a9
KW
630
631 if (bad_count || multi_byte_locale) {
780fcc9f 632 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 633 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 634 newctype,
8c6180a9
KW
635 (multi_byte_locale)
636 ? " Some characters in it are not recognized by"
637 " Perl."
638 : "",
639 (bad_count)
640 ? "\nThe following characters (and maybe others)"
641 " may not have the same meaning as the Perl"
642 " program expects:\n"
643 : "",
644 (bad_count)
645 ? bad_chars_list
646 : ""
647 );
cc9eaeb0 648 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
649 * output the message now. If not in that scope, we save the
650 * message to be output at the first operation using this locale,
651 * if that actually happens. Most programs don't use locales, so
652 * they are immune to bad ones. */
cc9eaeb0 653 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f
KW
654
655 /* We have to save 'newctype' because the setlocale() just
656 * below may destroy it. The next setlocale() further down
657 * should restore it properly so that the intermediate change
658 * here is transparent to this function's caller */
659 const char * const badlocale = savepv(newctype);
660
837ce802 661 do_setlocale_c(LC_CTYPE, "C");
780fcc9f
KW
662
663 /* The '0' below suppresses a bogus gcc compiler warning */
664 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 665
837ce802 666 do_setlocale_c(LC_CTYPE, badlocale);
c0f3a893 667 Safefree(badlocale);
bddebb56
KW
668
669 if (IN_LC(LC_CTYPE)) {
670 SvREFCNT_dec_NN(PL_warn_locale);
671 PL_warn_locale = NULL;
672 }
780fcc9f 673 }
baa60164 674 }
31f05a37 675 }
98994639
HS
676
677#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 678
98994639
HS
679}
680
98994639 681void
2726666d
KW
682Perl__warn_problematic_locale()
683{
2726666d
KW
684
685#ifdef USE_LOCALE_CTYPE
686
5f04a188
KW
687 dTHX;
688
689 /* Internal-to-core function that outputs the message in PL_warn_locale,
690 * and then NULLS it. Should be called only through the macro
691 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
692
2726666d 693 if (PL_warn_locale) {
2726666d
KW
694 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
695 SvPVX(PL_warn_locale),
696 0 /* dummy to avoid compiler warning */ );
2726666d
KW
697 SvREFCNT_dec_NN(PL_warn_locale);
698 PL_warn_locale = NULL;
699 }
700
701#endif
702
703}
704
a4f00dcc
KW
705STATIC void
706S_new_collate(pTHX_ const char *newcoll)
98994639 707{
7d4bcc4a
KW
708
709#ifndef USE_LOCALE_COLLATE
710
711 PERL_UNUSED_ARG(newcoll);
712 PERL_UNUSED_CONTEXT;
713
714#else
0d071d52 715
291a84fb 716 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
717 * core Perl this and that 'newcoll' is the name of the new locale.
718 *
d35fca5f
KW
719 * The design of locale collation is that every locale change is given an
720 * index 'PL_collation_ix'. The first time a string particpates in an
721 * operation that requires collation while locale collation is active, it
722 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
723 * magic includes the collation index, and the transformation of the string
724 * by strxfrm(), q.v. That transformation is used when doing comparisons,
725 * instead of the string itself. If a string changes, the magic is
726 * cleared. The next time the locale changes, the index is incremented,
727 * and so we know during a comparison that the transformation is not
728 * necessarily still valid, and so is recomputed. Note that if the locale
729 * changes enough times, the index could wrap (a U32), and it is possible
730 * that a transformation would improperly be considered valid, leading to
731 * an unlikely bug */
0d071d52 732
98994639
HS
733 if (! newcoll) {
734 if (PL_collation_name) {
735 ++PL_collation_ix;
736 Safefree(PL_collation_name);
737 PL_collation_name = NULL;
738 }
739 PL_collation_standard = TRUE;
00bf60ca 740 is_standard_collation:
98994639
HS
741 PL_collxfrm_base = 0;
742 PL_collxfrm_mult = 2;
165a1c52 743 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 744 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 745 PL_strxfrm_max_cp = 0;
98994639
HS
746 return;
747 }
748
d35fca5f 749 /* If this is not the same locale as currently, set the new one up */
98994639
HS
750 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
751 ++PL_collation_ix;
752 Safefree(PL_collation_name);
753 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 754 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
755 if (PL_collation_standard) {
756 goto is_standard_collation;
757 }
98994639 758
165a1c52 759 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 760 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 761 PL_strxfrm_max_cp = 0;
165a1c52 762
59c018b9
KW
763 /* A locale collation definition includes primary, secondary, tertiary,
764 * etc. weights for each character. To sort, the primary weights are
765 * used, and only if they compare equal, then the secondary weights are
766 * used, and only if they compare equal, then the tertiary, etc.
767 *
768 * strxfrm() works by taking the input string, say ABC, and creating an
769 * output transformed string consisting of first the primary weights,
770 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
771 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
772 * may not have weights at every level. In our example, let's say B
773 * doesn't have a tertiary weight, and A doesn't have a secondary
774 * weight. The constructed string is then going to be
775 * A¹B¹C¹ B²C² A³C³ ....
776 * This has the desired effect that strcmp() will look at the secondary
777 * or tertiary weights only if the strings compare equal at all higher
778 * priority weights. The spaces shown here, like in
c342d20e 779 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
780 * are not just for readability. In the general case, these must
781 * actually be bytes, which we will call here 'separator weights'; and
782 * they must be smaller than any other weight value, but since these
783 * are C strings, only the terminating one can be a NUL (some
784 * implementations may include a non-NUL separator weight just before
785 * the NUL). Implementations tend to reserve 01 for the separator
786 * weights. They are needed so that a shorter string's secondary
787 * weights won't be misconstrued as primary weights of a longer string,
788 * etc. By making them smaller than any other weight, the shorter
789 * string will sort first. (Actually, if all secondary weights are
790 * smaller than all primary ones, there is no need for a separator
791 * weight between those two levels, etc.)
792 *
793 * The length of the transformed string is roughly a linear function of
794 * the input string. It's not exactly linear because some characters
795 * don't have weights at all levels. When we call strxfrm() we have to
796 * allocate some memory to hold the transformed string. The
797 * calculations below try to find coefficients 'm' and 'b' for this
798 * locale so that m*x + b equals how much space we need, given the size
799 * of the input string in 'x'. If we calculate too small, we increase
800 * the size as needed, and call strxfrm() again, but it is better to
801 * get it right the first time to avoid wasted expensive string
802 * transformations. */
803
98994639 804 {
79f120c8
KW
805 /* We use the string below to find how long the tranformation of it
806 * is. Almost all locales are supersets of ASCII, or at least the
807 * ASCII letters. We use all of them, half upper half lower,
808 * because if we used fewer, we might hit just the ones that are
809 * outliers in a particular locale. Most of the strings being
810 * collated will contain a preponderance of letters, and even if
811 * they are above-ASCII, they are likely to have the same number of
812 * weight levels as the ASCII ones. It turns out that digits tend
813 * to have fewer levels, and some punctuation has more, but those
814 * are relatively sparse in text, and khw believes this gives a
815 * reasonable result, but it could be changed if experience so
816 * dictates. */
817 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
818 char * x_longer; /* Transformed 'longer' */
819 Size_t x_len_longer; /* Length of 'x_longer' */
820
821 char * x_shorter; /* We also transform a substring of 'longer' */
822 Size_t x_len_shorter;
823
a4a439fb 824 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
825 * are interested only in its length). It is used because it has
826 * the intelligence to handle all cases, but to work, it needs some
827 * values of 'm' and 'b' to get it started. For the purposes of
828 * this calculation we use a very conservative estimate of 'm' and
829 * 'b'. This assumes a weight can be multiple bytes, enough to
830 * hold any UV on the platform, and there are 5 levels, 4 weight
831 * bytes, and a trailing NUL. */
832 PL_collxfrm_base = 5;
833 PL_collxfrm_mult = 5 * sizeof(UV);
834
835 /* Find out how long the transformation really is */
a4a439fb
KW
836 x_longer = _mem_collxfrm(longer,
837 sizeof(longer) - 1,
838 &x_len_longer,
839
840 /* We avoid converting to UTF-8 in the
841 * called function by telling it the
842 * string is in UTF-8 if the locale is a
843 * UTF-8 one. Since the string passed
844 * here is invariant under UTF-8, we can
845 * claim it's UTF-8 even though it isn't.
846 * */
847 PL_in_utf8_COLLATE_locale);
79f120c8
KW
848 Safefree(x_longer);
849
850 /* Find out how long the transformation of a substring of 'longer'
851 * is. Together the lengths of these transformations are
852 * sufficient to calculate 'm' and 'b'. The substring is all of
853 * 'longer' except the first character. This minimizes the chances
854 * of being swayed by outliers */
a4a439fb 855 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 856 sizeof(longer) - 2,
a4a439fb
KW
857 &x_len_shorter,
858 PL_in_utf8_COLLATE_locale);
79f120c8
KW
859 Safefree(x_shorter);
860
861 /* If the results are nonsensical for this simple test, the whole
862 * locale definition is suspect. Mark it so that locale collation
863 * is not active at all for it. XXX Should we warn? */
864 if ( x_len_shorter == 0
865 || x_len_longer == 0
866 || x_len_shorter >= x_len_longer)
867 {
868 PL_collxfrm_mult = 0;
869 PL_collxfrm_base = 0;
870 }
871 else {
872 SSize_t base; /* Temporary */
873
874 /* We have both: m * strlen(longer) + b = x_len_longer
875 * m * strlen(shorter) + b = x_len_shorter;
876 * subtracting yields:
877 * m * (strlen(longer) - strlen(shorter))
878 * = x_len_longer - x_len_shorter
879 * But we have set things up so that 'shorter' is 1 byte smaller
880 * than 'longer'. Hence:
881 * m = x_len_longer - x_len_shorter
882 *
883 * But if something went wrong, make sure the multiplier is at
884 * least 1.
885 */
886 if (x_len_longer > x_len_shorter) {
887 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
888 }
889 else {
890 PL_collxfrm_mult = 1;
891 }
892
893 /* mx + b = len
894 * so: b = len - mx
895 * but in case something has gone wrong, make sure it is
896 * non-negative */
897 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
898 if (base < 0) {
899 base = 0;
900 }
901
902 /* Add 1 for the trailing NUL */
903 PL_collxfrm_base = base + 1;
904 }
58eebef2 905
7d4bcc4a
KW
906# ifdef DEBUGGING
907
58eebef2
KW
908 if (DEBUG_L_TEST || debug_initialization) {
909 PerlIO_printf(Perl_debug_log,
b07929e4
KW
910 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
911 "x_len_longer=%zu,"
912 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
913 __FILE__, __LINE__,
914 PL_in_utf8_COLLATE_locale,
915 x_len_shorter, x_len_longer,
916 PL_collxfrm_mult, PL_collxfrm_base);
917 }
7d4bcc4a
KW
918# endif
919
98994639
HS
920 }
921 }
922
923#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 924
98994639
HS
925}
926
d2b24094 927#ifdef WIN32
b8cc575c 928
a4f00dcc 929STATIC char *
b8cc575c 930S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
931{
932 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
933 * difference between the two unless the input locale is "", which normally
934 * means on Windows to get the machine default, which is set via the
935 * computer's "Regional and Language Options" (or its current equivalent).
936 * In POSIX, it instead means to find the locale from the user's
937 * environment. This routine changes the Windows behavior to first look in
938 * the environment, and, if anything is found, use that instead of going to
939 * the machine default. If there is no environment override, the machine
940 * default is used, by calling the real setlocale() with "".
941 *
942 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
943 * use the particular category's variable if set; otherwise to use the LANG
944 * variable. */
b385bb4d 945
175c4cf9 946 bool override_LC_ALL = FALSE;
89f7b9aa 947 char * result;
e5f10d49 948 unsigned int i;
89f7b9aa 949
b385bb4d 950 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
951
952# ifdef LC_ALL
953
b385bb4d
KW
954 locale = PerlEnv_getenv("LC_ALL");
955 if (! locale) {
e5f10d49
KW
956 if (category == LC_ALL) {
957 override_LC_ALL = TRUE;
958 }
959 else {
7d4bcc4a
KW
960
961# endif
7d4bcc4a 962
e5f10d49
KW
963 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
964 if (category == categories[i]) {
965 locale = PerlEnv_getenv(category_names[i]);
966 goto found_locale;
967 }
968 }
7d4bcc4a 969
b385bb4d 970 locale = PerlEnv_getenv("LANG");
481465ea 971 if (! locale) {
b385bb4d
KW
972 locale = "";
973 }
e5f10d49
KW
974
975 found_locale: ;
7d4bcc4a
KW
976
977# ifdef LC_ALL
978
e5f10d49 979 }
b385bb4d 980 }
7d4bcc4a
KW
981
982# endif
983
b385bb4d
KW
984 }
985
89f7b9aa 986 result = setlocale(category, locale);
bbc98134 987 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
a4f00dcc 988 setlocale_debug_string(category, locale, result)));
89f7b9aa 989
481465ea 990 if (! override_LC_ALL) {
89f7b9aa
KW
991 return result;
992 }
993
dfd77d7a 994 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
995 * LANG variable or the system default if there is no LANG. But these have
996 * lower priority than the other LC_foo variables, so override it for each
997 * one that is set. (If they are set to "", it means to use the same thing
998 * we just set LC_ALL to, so can skip) */
7d4bcc4a 999
948523db 1000 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
1001 result = PerlEnv_getenv(category_names[i]);
1002 if (result && strNE(result, "")) {
1003 setlocale(categories[i], result);
1004 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1005 __FILE__, __LINE__,
1006 setlocale_debug_string(categories[i], result, "not captured")));
1007 }
89f7b9aa 1008 }
7d4bcc4a 1009
bbc98134
KW
1010 result = setlocale(LC_ALL, NULL);
1011 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1012 __FILE__, __LINE__,
a4f00dcc 1013 setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 1014
bbc98134 1015 return result;
b385bb4d
KW
1016}
1017
1018#endif
1019
a4f00dcc
KW
1020char *
1021Perl_setlocale(int category, const char * locale)
1022{
1023 /* This wraps POSIX::setlocale() */
1024
1025 char * retval;
1159483a 1026 char * newlocale;
a4f00dcc
KW
1027 dTHX;
1028
a4f00dcc
KW
1029#ifdef USE_LOCALE_NUMERIC
1030
291a84fb
KW
1031 /* A NULL locale means only query what the current one is. We have the
1032 * LC_NUMERIC name saved, because we are normally switched into the C
1033 * locale for it. For an LC_ALL query, switch back to get the correct
1034 * results. All other categories don't require special handling */
a4f00dcc
KW
1035 if (locale == NULL) {
1036 if (category == LC_NUMERIC) {
1037 return savepv(PL_numeric_name);
1038 }
1039
7d4bcc4a 1040# ifdef LC_ALL
a4f00dcc
KW
1041
1042 else if (category == LC_ALL) {
1043 SET_NUMERIC_UNDERLYING();
1044 }
1045
7d4bcc4a 1046# endif
a4f00dcc
KW
1047
1048 }
1049
1050#endif
1051
e5f10d49
KW
1052 /* Save retval since subsequent setlocale() calls may overwrite it. */
1053 retval = savepv(do_setlocale_r(category, locale));
a4f00dcc
KW
1054
1055 DEBUG_L(PerlIO_printf(Perl_debug_log,
1056 "%s:%d: %s\n", __FILE__, __LINE__,
1057 setlocale_debug_string(category, locale, retval)));
1058 if (! retval) {
1059 /* Should never happen that a query would return an error, but be
1060 * sure and reset to C locale */
1061 if (locale == 0) {
1062 SET_NUMERIC_STANDARD();
1063 }
7d4bcc4a 1064
a4f00dcc
KW
1065 return NULL;
1066 }
1067
a4f00dcc
KW
1068 /* If locale == NULL, we are just querying the state, but may have switched
1069 * to NUMERIC_UNDERLYING. Switch back before returning. */
1070 if (locale == NULL) {
1071 SET_NUMERIC_STANDARD();
1072 return retval;
1073 }
a4f00dcc 1074
1159483a
KW
1075 /* Now that have switched locales, we have to update our records to
1076 * correspond. */
a4f00dcc 1077
1159483a 1078 switch (category) {
a4f00dcc 1079
1159483a 1080#ifdef USE_LOCALE_CTYPE
a4f00dcc 1081
1159483a
KW
1082 case LC_CTYPE:
1083 new_ctype(retval);
1084 break;
a4f00dcc 1085
1159483a 1086#endif
a4f00dcc
KW
1087#ifdef USE_LOCALE_COLLATE
1088
1159483a
KW
1089 case LC_COLLATE:
1090 new_collate(retval);
1091 break;
a4f00dcc 1092
1159483a 1093#endif
a4f00dcc
KW
1094#ifdef USE_LOCALE_NUMERIC
1095
1159483a
KW
1096 case LC_NUMERIC:
1097 new_numeric(retval);
1098 break;
a4f00dcc 1099
1159483a
KW
1100#endif
1101#ifdef LC_ALL
a4f00dcc 1102
1159483a 1103 case LC_ALL:
a4f00dcc 1104
1159483a
KW
1105 /* LC_ALL updates all the things we care about. The values may not
1106 * be the same as 'retval', as the locale "" may have set things
1107 * individually */
a4f00dcc 1108
1159483a 1109# ifdef USE_LOCALE_CTYPE
a4f00dcc 1110
1159483a
KW
1111 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1112 new_ctype(newlocale);
a4f00dcc 1113
1159483a
KW
1114# endif /* USE_LOCALE_CTYPE */
1115# ifdef USE_LOCALE_COLLATE
1116
1117 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1118 new_collate(newlocale);
a4f00dcc 1119
7d4bcc4a 1120# endif
1159483a 1121# ifdef USE_LOCALE_NUMERIC
a4f00dcc 1122
1159483a
KW
1123 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1124 new_numeric(newlocale);
a4f00dcc 1125
1159483a
KW
1126# endif /* USE_LOCALE_NUMERIC */
1127#endif /* LC_ALL */
a4f00dcc 1128
1159483a
KW
1129 default:
1130 break;
a4f00dcc
KW
1131 }
1132
1133 return retval;
1134
f7416781
KW
1135
1136}
1137
1138PERL_STATIC_INLINE const char *
1139S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1140{
1141 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1142 * growing it if necessary */
1143
1144 const Size_t string_size = strlen(string) + offset + 1;
1145
1146 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1147
1148 if (*buf_size == 0) {
1149 Newx(*buf, string_size, char);
1150 *buf_size = string_size;
1151 }
1152 else if (string_size > *buf_size) {
1153 Renew(*buf, string_size, char);
1154 *buf_size = string_size;
1155 }
1156
1157 Copy(string, *buf + offset, string_size - offset, char);
1158 return *buf;
1159}
1160
1161/*
1162
1163=head1 Locale-related functions and macros
1164
1165=for apidoc Perl_langinfo
1166
7d4bcc4a 1167This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
1168taking the same C<item> parameter values, and returning the same information.
1169But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1170of Perl's locale handling from your code, and can be used on systems that lack
1171a native C<nl_langinfo>.
1172
1173Expanding on these:
1174
1175=over
1176
1177=item *
1178
1179It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1180without you having to write extra code. The reason for the extra code would be
1181because these are from the C<LC_NUMERIC> locale category, which is normally
1182kept set to the C locale by Perl, no matter what the underlying locale is
1183supposed to be, and so to get the expected results, you have to temporarily
1184toggle into the underlying locale, and later toggle back. (You could use
1185plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1186but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1187keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1188expecting the radix (decimal point) character to be a dot.)
1189
1190=item *
1191
1192Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1193makes your code more portable. Of the fifty-some possible items specified by
1194the POSIX 2008 standard,
1195L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1196only two are completely unimplemented. It uses various techniques to recover
1197the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1198both of which are specified in C89, so should be always be available. Later
1199C<strftime()> versions have additional capabilities; C<""> is returned for
1200those not available on your system.
1201
1202The details for those items which may differ from what this emulation returns
1203and what a native C<nl_langinfo()> would return are:
1204
1205=over
1206
1207=item C<CODESET>
1208
1209=item C<ERA>
1210
1211Unimplemented, so returns C<"">.
1212
1213=item C<YESEXPR>
1214
c1566110
KW
1215=item C<YESSTR>
1216
f7416781
KW
1217=item C<NOEXPR>
1218
c1566110
KW
1219=item C<NOSTR>
1220
1221Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
1222removed from POSIX 2008, and are retained for backwards compatibility. Your
1223platform's C<nl_langinfo> may not support them.
f7416781
KW
1224
1225=item C<D_FMT>
1226
1227Always evaluates to C<%x>, the locale's appropriate date representation.
1228
1229=item C<T_FMT>
1230
1231Always evaluates to C<%X>, the locale's appropriate time representation.
1232
1233=item C<D_T_FMT>
1234
1235Always evaluates to C<%c>, the locale's appropriate date and time
1236representation.
1237
1238=item C<CRNCYSTR>
1239
1240The return may be incorrect for those rare locales where the currency symbol
1241replaces the radix character.
1242Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1243to work differently.
1244
1245=item C<ALT_DIGITS>
1246
1247Currently this gives the same results as Linux does.
1248Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1249to work differently.
1250
1251=item C<ERA_D_FMT>
1252
1253=item C<ERA_T_FMT>
1254
1255=item C<ERA_D_T_FMT>
1256
1257=item C<T_FMT_AMPM>
1258
1259These are derived by using C<strftime()>, and not all versions of that function
1260know about them. C<""> is returned for these on such systems.
1261
1262=back
1263
1264When using C<Perl_langinfo> on systems that don't have a native
1265C<nl_langinfo()>, you must
1266
1267 #include "perl_langinfo.h"
1268
1269before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1270C<#include> with this one. (Doing it this way keeps out the symbols that plain
1271C<langinfo.h> imports into the namespace for code that doesn't need it.)
1272
1273You also should not use the bare C<langinfo.h> item names, but should preface
1274them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1275The C<PERL_I<foo>> versions will also work for this function on systems that do
1276have a native C<nl_langinfo>.
1277
1278=item *
1279
1280It is thread-friendly, returning its result in a buffer that won't be
1281overwritten by another thread, so you don't have to code for that possibility.
1282The buffer can be overwritten by the next call to C<nl_langinfo> or
1283C<Perl_langinfo> in the same thread.
1284
1285=item *
1286
7d4bcc4a 1287ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
1288*>>, but you are (only by documentation) forbidden to write into the buffer.
1289By declaring this C<const>, the compiler enforces this restriction. The extra
1290C<const> is why this isn't an unequivocal drop-in replacement for
1291C<nl_langinfo>.
1292
1293=back
1294
1295The original impetus for C<Perl_langinfo()> was so that code that needs to
1296find out the current currency symbol, floating point radix character, or digit
1297grouping separator can use, on all systems, the simpler and more
1298thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1299pain to make thread-friendly. For other fields returned by C<localeconv>, it
1300is better to use the methods given in L<perlcall> to call
1301L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1302
1303=cut
1304
1305*/
1306
1307const char *
1308#ifdef HAS_NL_LANGINFO
1309Perl_langinfo(const nl_item item)
1310#else
1311Perl_langinfo(const int item)
1312#endif
1313{
f61748ac
KW
1314 return my_nl_langinfo(item, TRUE);
1315}
1316
1317const char *
1318#ifdef HAS_NL_LANGINFO
1319S_my_nl_langinfo(const nl_item item, bool toggle)
1320#else
1321S_my_nl_langinfo(const int item, bool toggle)
1322#endif
1323{
ae74815b 1324 dTHX;
f7416781 1325
ab340fff
KW
1326#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1327#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1328
ab340fff 1329 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1330 * for those items dependent on it. This must be copied to a buffer before
1331 * switching back, as some systems destroy the buffer when setlocale() is
1332 * called */
f7416781
KW
1333
1334 LOCALE_LOCK;
1335
1336 if (toggle) {
1337 if (item == PERL_RADIXCHAR || item == PERL_THOUSEP) {
837ce802 1338 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
f7416781
KW
1339 }
1340 else {
1341 toggle = FALSE;
1342 }
1343 }
1344
1345 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1346
1347 if (toggle) {
837ce802 1348 do_setlocale_c(LC_NUMERIC, "C");
f7416781
KW
1349 }
1350
1351 LOCALE_UNLOCK;
1352
ab340fff
KW
1353# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1354
1355 bool do_free = FALSE;
1356 locale_t cur = uselocale((locale_t) 0);
1357
1358 if (cur == LC_GLOBAL_LOCALE) {
1359 cur = duplocale(LC_GLOBAL_LOCALE);
1360 do_free = TRUE;
1361 }
1362
1363 if ( toggle
1364 && (item == PERL_RADIXCHAR || item == PERL_THOUSEP))
1365 {
1366 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1367 do_free = TRUE;
1368 }
1369
1370 save_to_buffer(nl_langinfo_l(item, cur),
1371 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1372 if (do_free) {
1373 freelocale(cur);
1374 }
1375
c1566110
KW
1376# endif
1377
1378 if (strEQ(PL_langinfo_buf, "")) {
1379 if (item == PERL_YESSTR) {
1380 return "yes";
1381 }
1382 if (item == PERL_NOSTR) {
1383 return "no";
1384 }
1385 }
1386
ab340fff
KW
1387 return PL_langinfo_buf;
1388
f7416781 1389#else /* Below, emulate nl_langinfo as best we can */
f7416781
KW
1390# ifdef HAS_LOCALECONV
1391
1392 const struct lconv* lc;
1393
1394# endif
1395# ifdef HAS_STRFTIME
1396
1397 struct tm tm;
1398 bool return_format = FALSE; /* Return the %format, not the value */
1399 const char * format;
1400
1401# endif
1402
1403 /* We copy the results to a per-thread buffer, even if not multi-threaded.
1404 * This is in part to simplify this code, and partly because we need a
1405 * buffer anyway for strftime(), and partly because a call of localeconv()
1406 * could otherwise wipe out the buffer, and the programmer would not be
1407 * expecting this, as this is a nl_langinfo() substitute after all, so s/he
1408 * might be thinking their localeconv() is safe until another localeconv()
1409 * call. */
1410
1411 switch (item) {
1412 Size_t len;
1413 const char * retval;
1414
1415 /* These 2 are unimplemented */
1416 case PERL_CODESET:
1417 case PERL_ERA: /* For use with strftime() %E modifier */
1418
1419 default:
1420 return "";
1421
1422 /* We use only an English set, since we don't know any more */
1423 case PERL_YESEXPR: return "^[+1yY]";
c1566110 1424 case PERL_YESSTR: return "yes";
f7416781 1425 case PERL_NOEXPR: return "^[-0nN]";
c1566110 1426 case PERL_NOSTR: return "no";
f7416781
KW
1427
1428# ifdef HAS_LOCALECONV
1429
1430 case PERL_CRNCYSTR:
1431
1432 LOCALE_LOCK;
1433
291a84fb
KW
1434 /* We don't bother with localeconv_l() because any system that has
1435 * it is likely to also have nl_langinfo() */
1436
f7416781
KW
1437 lc = localeconv();
1438 if (! lc || ! lc->currency_symbol || strEQ("", lc->currency_symbol))
1439 {
1440 LOCALE_UNLOCK;
1441 return "";
1442 }
1443
1444 /* Leave the first spot empty to be filled in below */
1445 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1446 &PL_langinfo_bufsize, 1);
1447 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1448 { /* khw couldn't figure out how the localedef specifications
1449 would show that the $ should replace the radix; this is
1450 just a guess as to how it might work.*/
1451 *PL_langinfo_buf = '.';
1452 }
1453 else if (lc->p_cs_precedes) {
1454 *PL_langinfo_buf = '-';
1455 }
1456 else {
1457 *PL_langinfo_buf = '+';
1458 }
1459
1460 LOCALE_UNLOCK;
1461 break;
1462
1463 case PERL_RADIXCHAR:
1464 case PERL_THOUSEP:
1465
1466 LOCALE_LOCK;
1467
1468 if (toggle) {
837ce802 1469 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
f7416781
KW
1470 }
1471
1472 lc = localeconv();
1473 if (! lc) {
1474 retval = "";
1475 }
33394adc
KW
1476 else {
1477 retval = (item == PERL_RADIXCHAR)
1478 ? lc->decimal_point
1479 : lc->thousands_sep;
1480 if (! retval) {
1481 retval = "";
1482 }
f7416781
KW
1483 }
1484
1485 save_to_buffer(retval, &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1486
1487 if (toggle) {
837ce802 1488 do_setlocale_c(LC_NUMERIC, "C");
f7416781
KW
1489 }
1490
1491 LOCALE_UNLOCK;
1492
1493 break;
1494
1495# endif
1496# ifdef HAS_STRFTIME
1497
1498 /* These are defined by C89, so we assume that strftime supports them,
1499 * and so are returned unconditionally; they may not be what the locale
1500 * actually says, but should give good enough results for someone using
1501 * them as formats (as opposed to trying to parse them to figure out
7d4bcc4a 1502 * what the locale says). The other format items are actually tested to
f7416781
KW
1503 * verify they work on the platform */
1504 case PERL_D_FMT: return "%x";
1505 case PERL_T_FMT: return "%X";
1506 case PERL_D_T_FMT: return "%c";
1507
1508 /* These formats are only available in later strfmtime's */
1509 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1510 case PERL_T_FMT_AMPM:
1511
1512 /* The rest can be gotten from most versions of strftime(). */
1513 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1514 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1515 case PERL_ABDAY_7:
1516 case PERL_ALT_DIGITS:
1517 case PERL_AM_STR: case PERL_PM_STR:
1518 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1519 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1520 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1521 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1522 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1523 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1524 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1525 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1526 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11: case PERL_MON_12:
1527
1528 LOCALE_LOCK;
1529
1530 init_tm(&tm); /* Precaution against core dumps */
1531 tm.tm_sec = 30;
1532 tm.tm_min = 30;
1533 tm.tm_hour = 6;
1534 tm.tm_year = 2017 - 1900;
1535 tm.tm_wday = 0;
1536 tm.tm_mon = 0;
1537 switch (item) {
1538 default:
1539 LOCALE_UNLOCK;
1540 Perl_croak(aTHX_ "panic: %s: %d: switch case: %d problem",
1541 __FILE__, __LINE__, item);
1542 NOT_REACHED; /* NOTREACHED */
1543
1544 case PERL_PM_STR: tm.tm_hour = 18;
1545 case PERL_AM_STR:
1546 format = "%p";
1547 break;
1548
1549 case PERL_ABDAY_7: tm.tm_wday++;
1550 case PERL_ABDAY_6: tm.tm_wday++;
1551 case PERL_ABDAY_5: tm.tm_wday++;
1552 case PERL_ABDAY_4: tm.tm_wday++;
1553 case PERL_ABDAY_3: tm.tm_wday++;
1554 case PERL_ABDAY_2: tm.tm_wday++;
1555 case PERL_ABDAY_1:
1556 format = "%a";
1557 break;
1558
1559 case PERL_DAY_7: tm.tm_wday++;
1560 case PERL_DAY_6: tm.tm_wday++;
1561 case PERL_DAY_5: tm.tm_wday++;
1562 case PERL_DAY_4: tm.tm_wday++;
1563 case PERL_DAY_3: tm.tm_wday++;
1564 case PERL_DAY_2: tm.tm_wday++;
1565 case PERL_DAY_1:
1566 format = "%A";
1567 break;
1568
1569 case PERL_ABMON_12: tm.tm_mon++;
1570 case PERL_ABMON_11: tm.tm_mon++;
1571 case PERL_ABMON_10: tm.tm_mon++;
1572 case PERL_ABMON_9: tm.tm_mon++;
1573 case PERL_ABMON_8: tm.tm_mon++;
1574 case PERL_ABMON_7: tm.tm_mon++;
1575 case PERL_ABMON_6: tm.tm_mon++;
1576 case PERL_ABMON_5: tm.tm_mon++;
1577 case PERL_ABMON_4: tm.tm_mon++;
1578 case PERL_ABMON_3: tm.tm_mon++;
1579 case PERL_ABMON_2: tm.tm_mon++;
1580 case PERL_ABMON_1:
1581 format = "%b";
1582 break;
1583
1584 case PERL_MON_12: tm.tm_mon++;
1585 case PERL_MON_11: tm.tm_mon++;
1586 case PERL_MON_10: tm.tm_mon++;
1587 case PERL_MON_9: tm.tm_mon++;
1588 case PERL_MON_8: tm.tm_mon++;
1589 case PERL_MON_7: tm.tm_mon++;
1590 case PERL_MON_6: tm.tm_mon++;
1591 case PERL_MON_5: tm.tm_mon++;
1592 case PERL_MON_4: tm.tm_mon++;
1593 case PERL_MON_3: tm.tm_mon++;
1594 case PERL_MON_2: tm.tm_mon++;
1595 case PERL_MON_1:
1596 format = "%B";
1597 break;
1598
1599 case PERL_T_FMT_AMPM:
1600 format = "%r";
1601 return_format = TRUE;
1602 break;
1603
1604 case PERL_ERA_D_FMT:
1605 format = "%Ex";
1606 return_format = TRUE;
1607 break;
1608
1609 case PERL_ERA_T_FMT:
1610 format = "%EX";
1611 return_format = TRUE;
1612 break;
1613
1614 case PERL_ERA_D_T_FMT:
1615 format = "%Ec";
1616 return_format = TRUE;
1617 break;
1618
1619 case PERL_ALT_DIGITS:
1620 tm.tm_wday = 0;
1621 format = "%Ow"; /* Find the alternate digit for 0 */
1622 break;
1623 }
1624
1625 /* We can't use my_strftime() because it doesn't look at tm_wday */
1626 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1627 format, &tm))
1628 {
1629 /* A zero return means one of:
1630 * a) there wasn't enough space in PL_langinfo_buf
1631 * b) the format, like a plain %p, returns empty
1632 * c) it was an illegal format, though some implementations of
1633 * strftime will just return the illegal format as a plain
1634 * character sequence.
1635 *
1636 * To quickly test for case 'b)', try again but precede the
1637 * format with a plain character. If that result is still
1638 * empty, the problem is either 'a)' or 'c)' */
1639
1640 Size_t format_size = strlen(format) + 1;
1641 Size_t mod_size = format_size + 1;
1642 char * mod_format;
1643 char * temp_result;
1644
1645 Newx(mod_format, mod_size, char);
1646 Newx(temp_result, PL_langinfo_bufsize, char);
1647 *mod_format = '\a';
1648 my_strlcpy(mod_format + 1, format, mod_size);
1649 len = strftime(temp_result,
1650 PL_langinfo_bufsize,
1651 mod_format, &tm);
1652 Safefree(mod_format);
1653 Safefree(temp_result);
1654
1655 /* If 'len' is non-zero, it means that we had a case like %p
1656 * which means the current locale doesn't use a.m. or p.m., and
1657 * that is valid */
1658 if (len == 0) {
1659
1660 /* Here, still didn't work. If we get well beyond a
1661 * reasonable size, bail out to prevent an infinite loop. */
1662
1663 if (PL_langinfo_bufsize > 100 * format_size) {
1664 *PL_langinfo_buf = '\0';
1665 }
1666 else { /* Double the buffer size to retry; Add 1 in case
1667 original was 0, so we aren't stuck at 0. */
1668 PL_langinfo_bufsize *= 2;
1669 PL_langinfo_bufsize++;
1670 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1671 continue;
1672 }
1673 }
1674
1675 break;
1676 }
1677
1678 /* Here, we got a result.
1679 *
1680 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1681 * alternate format for wday 0. If the value is the same as the
1682 * normal 0, there isn't an alternate, so clear the buffer. */
1683 if ( item == PERL_ALT_DIGITS
1684 && strEQ(PL_langinfo_buf, "0"))
1685 {
1686 *PL_langinfo_buf = '\0';
1687 }
1688
1689 /* ALT_DIGITS is problematic. Experiments on it showed that
1690 * strftime() did not always work properly when going from alt-9 to
1691 * alt-10. Only a few locales have this item defined, and in all
1692 * of them on Linux that khw was able to find, nl_langinfo() merely
1693 * returned the alt-0 character, possibly doubled. Most Unicode
1694 * digits are in blocks of 10 consecutive code points, so that is
1695 * sufficient information for those scripts, as we can infer alt-1,
1696 * alt-2, .... But for a Japanese locale, a CJK ideographic 0 is
1697 * returned, and the CJK digits are not in code point order, so you
1698 * can't really infer anything. The localedef for this locale did
1699 * specify the succeeding digits, so that strftime() works properly
1700 * on them, without needing to infer anything. But the
1701 * nl_langinfo() return did not give sufficient information for the
1702 * caller to understand what's going on. So until there is
1703 * evidence that it should work differently, this returns the alt-0
1704 * string for ALT_DIGITS.
1705 *
1706 * wday was chosen because its range is all a single digit. Things
1707 * like tm_sec have two digits as the minimum: '00' */
1708
1709 LOCALE_UNLOCK;
1710
1711 /* If to return the format, not the value, overwrite the buffer
1712 * with it. But some strftime()s will keep the original format if
1713 * illegal, so change those to "" */
1714 if (return_format) {
1715 if (strEQ(PL_langinfo_buf, format)) {
1716 *PL_langinfo_buf = '\0';
1717 }
1718 else {
1719 save_to_buffer(format, &PL_langinfo_buf,
1720 &PL_langinfo_bufsize, 0);
1721 }
1722 }
1723
1724 break;
1725
1726# endif
1727
1728 }
1729
1730 return PL_langinfo_buf;
1731
1732#endif
1733
a4f00dcc 1734}
b385bb4d 1735
98994639
HS
1736/*
1737 * Initialize locale awareness.
1738 */
1739int
1740Perl_init_i18nl10n(pTHX_ int printwarn)
1741{
0e92a118
KW
1742 /* printwarn is
1743 *
1744 * 0 if not to output warning when setup locale is bad
1745 * 1 if to output warning based on value of PERL_BADLANG
1746 * >1 if to output regardless of PERL_BADLANG
1747 *
1748 * returns
98994639 1749 * 1 = set ok or not applicable,
0e92a118
KW
1750 * 0 = fallback to a locale of lower priority
1751 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1752 *
1753 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1754 * set, debugging information is output.
1755 *
1756 * This looks more complicated than it is, mainly due to the #ifdefs.
1757 *
1758 * We try to set LC_ALL to the value determined by the environment. If
1759 * there is no LC_ALL on this platform, we try the individual categories we
1760 * know about. If this works, we are done.
1761 *
1762 * But if it doesn't work, we have to do something else. We search the
1763 * environment variables ourselves instead of relying on the system to do
1764 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1765 * think there is one), and the ultimate fallback "C". This is all done in
1766 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
1767 * more complex. The 'trial_locales' array is initialized with just one
1768 * element; it causes the behavior described in the paragraph above this to
1769 * happen. If that fails, we add elements to 'trial_locales', and do extra
1770 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
1771 *
1772 * On Ultrix, the locale MUST come from the environment, so there is
1773 * preliminary code to set it. I (khw) am not sure that it is necessary,
1774 * and that this couldn't be folded into the loop, but barring any real
1775 * platforms to test on, it's staying as-is
1776 *
1777 * A slight complication is that in embedded Perls, the locale may already
1778 * be set-up, and we don't want to get it from the normal environment
1779 * variables. This is handled by having a special environment variable
1780 * indicate we're in this situation. We simply set setlocale's 2nd
1781 * parameter to be a NULL instead of "". That indicates to setlocale that
1782 * it is not to change anything, but to return the current value,
1783 * effectively initializing perl's db to what the locale already is.
1784 *
1785 * We play the same trick with NULL if a LC_ALL succeeds. We call
1786 * setlocale() on the individual categores with NULL to get their existing
1787 * values for our db, instead of trying to change them.
1788 * */
98994639 1789
0e92a118
KW
1790 int ok = 1;
1791
7d4bcc4a
KW
1792#ifndef USE_LOCALE
1793
1794 PERL_UNUSED_ARG(printwarn);
1795
1796#else /* USE_LOCALE */
7d4bcc4a
KW
1797# ifdef __GLIBC__
1798
175c4cf9 1799 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
1800
1801# endif
65ebb059 1802
ccd65d51
KW
1803 /* NULL uses the existing already set up locale */
1804 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
1805 ? NULL
1806 : "";
c3fcd832
KW
1807 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
1808 unsigned int trial_locales_count;
175c4cf9
KW
1809 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
1810 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 1811 bool setlocale_failure = FALSE;
65ebb059 1812 unsigned int i;
175c4cf9
KW
1813
1814 /* A later getenv() could zap this, so only use here */
1815 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
1816
1817 const bool locwarn = (printwarn > 1
e5f10d49
KW
1818 || ( printwarn
1819 && ( ! bad_lang_use_once
22ff3130 1820 || (
e5f10d49
KW
1821 /* disallow with "" or "0" */
1822 *bad_lang_use_once
1823 && strNE("0", bad_lang_use_once)))));
0e92a118 1824 bool done = FALSE;
291a84fb 1825 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 1826 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
1827
1828 /* current locale for given category; should have been copied so aren't
1829 * volatile */
8de4332b 1830 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 1831
5d1187d1 1832 char * locale_param;
7d4bcc4a
KW
1833
1834# ifdef WIN32
1835
6bce99ee
JH
1836 /* In some systems you can find out the system default locale
1837 * and use that as the fallback locale. */
7d4bcc4a
KW
1838# define SYSTEM_DEFAULT_LOCALE
1839# endif
1840# ifdef SYSTEM_DEFAULT_LOCALE
1841
65ebb059 1842 const char *system_default_locale = NULL;
98994639 1843
7d4bcc4a 1844# endif
948523db
KW
1845
1846# ifndef DEBUGGING
1847# define DEBUG_LOCALE_INIT(a,b,c)
1848# else
7d4bcc4a 1849
8298454c 1850 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
1851
1852# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
1853 STMT_START { \
1854 if (debug_initialization) { \
1855 PerlIO_printf(Perl_debug_log, \
1856 "%s:%d: %s\n", \
1857 __FILE__, __LINE__, \
a4f00dcc 1858 setlocale_debug_string(category, \
2fcc0ca9
KW
1859 locale, \
1860 result)); \
1861 } \
1862 } STMT_END
2fcc0ca9 1863
948523db
KW
1864/* Make sure the parallel arrays are properly set up */
1865# ifdef USE_LOCALE_NUMERIC
1866 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
1867 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
1868# endif
1869# ifdef USE_LOCALE_CTYPE
1870 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
1871 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
1872# endif
1873# ifdef USE_LOCALE_COLLATE
1874 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
1875 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
1876# endif
1877# ifdef USE_LOCALE_TIME
1878 assert(categories[LC_TIME_INDEX] == LC_TIME);
1879 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
1880# endif
1881# ifdef USE_LOCALE_MESSAGES
1882 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
1883 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
1884# endif
1885# ifdef USE_LOCALE_MONETARY
1886 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
1887 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
1888# endif
1889# ifdef LC_ALL
1890 assert(categories[LC_ALL_INDEX] == LC_ALL);
1891 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
1892 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
1893# endif
1894# endif /* DEBUGGING */
7d4bcc4a
KW
1895# ifndef LOCALE_ENVIRON_REQUIRED
1896
0e92a118 1897 PERL_UNUSED_VAR(done);
5d1187d1 1898 PERL_UNUSED_VAR(locale_param);
7d4bcc4a
KW
1899
1900# else
98994639
HS
1901
1902 /*
1903 * Ultrix setlocale(..., "") fails if there are no environment
1904 * variables from which to get a locale name.
1905 */
1906
7d4bcc4a
KW
1907# ifdef LC_ALL
1908
98994639 1909 if (lang) {
948523db
KW
1910 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
1911 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
1912 if (sl_result[LC_ALL_INDEX])
98994639
HS
1913 done = TRUE;
1914 else
1915 setlocale_failure = TRUE;
1916 }
5d1187d1 1917 if (! setlocale_failure) {
948523db 1918 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
1919 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
1920 ? setlocale_init
1921 : NULL;
1922 sl_result[i] = do_setlocale_r(categories[i], locale_param);
1923 if (! sl_result[i]) {
1924 setlocale_failure = TRUE;
1925 }
1926 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
c835d6be 1927 }
7d4bcc4a
KW
1928 }
1929
1930# endif /* LC_ALL */
e5f10d49 1931# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 1932
65ebb059 1933 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
1934 * the list. Normally the loop is executed just once. But if setting the
1935 * locale fails, inside the loop we add fallback trials to the array and so
1936 * will execute the loop multiple times */
c3fcd832
KW
1937 trial_locales[0] = setlocale_init;
1938 trial_locales_count = 1;
7d4bcc4a 1939
65ebb059
KW
1940 for (i= 0; i < trial_locales_count; i++) {
1941 const char * trial_locale = trial_locales[i];
1942
1943 if (i > 0) {
1944
1945 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
1946 * when i==0, but I (khw) don't think that behavior makes much
1947 * sense */
1948 setlocale_failure = FALSE;
1949
7d4bcc4a 1950# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 1951# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 1952
65ebb059
KW
1953 /* On Windows machines, an entry of "" after the 0th means to use
1954 * the system default locale, which we now proceed to get. */
1955 if (strEQ(trial_locale, "")) {
1956 unsigned int j;
1957
1958 /* Note that this may change the locale, but we are going to do
1959 * that anyway just below */
837ce802 1960 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 1961 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 1962
7d4bcc4a 1963 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
1964 * try */
1965 if (! system_default_locale) {
1966 goto next_iteration;
1967 }
1968 for (j = 0; j < trial_locales_count; j++) {
1969 if (strEQ(system_default_locale, trial_locales[j])) {
1970 goto next_iteration;
1971 }
1972 }
1973
1974 trial_locale = system_default_locale;
1975 }
7d4bcc4a
KW
1976# endif /* WIN32 */
1977# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
1978
1979 } /* For i > 0 */
65ebb059 1980
7d4bcc4a
KW
1981# ifdef LC_ALL
1982
948523db
KW
1983 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
1984 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
1985 if (! sl_result[LC_ALL_INDEX]) {
49c85077 1986 setlocale_failure = TRUE;
7cd8b568
KW
1987 }
1988 else {
1989 /* Since LC_ALL succeeded, it should have changed all the other
1990 * categories it can to its value; so we massage things so that the
1991 * setlocales below just return their category's current values.
1992 * This adequately handles the case in NetBSD where LC_COLLATE may
1993 * not be defined for a locale, and setting it individually will
7d4bcc4a 1994 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
1995 * the POSIX locale. */
1996 trial_locale = NULL;
1997 }
7d4bcc4a
KW
1998
1999# endif /* LC_ALL */
98994639 2000
e5f10d49
KW
2001 if (! setlocale_failure) {
2002 unsigned int j;
2003 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2004 curlocales[j]
2005 = savepv(do_setlocale_r(categories[j], trial_locale));
2006 if (! curlocales[j]) {
2007 setlocale_failure = TRUE;
2008 }
2009 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2010 }
c835d6be 2011
e5f10d49
KW
2012 if (! setlocale_failure) { /* All succeeded */
2013 break; /* Exit trial_locales loop */
49c85077 2014 }
65ebb059 2015 }
98994639 2016
49c85077
KW
2017 /* Here, something failed; will need to try a fallback. */
2018 ok = 0;
65ebb059 2019
49c85077
KW
2020 if (i == 0) {
2021 unsigned int j;
98994639 2022
65ebb059 2023 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2024
2025# ifdef LC_ALL
98994639 2026
49c85077
KW
2027 PerlIO_printf(Perl_error_log,
2028 "perl: warning: Setting locale failed.\n");
98994639 2029
7d4bcc4a 2030# else /* !LC_ALL */
98994639 2031
49c85077
KW
2032 PerlIO_printf(Perl_error_log,
2033 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2034
e5f10d49
KW
2035 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2036 if (! curlocales[j]) {
2037 PerlIO_printf(Perl_error_log, category_names[j]);
2038 }
2039 else {
2040 Safefree(curlocales[j]);
2041 }
2042 }
7d4bcc4a 2043
a782673d 2044 PerlIO_printf(Perl_error_log, "and possibly others\n");
98994639 2045
7d4bcc4a 2046# endif /* LC_ALL */
98994639 2047
49c85077
KW
2048 PerlIO_printf(Perl_error_log,
2049 "perl: warning: Please check that your locale settings:\n");
98994639 2050
7d4bcc4a
KW
2051# ifdef __GLIBC__
2052
49c85077
KW
2053 PerlIO_printf(Perl_error_log,
2054 "\tLANGUAGE = %c%s%c,\n",
2055 language ? '"' : '(',
2056 language ? language : "unset",
2057 language ? '"' : ')');
7d4bcc4a 2058# endif
98994639 2059
49c85077
KW
2060 PerlIO_printf(Perl_error_log,
2061 "\tLC_ALL = %c%s%c,\n",
2062 lc_all ? '"' : '(',
2063 lc_all ? lc_all : "unset",
2064 lc_all ? '"' : ')');
98994639 2065
7d4bcc4a
KW
2066# if defined(USE_ENVIRON_ARRAY)
2067
49c85077 2068 {
cd999af9 2069 char **e;
d5e32b93
KW
2070
2071 /* Look through the environment for any variables of the
2072 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2073 * already handled above. These are assumed to be locale
2074 * settings. Output them and their values. */
cd999af9 2075 for (e = environ; *e; e++) {
d5e32b93
KW
2076 const STRLEN prefix_len = sizeof("LC_") - 1;
2077 STRLEN uppers_len;
2078
cd999af9 2079 if ( strBEGINs(*e, "LC_")
c8b388b0 2080 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2081 && (uppers_len = strspn(*e + prefix_len,
2082 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2083 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2084 {
2085 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2086 (int) (prefix_len + uppers_len), *e,
2087 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2088 }
2089 }
49c85077 2090 }
7d4bcc4a
KW
2091
2092# else
2093
49c85077
KW
2094 PerlIO_printf(Perl_error_log,
2095 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2096
2097# endif
98994639 2098
49c85077
KW
2099 PerlIO_printf(Perl_error_log,
2100 "\tLANG = %c%s%c\n",
2101 lang ? '"' : '(',
2102 lang ? lang : "unset",
2103 lang ? '"' : ')');
98994639 2104
49c85077
KW
2105 PerlIO_printf(Perl_error_log,
2106 " are supported and installed on your system.\n");
2107 }
98994639 2108
65ebb059 2109 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2110 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2111 * usually change the upper bound of the loop we are in.
2112 *
2113 * Since the system's default way of setting the locale has not
2114 * found one that works, We use Perl's defined ordering: LC_ALL,
2115 * LANG, and the C locale. We don't try the same locale twice, so
2116 * don't add to the list if already there. (On POSIX systems, the
2117 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2118 * but there's no harm done by doing it explicitly.
2119 *
2120 * Note that this tries the LC_ALL environment variable even on
2121 * systems which have no LC_ALL locale setting. This may or may
2122 * not have been originally intentional, but there's no real need
2123 * to change the behavior. */
65ebb059
KW
2124 if (lc_all) {
2125 for (j = 0; j < trial_locales_count; j++) {
2126 if (strEQ(lc_all, trial_locales[j])) {
2127 goto done_lc_all;
2128 }
2129 }
2130 trial_locales[trial_locales_count++] = lc_all;
2131 }
2132 done_lc_all:
98994639 2133
65ebb059
KW
2134 if (lang) {
2135 for (j = 0; j < trial_locales_count; j++) {
2136 if (strEQ(lang, trial_locales[j])) {
2137 goto done_lang;
2138 }
2139 }
2140 trial_locales[trial_locales_count++] = lang;
2141 }
2142 done_lang:
2143
7d4bcc4a
KW
2144# if defined(WIN32) && defined(LC_ALL)
2145
65ebb059
KW
2146 /* For Windows, we also try the system default locale before "C".
2147 * (If there exists a Windows without LC_ALL we skip this because
2148 * it gets too complicated. For those, the "C" is the next
2149 * fallback possibility). The "" is the same as the 0th element of
2150 * the array, but the code at the loop above knows to treat it
2151 * differently when not the 0th */
2152 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2153
2154# endif
65ebb059
KW
2155
2156 for (j = 0; j < trial_locales_count; j++) {
2157 if (strEQ("C", trial_locales[j])) {
2158 goto done_C;
2159 }
2160 }
2161 trial_locales[trial_locales_count++] = "C";
98994639 2162
65ebb059
KW
2163 done_C: ;
2164 } /* end of first time through the loop */
98994639 2165
7d4bcc4a
KW
2166# ifdef WIN32
2167
65ebb059 2168 next_iteration: ;
7d4bcc4a
KW
2169
2170# endif
65ebb059
KW
2171
2172 } /* end of looping through the trial locales */
2173
2174 if (ok < 1) { /* If we tried to fallback */
2175 const char* msg;
2176 if (! setlocale_failure) { /* fallback succeeded */
2177 msg = "Falling back to";
2178 }
2179 else { /* fallback failed */
e5f10d49 2180 unsigned int j;
98994639 2181
65ebb059
KW
2182 /* We dropped off the end of the loop, so have to decrement i to
2183 * get back to the value the last time through */
2184 i--;
98994639 2185
65ebb059
KW
2186 ok = -1;
2187 msg = "Failed to fall back to";
2188
2189 /* To continue, we should use whatever values we've got */
7d4bcc4a 2190
e5f10d49
KW
2191 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2192 Safefree(curlocales[j]);
2193 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2194 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2195 }
65ebb059
KW
2196 }
2197
2198 if (locwarn) {
2199 const char * description;
2200 const char * name = "";
2201 if (strEQ(trial_locales[i], "C")) {
2202 description = "the standard locale";
2203 name = "C";
2204 }
7d4bcc4a
KW
2205
2206# ifdef SYSTEM_DEFAULT_LOCALE
2207
65ebb059
KW
2208 else if (strEQ(trial_locales[i], "")) {
2209 description = "the system default locale";
2210 if (system_default_locale) {
2211 name = system_default_locale;
2212 }
2213 }
7d4bcc4a
KW
2214
2215# endif /* SYSTEM_DEFAULT_LOCALE */
2216
65ebb059
KW
2217 else {
2218 description = "a fallback locale";
2219 name = trial_locales[i];
2220 }
2221 if (name && strNE(name, "")) {
2222 PerlIO_printf(Perl_error_log,
2223 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2224 }
2225 else {
2226 PerlIO_printf(Perl_error_log,
2227 "perl: warning: %s %s.\n", msg, description);
2228 }
2229 }
2230 } /* End of tried to fallback */
98994639 2231
e5f10d49
KW
2232 /* Done with finding the locales; update our records */
2233
7d4bcc4a
KW
2234# ifdef USE_LOCALE_CTYPE
2235
948523db 2236 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2237
e5f10d49 2238# endif
7d4bcc4a
KW
2239# ifdef USE_LOCALE_COLLATE
2240
948523db 2241 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2242
e5f10d49 2243# endif
7d4bcc4a
KW
2244# ifdef USE_LOCALE_NUMERIC
2245
948523db 2246 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2247
2248# endif
2249
e5f10d49 2250
948523db 2251 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
e5f10d49
KW
2252 Safefree(curlocales[i]);
2253 }
b310b053 2254
7d4bcc4a
KW
2255# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2256
49c85077
KW
2257 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
2258 * locale is UTF-8. If PL_utf8locale and PL_unicode (set by -C or by
2259 * $ENV{PERL_UNICODE}) are true, perl.c:S_parse_body() will turn on the
2260 * PerlIO :utf8 layer on STDIN, STDOUT, STDERR, _and_ the default open
2261 * discipline. */
c1284011 2262 PL_utf8locale = _is_cur_LC_category_utf8(LC_CTYPE);
49c85077 2263
a05d7ebb 2264 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2265 This is an alternative to using the -C command line switch
2266 (the -C if present will override this). */
2267 {
dd374669 2268 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2269 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2270 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2271 PL_utf8cache = -1;
b310b053
JH
2272 }
2273
7d4bcc4a 2274# endif
7d4bcc4a
KW
2275# ifdef __GLIBC__
2276
175c4cf9 2277 Safefree(language);
7d4bcc4a
KW
2278
2279# endif
175c4cf9
KW
2280
2281 Safefree(lc_all);
2282 Safefree(lang);
2283
e3305790 2284#endif /* USE_LOCALE */
2fcc0ca9 2285#ifdef DEBUGGING
7d4bcc4a 2286
2fcc0ca9 2287 /* So won't continue to output stuff */
27cdc72e 2288 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2289
2fcc0ca9
KW
2290#endif
2291
98994639
HS
2292 return ok;
2293}
2294
98994639
HS
2295#ifdef USE_LOCALE_COLLATE
2296
a4a439fb 2297char *
a4a439fb
KW
2298Perl__mem_collxfrm(pTHX_ const char *input_string,
2299 STRLEN len, /* Length of 'input_string' */
2300 STRLEN *xlen, /* Set to length of returned string
2301 (not including the collation index
2302 prefix) */
2303 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2304 )
98994639 2305{
a4a439fb
KW
2306
2307 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2308 * differences. First, it handles embedded NULs. Second, it allocates a bit
2309 * more memory than needed for the transformed data itself. The real
55e5378d 2310 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2311 * the length of that, and doesn't include the collation index size.
2312 * Please see sv_collxfrm() to see how this is used. */
2313
55e5378d
KW
2314#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2315
6696cfa7
KW
2316 char * s = (char *) input_string;
2317 STRLEN s_strlen = strlen(input_string);
79f120c8 2318 char *xbuf = NULL;
55e5378d 2319 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2320 STRLEN length_in_chars;
c664130f 2321 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2322
a4a439fb
KW
2323 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2324
2325 /* Must be NUL-terminated */
2326 assert(*(input_string + len) == '\0');
7918f24d 2327
79f120c8
KW
2328 /* If this locale has defective collation, skip */
2329 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2330 DEBUG_L(PerlIO_printf(Perl_debug_log,
2331 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2332 goto bad;
2333 }
2334
6696cfa7
KW
2335 /* Replace any embedded NULs with the control that sorts before any others.
2336 * This will give as good as possible results on strings that don't
2337 * otherwise contain that character, but otherwise there may be
2338 * less-than-perfect results with that character and NUL. This is
fdc080f3 2339 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2340 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2341 NUL */
6696cfa7
KW
2342 char * e = s + len;
2343 char * sans_nuls;
fdc080f3 2344 STRLEN sans_nuls_len;
94762aa0 2345 int try_non_controls;
afc4976f
KW
2346 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2347 making sure 2nd byte is NUL.
2348 */
2349 STRLEN this_replacement_len;
2350
1e4c9676
KW
2351 /* If we don't know what non-NUL control character sorts lowest for
2352 * this locale, find it */
f28f4d2a 2353 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2354 int j;
afc4976f 2355 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2356 includes the collation index
2357 prefixed. */
2358
91c0e2e0 2359 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2360
2361 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2362 * NUL, in which case we instead look for any character. Controls
2363 * are preferred because collation order is, in general, context
2364 * sensitive, with adjoining characters affecting the order, and
2365 * controls are less likely to have such interactions, allowing the
2366 * NUL-replacement to stand on its own. (Another way to look at it
2367 * is to imagine what would happen if the NUL were replaced by a
2368 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2369 for (try_non_controls = 0;
2370 try_non_controls < 2;
2371 try_non_controls++)
2372 {
d4ff9586
KW
2373 /* Look through all legal code points (NUL isn't) */
2374 for (j = 1; j < 256; j++) {
2375 char * x; /* j's xfrm plus collation index */
2376 STRLEN x_len; /* length of 'x' */
2377 STRLEN trial_len = 1;
736a4fed 2378 char cur_source[] = { '\0', '\0' };
d4ff9586 2379
736a4fed
KW
2380 /* Skip non-controls the first time through the loop. The
2381 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2382 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2383 ? ! isCNTRL_L1(j)
2384 : ! isCNTRL_LC(j))
2385 {
d4ff9586 2386 continue;
6696cfa7 2387 }
6696cfa7 2388
736a4fed
KW
2389 /* Create a 1-char string of the current code point */
2390 cur_source[0] = (char) j;
2391
d4ff9586
KW
2392 /* Then transform it */
2393 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2394 0 /* The string is not in UTF-8 */);
6696cfa7 2395
1e4c9676 2396 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2397 * */
2398 if (! x) {
2399 continue;
2400 }
6696cfa7 2401
d4ff9586
KW
2402 /* If this character's transformation is lower than
2403 * the current lowest, this one becomes the lowest */
2404 if ( cur_min_x == NULL
2405 || strLT(x + COLLXFRM_HDR_LEN,
2406 cur_min_x + COLLXFRM_HDR_LEN))
2407 {
f28f4d2a 2408 PL_strxfrm_NUL_replacement = j;
d4ff9586 2409 cur_min_x = x;
d4ff9586
KW
2410 }
2411 else {
2412 Safefree(x);
2413 }
1e4c9676 2414 } /* end of loop through all 255 characters */
6696cfa7 2415
1e4c9676 2416 /* Stop looking if found */
94762aa0
KW
2417 if (cur_min_x) {
2418 break;
2419 }
2420
2421 /* Unlikely, but possible, if there aren't any controls that
2422 * work in the locale, repeat the loop, looking for any
2423 * character that works */
2424 DEBUG_L(PerlIO_printf(Perl_debug_log,
2425 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2426 } /* End of loop to try first the controls, then any char */
6696cfa7 2427
94762aa0
KW
2428 if (! cur_min_x) {
2429 DEBUG_L(PerlIO_printf(Perl_debug_log,
2430 "_mem_collxfrm: Couldn't find any character to replace"
2431 " embedded NULs in locale %s with", PL_collation_name));
2432 goto bad;
58eebef2
KW
2433 }
2434
94762aa0
KW
2435 DEBUG_L(PerlIO_printf(Perl_debug_log,
2436 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2437 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2438
6696cfa7 2439 Safefree(cur_min_x);
1e4c9676 2440 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2441
2442 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2443 * UTF8-ness of the original */
f28f4d2a
KW
2444 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2445 this_replacement_char[0] =
2446 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2447 this_replacement_char[1] =
2448 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2449 this_replacement_len = 2;
2450 }
2451 else {
f28f4d2a 2452 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2453 /* this_replacement_char[1] = '\0' was done at initialization */
2454 this_replacement_len = 1;
6696cfa7
KW
2455 }
2456
2457 /* The worst case length for the replaced string would be if every
2458 * character in it is NUL. Multiply that by the length of each
2459 * replacement, and allow for a trailing NUL */
afc4976f 2460 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2461 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2462 *sans_nuls = '\0';
2463
6696cfa7
KW
2464 /* Replace each NUL with the lowest collating control. Loop until have
2465 * exhausted all the NULs */
2466 while (s + s_strlen < e) {
6069d6c5 2467 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2468
2469 /* Do the actual replacement */
6069d6c5 2470 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2471
2472 /* Move past the input NUL */
2473 s += s_strlen + 1;
2474 s_strlen = strlen(s);
2475 }
2476
2477 /* And add anything that trails the final NUL */
6069d6c5 2478 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2479
2480 /* Switch so below we transform this modified string */
2481 s = sans_nuls;
2482 len = strlen(s);
1e4c9676 2483 } /* End of replacing NULs */
6696cfa7 2484
a4a439fb
KW
2485 /* Make sure the UTF8ness of the string and locale match */
2486 if (utf8 != PL_in_utf8_COLLATE_locale) {
2487 const char * const t = s; /* Temporary so we can later find where the
2488 input was */
2489
2490 /* Here they don't match. Change the string's to be what the locale is
2491 * expecting */
2492
2493 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2494 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2495 utf8 = TRUE;
2496 }
2497 else { /* locale is not UTF-8; but input is; downgrade the input */
2498
2499 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2500
2501 /* If the downgrade was successful we are done, but if the input
2502 * contains things that require UTF-8 to represent, have to do
2503 * damage control ... */
2504 if (UNLIKELY(utf8)) {
2505
2506 /* What we do is construct a non-UTF-8 string with
2507 * 1) the characters representable by a single byte converted
2508 * to be so (if necessary);
2509 * 2) and the rest converted to collate the same as the
2510 * highest collating representable character. That makes
2511 * them collate at the end. This is similar to how we
2512 * handle embedded NULs, but we use the highest collating
2513 * code point instead of the smallest. Like the NUL case,
2514 * this isn't perfect, but is the best we can reasonably
2515 * do. Every above-255 code point will sort the same as
2516 * the highest-sorting 0-255 code point. If that code
2517 * point can combine in a sequence with some other code
2518 * points for weight calculations, us changing something to
2519 * be it can adversely affect the results. But in most
2520 * cases, it should work reasonably. And note that this is
2521 * really an illegal situation: using code points above 255
2522 * on a locale where only 0-255 are valid. If two strings
2523 * sort entirely equal, then the sort order for the
2524 * above-255 code points will be in code point order. */
2525
2526 utf8 = FALSE;
2527
2528 /* If we haven't calculated the code point with the maximum
2529 * collating order for this locale, do so now */
2530 if (! PL_strxfrm_max_cp) {
2531 int j;
2532
2533 /* The current transformed string that collates the
2534 * highest (except it also includes the prefixed collation
2535 * index. */
2536 char * cur_max_x = NULL;
2537
2538 /* Look through all legal code points (NUL isn't) */
2539 for (j = 1; j < 256; j++) {
2540 char * x;
2541 STRLEN x_len;
736a4fed 2542 char cur_source[] = { '\0', '\0' };
a4a439fb 2543
736a4fed
KW
2544 /* Create a 1-char string of the current code point */
2545 cur_source[0] = (char) j;
a4a439fb
KW
2546
2547 /* Then transform it */
2548 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2549
2550 /* If something went wrong (which it shouldn't), just
2551 * ignore this code point */
94762aa0 2552 if (! x) {
a4a439fb
KW
2553 continue;
2554 }
2555
2556 /* If this character's transformation is higher than
2557 * the current highest, this one becomes the highest */
2558 if ( cur_max_x == NULL
55e5378d
KW
2559 || strGT(x + COLLXFRM_HDR_LEN,
2560 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2561 {
2562 PL_strxfrm_max_cp = j;
2563 cur_max_x = x;
2564 }
2565 else {
2566 Safefree(x);
2567 }
2568 }
2569
94762aa0
KW
2570 if (! cur_max_x) {
2571 DEBUG_L(PerlIO_printf(Perl_debug_log,
2572 "_mem_collxfrm: Couldn't find any character to"
2573 " replace above-Latin1 chars in locale %s with",
2574 PL_collation_name));
2575 goto bad;
2576 }
2577
58eebef2
KW
2578 DEBUG_L(PerlIO_printf(Perl_debug_log,
2579 "_mem_collxfrm: highest 1-byte collating character"
2580 " in locale %s is 0x%02X\n",
2581 PL_collation_name,
2582 PL_strxfrm_max_cp));
58eebef2 2583
a4a439fb
KW
2584 Safefree(cur_max_x);
2585 }
2586
2587 /* Here we know which legal code point collates the highest.
2588 * We are ready to construct the non-UTF-8 string. The length
2589 * will be at least 1 byte smaller than the input string
2590 * (because we changed at least one 2-byte character into a
2591 * single byte), but that is eaten up by the trailing NUL */
2592 Newx(s, len, char);
2593
2594 {
2595 STRLEN i;
2596 STRLEN d= 0;
042d9e50 2597 char * e = (char *) t + len;
a4a439fb
KW
2598
2599 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2600 U8 cur_char = t[i];
2601 if (UTF8_IS_INVARIANT(cur_char)) {
2602 s[d++] = cur_char;
2603 }
042d9e50 2604 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2605 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2606 }
2607 else { /* Replace illegal cp with highest collating
2608 one */
2609 s[d++] = PL_strxfrm_max_cp;
2610 }
2611 }
2612 s[d++] = '\0';
2613 Renew(s, d, char); /* Free up unused space */
2614 }
2615 }
2616 }
2617
2618 /* Here, we have constructed a modified version of the input. It could
2619 * be that we already had a modified copy before we did this version.
2620 * If so, that copy is no longer needed */
2621 if (t != input_string) {
2622 Safefree(t);
2623 }
2624 }
2625
17f41037
KW
2626 length_in_chars = (utf8)
2627 ? utf8_length((U8 *) s, (U8 *) s + len)
2628 : len;
2629
59c018b9
KW
2630 /* The first element in the output is the collation id, used by
2631 * sv_collxfrm(); then comes the space for the transformed string. The
2632 * equation should give us a good estimate as to how much is needed */
55e5378d 2633 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2634 + PL_collxfrm_base
17f41037 2635 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2636 Newx(xbuf, xAlloc, char);
c7202dee
KW
2637 if (UNLIKELY(! xbuf)) {
2638 DEBUG_L(PerlIO_printf(Perl_debug_log,
2639 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2640 goto bad;
c7202dee 2641 }
98994639 2642
d35fca5f 2643 /* Store the collation id */
98994639 2644 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2645
2646 /* Then the transformation of the input. We loop until successful, or we
2647 * give up */
4ebeff16 2648 for (;;) {
1adab0a7 2649
55e5378d 2650 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2651
2652 /* If the transformed string occupies less space than we told strxfrm()
2653 * was available, it means it successfully transformed the whole
2654 * string. */
55e5378d 2655 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2656
1adab0a7
KW
2657 /* Some systems include a trailing NUL in the returned length.
2658 * Ignore it, using a loop in case multiple trailing NULs are
2659 * returned. */
2660 while ( (*xlen) > 0
2661 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2662 {
2663 (*xlen)--;
2664 }
2665
17f41037
KW
2666 /* If the first try didn't get it, it means our prediction was low.
2667 * Modify the coefficients so that we predict a larger value in any
2668 * future transformations */
2669 if (! first_time) {
2670 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2671 STRLEN computed_guess = PL_collxfrm_base
2672 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2673
2674 /* On zero-length input, just keep current slope instead of
2675 * dividing by 0 */
2676 const STRLEN new_m = (length_in_chars != 0)
2677 ? needed / length_in_chars
2678 : PL_collxfrm_mult;
17f41037
KW
2679
2680 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2681 "%s: %d: initial size of %zu bytes for a length "
2682 "%zu string was insufficient, %zu needed\n",
17f41037 2683 __FILE__, __LINE__,
b07929e4 2684 computed_guess, length_in_chars, needed));
17f41037
KW
2685
2686 /* If slope increased, use it, but discard this result for
2687 * length 1 strings, as we can't be sure that it's a real slope
2688 * change */
2689 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
2690
2691# ifdef DEBUGGING
2692
17f41037
KW
2693 STRLEN old_m = PL_collxfrm_mult;
2694 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
2695
2696# endif
2697
17f41037
KW
2698 PL_collxfrm_mult = new_m;
2699 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2700 computed_guess = PL_collxfrm_base
2701 + (PL_collxfrm_mult * length_in_chars);
2702 if (computed_guess < needed) {
2703 PL_collxfrm_base += needed - computed_guess;
2704 }
2705
2706 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2707 "%s: %d: slope is now %zu; was %zu, base "
2708 "is now %zu; was %zu\n",
17f41037 2709 __FILE__, __LINE__,
b07929e4
KW
2710 PL_collxfrm_mult, old_m,
2711 PL_collxfrm_base, old_b));
17f41037
KW
2712 }
2713 else { /* Slope didn't change, but 'b' did */
2714 const STRLEN new_b = needed
2715 - computed_guess
2716 + PL_collxfrm_base;
2717 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2718 "%s: %d: base is now %zu; was %zu\n",
17f41037 2719 __FILE__, __LINE__,
b07929e4 2720 new_b, PL_collxfrm_base));
17f41037
KW
2721 PL_collxfrm_base = new_b;
2722 }
2723 }
2724
4ebeff16
KW
2725 break;
2726 }
bb0f664e 2727
c7202dee
KW
2728 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2729 DEBUG_L(PerlIO_printf(Perl_debug_log,
2730 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2731 *xlen, PERL_INT_MAX));
4ebeff16 2732 goto bad;
c7202dee 2733 }
d35fca5f 2734
c664130f 2735 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
2736 * (usually not including the trailing NUL) when it fails due to not
2737 * enough space being provided. Assume that this is the case unless
2738 * it's been proven otherwise */
c664130f 2739 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 2740 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
2741 }
2742 else { /* Here, either:
2743 * 1) The strxfrm() has previously shown bad behavior; or
2744 * 2) It isn't the first time through the loop, which means
2745 * that the strxfrm() is now showing bad behavior, because
2746 * we gave it what it said was needed in the previous
2747 * iteration, and it came back saying it needed still more.
2748 * (Many versions of cygwin fit this. When the buffer size
2749 * isn't sufficient, they return the input size instead of
2750 * how much is needed.)
d4ff9586
KW
2751 * Increase the buffer size by a fixed percentage and try again.
2752 * */
6ddd902c 2753 xAlloc += (xAlloc / 4) + 1;
c664130f 2754 PL_strxfrm_is_behaved = FALSE;
c664130f 2755
7d4bcc4a
KW
2756# ifdef DEBUGGING
2757
58eebef2
KW
2758 if (DEBUG_Lv_TEST || debug_initialization) {
2759 PerlIO_printf(Perl_debug_log,
2760 "_mem_collxfrm required more space than previously calculated"
b07929e4 2761 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 2762 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 2763 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 2764 }
7d4bcc4a
KW
2765
2766# endif
2767
58eebef2 2768 }
c664130f 2769
4ebeff16 2770 Renew(xbuf, xAlloc, char);
c7202dee
KW
2771 if (UNLIKELY(! xbuf)) {
2772 DEBUG_L(PerlIO_printf(Perl_debug_log,
2773 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 2774 goto bad;
c7202dee 2775 }
c664130f
KW
2776
2777 first_time = FALSE;
4ebeff16 2778 }
98994639 2779
6696cfa7 2780
7d4bcc4a
KW
2781# ifdef DEBUGGING
2782
58eebef2 2783 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
2784
2785 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
2786 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
2787 PerlIO_printf(Perl_debug_log, "%s\n",
2788 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
2789 *xlen, 1));
58eebef2 2790 }
7d4bcc4a
KW
2791
2792# endif
58eebef2 2793
3c5f993e 2794 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 2795 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 2796
6696cfa7
KW
2797 if (s != input_string) {
2798 Safefree(s);
98994639
HS
2799 }
2800
98994639
HS
2801 return xbuf;
2802
2803 bad:
2804 Safefree(xbuf);
6696cfa7
KW
2805 if (s != input_string) {
2806 Safefree(s);
2807 }
98994639 2808 *xlen = 0;
7d4bcc4a
KW
2809
2810# ifdef DEBUGGING
2811
58eebef2 2812 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 2813 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 2814 }
7d4bcc4a
KW
2815
2816# endif
2817
98994639
HS
2818 return NULL;
2819}
2820
7d4bcc4a 2821# ifdef DEBUGGING
c7202dee 2822
4cbaac56 2823STATIC void
c7202dee
KW
2824S_print_collxfrm_input_and_return(pTHX_
2825 const char * const s,
2826 const char * const e,
2827 const STRLEN * const xlen,
2828 const bool is_utf8)
2829{
c7202dee
KW
2830
2831 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
2832
511e4ff7
DM
2833 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
2834 (UV)PL_collation_ix);
c7202dee 2835 if (xlen) {
08b6dc1d 2836 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
2837 }
2838 else {
2839 PerlIO_printf(Perl_debug_log, "NULL");
2840 }
2841 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
2842 PL_collation_name);
9c8a6dc2
KW
2843 print_bytes_for_locale(s, e, is_utf8);
2844
2845 PerlIO_printf(Perl_debug_log, "'\n");
2846}
2847
2848STATIC void
2849S_print_bytes_for_locale(pTHX_
2850 const char * const s,
2851 const char * const e,
2852 const bool is_utf8)
2853{
2854 const char * t = s;
2855 bool prev_was_printable = TRUE;
2856 bool first_time = TRUE;
2857
2858 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
2859
2860 while (t < e) {
2861 UV cp = (is_utf8)
2862 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
2863 : * (U8 *) t;
2864 if (isPRINT(cp)) {
2865 if (! prev_was_printable) {
2866 PerlIO_printf(Perl_debug_log, " ");
2867 }
2868 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
2869 prev_was_printable = TRUE;
2870 }
2871 else {
2872 if (! first_time) {
2873 PerlIO_printf(Perl_debug_log, " ");
2874 }
147e3846 2875 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
2876 prev_was_printable = FALSE;
2877 }
2878 t += (is_utf8) ? UTF8SKIP(t) : 1;
2879 first_time = FALSE;
2880 }
c7202dee
KW
2881}
2882
7d4bcc4a 2883# endif /* #ifdef DEBUGGING */
98994639 2884#endif /* USE_LOCALE_COLLATE */
58eebef2 2885
8ef6e574
KW
2886#ifdef USE_LOCALE
2887
c1284011
KW
2888bool
2889Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
2890{
2891 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
2892 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 2893 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
2894 * could give the wrong result. The result will very likely be correct for
2895 * languages that have commonly used non-ASCII characters, but for notably
2896 * English, it comes down to if the locale's name ends in something like
2897 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61 2898
8de4332b 2899 const char *save_input_locale = NULL;
7d74bb61
KW
2900 STRLEN final_pos;
2901
7d4bcc4a
KW
2902# ifdef LC_ALL
2903
7d74bb61 2904 assert(category != LC_ALL);
7d4bcc4a
KW
2905
2906# endif
7d74bb61
KW
2907
2908 /* First dispose of the trivial cases */
837ce802 2909 save_input_locale = do_setlocale_r(category, NULL);
7d74bb61 2910 if (! save_input_locale) {
69014004
KW
2911 DEBUG_L(PerlIO_printf(Perl_debug_log,
2912 "Could not find current locale for category %d\n",
2913 category));
7d74bb61
KW
2914 return FALSE; /* XXX maybe should croak */
2915 }
b07fffd1 2916 save_input_locale = stdize_locale(savepv(save_input_locale));
a39edc4c 2917 if (isNAME_C_OR_POSIX(save_input_locale)) {
69014004
KW
2918 DEBUG_L(PerlIO_printf(Perl_debug_log,
2919 "Current locale for category %d is %s\n",
2920 category, save_input_locale));
b07fffd1 2921 Safefree(save_input_locale);
7d74bb61
KW
2922 return FALSE;
2923 }
2924
7d4bcc4a 2925# if defined(USE_LOCALE_CTYPE) \
1d958db2 2926 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 2927
1d958db2 2928 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
2929
2930 char *save_ctype_locale = NULL;
119ee68b 2931 bool is_utf8;
7d74bb61 2932
119ee68b 2933 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
2934
2935 /* Get the current LC_CTYPE locale */
837ce802 2936 save_ctype_locale = do_setlocale_c(LC_CTYPE, NULL);
7d74bb61 2937 if (! save_ctype_locale) {
69014004
KW
2938 DEBUG_L(PerlIO_printf(Perl_debug_log,
2939 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
2940 goto cant_use_nllanginfo;
2941 }
4f72bb37 2942 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
2943
2944 /* If LC_CTYPE and the desired category use the same locale, this
2945 * means that finding the value for LC_CTYPE is the same as finding
2946 * the value for the desired category. Otherwise, switch LC_CTYPE
2947 * to the desired category's locale */
2948 if (strEQ(save_ctype_locale, save_input_locale)) {
2949 Safefree(save_ctype_locale);
2950 save_ctype_locale = NULL;
2951 }
837ce802 2952 else if (! do_setlocale_c(LC_CTYPE, save_input_locale)) {
69014004
KW
2953 DEBUG_L(PerlIO_printf(Perl_debug_log,
2954 "Could not change LC_CTYPE locale to %s\n",
2955 save_input_locale));
7d74bb61
KW
2956 Safefree(save_ctype_locale);
2957 goto cant_use_nllanginfo;
2958 }
2959 }
2960
69014004
KW
2961 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
2962 save_input_locale));
2963
7d74bb61 2964 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
2965 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
2966 * should give the correct results */
119ee68b 2967
7d4bcc4a
KW
2968# if defined(HAS_NL_LANGINFO) && defined(CODESET)
2969
291a84fb 2970 { /* The task is easiest if the platform has this POSIX 2001 function */
c70a3e68
KW
2971 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
2972 /* FALSE => already in dest locale */
119ee68b 2973
c70a3e68
KW
2974 DEBUG_L(PerlIO_printf(Perl_debug_log,
2975 "\tnllanginfo returned CODESET '%s'\n", codeset));
2976
2977 if (codeset && strNE(codeset, "")) {
1d958db2
KW
2978 /* If we switched LC_CTYPE, switch back */
2979 if (save_ctype_locale) {
837ce802 2980 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2
KW
2981 Safefree(save_ctype_locale);
2982 }
2983
abc1d81d
KW
2984 is_utf8 = ( ( strlen(codeset) == STRLENs("UTF-8")
2985 && foldEQ(codeset, STR_WITH_LEN("UTF-8")))
2986 || ( strlen(codeset) == STRLENs("UTF8")
2987 && foldEQ(codeset, STR_WITH_LEN("UTF8"))));
1d958db2 2988
69014004
KW
2989 DEBUG_L(PerlIO_printf(Perl_debug_log,
2990 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
2991 codeset, is_utf8));
1d958db2
KW
2992 Safefree(save_input_locale);
2993 return is_utf8;
2994 }
119ee68b
KW
2995 }
2996
7d4bcc4a
KW
2997# endif
2998# ifdef MB_CUR_MAX
1d958db2
KW
2999
3000 /* Here, either we don't have nl_langinfo, or it didn't return a
3001 * codeset. Try MB_CUR_MAX */
3002
119ee68b
KW
3003 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3004 * Unicode code point. Since UTF-8 is the only non-single byte
3005 * encoding we handle, we just say any such encoding is UTF-8, and if
3006 * turns out to be wrong, other things will fail */
5f7616bd 3007 is_utf8 = (unsigned) MB_CUR_MAX >= STRLENs(MAX_UNICODE_UTF8);
119ee68b 3008
69014004
KW
3009 DEBUG_L(PerlIO_printf(Perl_debug_log,
3010 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
3011 (int) MB_CUR_MAX, is_utf8));
3012
119ee68b
KW
3013 Safefree(save_input_locale);
3014
7d4bcc4a 3015# ifdef HAS_MBTOWC
119ee68b
KW
3016
3017 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
3018 * since they are both in the C99 standard. We can feed a known byte
3019 * string to the latter function, and check that it gives the expected
3020 * result */
3021 if (is_utf8) {
3022 wchar_t wc;
51fc4b19
KW
3023 int len;
3024
856b881c 3025 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 3026 errno = 0;
b1d4925c 3027 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
51fc4b19 3028
b1d4925c
KW
3029
3030 if ( len != STRLENs(REPLACEMENT_CHARACTER_UTF8)
5579633c 3031 || wc != (wchar_t) UNICODE_REPLACEMENT)
119ee68b
KW
3032 {
3033 is_utf8 = FALSE;
b1d4925c
KW
3034 DEBUG_L(PerlIO_printf(Perl_debug_log, "\replacement=U+%x\n",
3035 (unsigned int)wc));
69014004
KW
3036 DEBUG_L(PerlIO_printf(Perl_debug_log,
3037 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
51fc4b19 3038 len, errno));
119ee68b
KW
3039 }
3040 }
7d4bcc4a
KW
3041
3042# endif
119ee68b 3043
1d958db2
KW
3044 /* If we switched LC_CTYPE, switch back */
3045 if (save_ctype_locale) {
837ce802 3046 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2 3047 Safefree(save_ctype_locale);
119ee68b 3048 }
7d74bb61 3049
1d958db2 3050 return is_utf8;
7d4bcc4a
KW
3051
3052# endif
3053
7d74bb61 3054 }
119ee68b 3055
7d74bb61
KW
3056 cant_use_nllanginfo:
3057
7d4bcc4a 3058# else /* nl_langinfo should work if available, so don't bother compiling this
0080c90a
KW
3059 fallback code. The final fallback of looking at the name is
3060 compiled, and will be executed if nl_langinfo fails */
7d74bb61 3061
97f4de96
KW
3062 /* nl_langinfo not available or failed somehow. Next try looking at the
3063 * currency symbol to see if it disambiguates things. Often that will be
3064 * in the native script, and if the symbol isn't in UTF-8, we know that the
3065 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
3066 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
3067 * */
fa9b773e 3068
7d4bcc4a
KW
3069# ifdef HAS_LOCALECONV
3070# ifdef USE_LOCALE_MONETARY
3071
fa9b773e
KW
3072 {
3073 char *save_monetary_locale = NULL;
fa9b773e 3074 bool only_ascii = FALSE;
13542a67
KW
3075 bool is_utf8 = FALSE;
3076 struct lconv* lc;
fa9b773e 3077
97f4de96
KW
3078 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
3079 * the desired category, if it isn't that locale already */
3080
fa9b773e
KW
3081 if (category != LC_MONETARY) {
3082
837ce802 3083 save_monetary_locale = do_setlocale_c(LC_MONETARY, NULL);
fa9b773e 3084 if (! save_monetary_locale) {
69014004
KW
3085 DEBUG_L(PerlIO_printf(Perl_debug_log,
3086 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
3087 goto cant_use_monetary;
3088 }
4f72bb37 3089 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 3090
13542a67
KW
3091 if (strEQ(save_monetary_locale, save_input_locale)) {
3092 Safefree(save_monetary_locale);
3093 save_monetary_locale = NULL;
3094 }
837ce802 3095 else if (! do_setlocale_c(LC_MONETARY, save_input_locale)) {
59c234b4
KW
3096 DEBUG_L(PerlIO_printf(Perl_debug_log,
3097 "Could not change LC_MONETARY locale to %s\n",
3098 save_input_locale));
3099 Safefree(save_monetary_locale);
3100 goto cant_use_monetary;
fa9b773e
KW
3101 }
3102 }
3103
3104 /* Here the current LC_MONETARY is set to the locale of the category
3105 * whose information is desired. */
3106
13542a67
KW
3107 lc = localeconv();
3108 if (! lc
3109 || ! lc->currency_symbol
c5f058df 3110 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
3111 {
3112 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));
3113 only_ascii = TRUE;
3114 }
3115 else {
3116 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
3117 }
3118
3119 /* If we changed it, restore LC_MONETARY to its original locale */
3120 if (save_monetary_locale) {
837ce802 3121 do_setlocale_c(LC_MONETARY, save_monetary_locale);
fa9b773e
KW
3122 Safefree(save_monetary_locale);
3123 }
3124
13542a67 3125 if (! only_ascii) {
fa9b773e 3126
59c234b4
KW
3127 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3128 * otherwise assume the locale is UTF-8 if and only if the symbol
3129 * is non-ascii UTF-8. */
3130 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
3131 save_input_locale, is_utf8));
3132 Safefree(save_input_locale);
3133 return is_utf8;
13542a67 3134 }
fa9b773e
KW
3135 }
3136 cant_use_monetary:
3137
7d4bcc4a
KW
3138# endif /* USE_LOCALE_MONETARY */
3139# endif /* HAS_LOCALECONV */
fa9b773e 3140
7d4bcc4a 3141# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e
KW
3142
3143/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3144 * the names of the months and weekdays, timezone, and am/pm indicator */
3145 {
3146 char *save_time_locale = NULL;
3147 int hour = 10;
3148 bool is_dst = FALSE;
3149 int dom = 1;
3150 int month = 0;
3151 int i;
3152 char * formatted_time;
3153
3154
3155 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
3156 * desired category, if it isn't that locale already */
3157
3158 if (category != LC_TIME) {
3159
837ce802 3160 save_time_locale = do_setlocale_c(LC_TIME, NULL);
15f7e74e
KW
3161 if (! save_time_locale) {
3162 DEBUG_L(PerlIO_printf(Perl_debug_log,
3163 "Could not find current locale for LC_TIME\n"));
3164 goto cant_use_time;
3165 }
3166 save_time_locale = stdize_locale(savepv(save_time_locale));
3167
3168 if (strEQ(save_time_locale, save_input_locale)) {
3169 Safefree(save_time_locale);
3170 save_time_locale = NULL;
3171 }
837ce802 3172 else if (! do_setlocale_c(LC_TIME, save_input_locale)) {
15f7e74e
KW
3173 DEBUG_L(PerlIO_printf(Perl_debug_log,
3174 "Could not change LC_TIME locale to %s\n",
3175 save_input_locale));
3176 Safefree(save_time_locale);
3177 goto cant_use_time;
3178 }
3179 }
3180
3181 /* Here the current LC_TIME is set to the locale of the category
3182 * whose information is desired. Look at all the days of the week and
9f10db87 3183 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
3184 * characters. The first such a one found will tell us if the locale
3185 * is UTF-8 or not */
3186
3187 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3188 formatted_time = my_strftime("%A %B %Z %p",
3ae5cd07 3189 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
c5f058df
KW
3190 if ( ! formatted_time
3191 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3192 {
15f7e74e
KW
3193
3194 /* Here, we didn't find a non-ASCII. Try the next time through
3195 * with the complemented dst and am/pm, and try with the next
3196 * weekday. After we have gotten all weekdays, try the next
3197 * month */
3198 is_dst = ! is_dst;
3199 hour = (hour + 12) % 24;
3200 dom++;
3201 if (i > 6) {
3202 month++;
3203 }
3204 continue;
3205 }
3206
3207 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3208 * false otherwise. But first, restore LC_TIME to its original
3209 * locale if we changed it */
3210 if (save_time_locale) {
837ce802 3211 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3212 Safefree(save_time_locale);
3213 }
3214
3215 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3216 save_input_locale,
3217 is_utf8_string((U8 *) formatted_time, 0)));
3218 Safefree(save_input_locale);
3219 return is_utf8_string((U8 *) formatted_time, 0);
3220 }
3221
3222 /* Falling off the end of the loop indicates all the names were just
3223 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3224 * to its original locale */
3225 if (save_time_locale) {
837ce802 3226 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3227 Safefree(save_time_locale);
3228 }
3229 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));
3230 }
3231 cant_use_time:
3232
7d4bcc4a 3233# endif
15f7e74e 3234
7d4bcc4a 3235# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
3236
3237/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
3238 * on our dromedary test machine, which has over 700 locales. There, this
3239 * added no value to looking at the currency symbol and the time strings. I
3240 * left it in so as to avoid rewriting it if real-world experience indicates
3241 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
3242 * haven't found illegal utf8, we continue on and examine all the strerror()
3243 * messages on the platform for utf8ness. If all are ASCII, we still don't
3244 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
3245 * utf8ness. The reason this doesn't help much is that the messages may not
3246 * have been translated into the locale. The currency symbol and time strings
3247 * are much more likely to have been translated. */
3248 {
855aeb93 3249 int e;
5857e934
KW
3250 bool is_utf8 = FALSE;
3251 bool non_ascii = FALSE;
855aeb93 3252 char *save_messages_locale = NULL;
5857e934 3253 const char * errmsg = NULL;
855aeb93 3254
5857e934
KW
3255 /* Like above, we set LC_MESSAGES to the locale of the desired
3256 * category, if it isn't that locale already */
855aeb93
JH
3257
3258 if (category != LC_MESSAGES) {
3259
837ce802 3260 save_messages_locale = do_setlocale_c(LC_MESSAGES, NULL);
855aeb93 3261 if (! save_messages_locale) {
5857e934
KW
3262 DEBUG_L(PerlIO_printf(Perl_debug_log,
3263 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
3264 goto cant_use_messages;
3265 }
5857e934 3266 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
3267
3268 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
3269 Safefree(save_messages_locale);
3270 save_messages_locale = NULL;
855aeb93 3271 }
837ce802 3272 else if (! do_setlocale_c(LC_MESSAGES, save_input_locale)) {
5857e934
KW
3273 DEBUG_L(PerlIO_printf(Perl_debug_log,
3274 "Could not change LC_MESSAGES locale to %s\n",
3275 save_input_locale));
855aeb93
JH
3276 Safefree(save_messages_locale);
3277 goto cant_use_messages;
3278 }
3279 }
3280
3281 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
3282 * whose information is desired. Look through all the messages. We
3283 * can't use Strerror() here because it may expand to code that
3284 * segfaults in miniperl */
855aeb93 3285
5857e934
KW
3286 for (e = 0; e <= sys_nerr; e++) {
3287 errno = 0;
3288 errmsg = sys_errlist[e];
3289 if (errno || !errmsg) {
855aeb93
JH
3290 break;
3291 }
5857e934 3292 errmsg = savepv(errmsg);
c5f058df 3293 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
3294 non_ascii = TRUE;
3295 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3296 break;
855aeb93
JH
3297 }
3298 }
5857e934 3299 Safefree(errmsg);
855aeb93
JH
3300
3301 /* And, if we changed it, restore LC_MESSAGES to its original locale */
3302 if (save_messages_locale) {
837ce802 3303 do_setlocale_c(LC_MESSAGES, save_messages_locale);
855aeb93
JH
3304 Safefree(save_messages_locale);
3305 }
3306
5857e934
KW
3307 if (non_ascii) {
3308
3309 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3310 * any non-ascii means it is one; otherwise we assume it isn't */
3311 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3312 save_input_locale,
3313 is_utf8));
3314 Safefree(save_input_locale);
3315 return is_utf8;
3316 }
855aeb93 3317
5857e934 3318 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
3319 }
3320 cant_use_messages:
3321
7d4bcc4a
KW
3322# endif
3323# endif /* the code that is compiled when no nl_langinfo */
0080c90a 3324
7d4bcc4a 3325# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3326 UTF-8 locale */
7d4bcc4a 3327
97f4de96
KW
3328 /* As a last resort, look at the locale name to see if it matches
3329 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3330 * return of setlocale(), is actually defined to be opaque, so we can't
3331 * really rely on the absence of various substrings in the name to indicate
3332 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3333 * be a UTF-8 locale. Similarly for the other common names */
3334
3335 final_pos = strlen(save_input_locale) - 1;
3336 if (final_pos >= 3) {
8de4332b 3337 const char *name = save_input_locale;
97f4de96
KW
3338
3339 /* Find next 'U' or 'u' and look from there */
3340 while ((name += strcspn(name, "Uu") + 1)
3341 <= save_input_locale + final_pos - 2)
3342 {
c72493e0 3343 if ( isALPHA_FOLD_NE(*name, 't')
305b8651 3344 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
3345 {
3346 continue;
3347 }
3348 name += 2;
3349 if (*(name) == '-') {
3350 if ((name > save_input_locale + final_pos - 1)) {
3351 break;
3352 }
3353 name++;
3354 }
3355 if (*(name) == '8') {
97f4de96
KW
3356 DEBUG_L(PerlIO_printf(Perl_debug_log,
3357 "Locale %s ends with UTF-8 in name\n",
3358 save_input_locale));
00c54b9c 3359 Safefree(save_input_locale);
97f4de96
KW
3360 return TRUE;
3361 }
3362 }
3363 DEBUG_L(PerlIO_printf(Perl_debug_log,
3364 "Locale %s doesn't end with UTF-8 in name\n",
3365 save_input_locale));
3366 }
3367
7d4bcc4a
KW
3368# endif
3369# ifdef WIN32
3370
97f4de96 3371 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
8a0832a1 3372 if (memENDs(save_input_locale, final_pos, "65001")) {
97f4de96 3373 DEBUG_L(PerlIO_printf(Perl_debug_log,
8a0832a1 3374 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
97f4de96
KW
3375 save_input_locale));
3376 Safefree(save_input_locale);
3377 return TRUE;
3378 }
7d4bcc4a
KW
3379
3380# endif
97f4de96
KW
3381
3382 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3383 * since we are about to return FALSE anyway, there is no point in doing
3384 * this extra work */
7d4bcc4a
KW
3385
3386# if 0
97f4de96
KW
3387 if (instr(save_input_locale, "8859")) {
3388 DEBUG_L(PerlIO_printf(Perl_debug_log,
3389 "Locale %s has 8859 in name, not UTF-8 locale\n",
3390 save_input_locale));
3391 Safefree(save_input_locale);
3392 return FALSE;
3393 }
7d4bcc4a 3394# endif
97f4de96 3395
69014004
KW
3396 DEBUG_L(PerlIO_printf(Perl_debug_log,
3397 "Assuming locale %s is not a UTF-8 locale\n",
3398 save_input_locale));
fa9b773e 3399 Safefree(save_input_locale);
7d74bb61
KW
3400 return FALSE;
3401}
3402
8ef6e574 3403#endif
7d74bb61 3404
d6ded950
KW
3405
3406bool
3407Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3408{
1a4f13e1 3409 dVAR;
d6ded950
KW
3410 /* Internal function which returns if we are in the scope of a pragma that
3411 * enables the locale category 'category'. 'compiling' should indicate if
3412 * this is during the compilation phase (TRUE) or not (FALSE). */
3413
3414 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
3415
3416 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
3417 if (! categories || categories == &PL_sv_placeholder) {
3418 return FALSE;
3419 }
3420
3421 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
3422 * a valid unsigned */
3423 assert(category >= -1);
3424 return cBOOL(SvUV(categories) & (1U << (category + 1)));
3425}
3426
2c6ee1a7 3427char *
6ebbc862
KW
3428Perl_my_strerror(pTHX_ const int errnum)
3429{
3430 /* Returns a mortalized copy of the text of the error message associated
3431 * with 'errnum'. It uses the current locale's text unless the platform
3432 * doesn't have the LC_MESSAGES category or we are not being called from
3433 * within the scope of 'use locale'. In the former case, it uses whatever
3434 * strerror returns; in the latter case it uses the text from the C locale.
3435 *
3436 * The function just calls strerror(), but temporarily switches, if needed,
3437 * to the C locale */
3438
3439 char *errstr;
52770946 3440 dVAR;
6ebbc862 3441
52770946 3442#ifndef USE_LOCALE_MESSAGES
6ebbc862 3443
52770946
KW
3444 /* If platform doesn't have messages category, we don't do any switching to
3445 * the C locale; we just use whatever strerror() returns */
3446
3447 errstr = savepv(Strerror(errnum));
3448
3449#else /* Has locale messages */
3450
3451 const bool within_locale_scope = IN_LC(LC_MESSAGES);
2c6ee1a7 3452
7aaa36b1
KW
3453# if defined(HAS_POSIX_2008_LOCALE) && defined(HAS_STRERROR_L)
3454
43cb6651
KW
3455 /* This function is trivial if we don't have to worry about thread safety
3456 * and have strerror_l(), as it handles the switch of locales so we don't
3457 * have to deal with that. We don't have to worry about thread safety if
3458 * this is an unthreaded build, or if strerror_r() is also available. Both
3459 * it and strerror_l() are thread-safe. Plain strerror() isn't thread
3460 * safe. But on threaded builds when strerror_r() is available, the
3461 * apparent call to strerror() below is actually a macro that
3462 * behind-the-scenes calls strerror_r().
3463 */
3464
3465# if ! defined(USE_ITHREADS) || defined(HAS_STRERROR_R)
7aaa36b1
KW
3466
3467 if (within_locale_scope) {
4eb27fc5 3468 errstr = savepv(strerror(errnum));
7aaa36b1
KW
3469 }
3470 else {
4eb27fc5 3471 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
7aaa36b1
KW
3472 }
3473
43cb6651
KW
3474# else
3475
3476 /* Here we have strerror_l(), but not strerror_r() and we are on a
3477 * threaded-build. We use strerror_l() for everything, constructing a
3478 * locale to pass to it if necessary */
3479
3480 bool do_free = FALSE;
3481 locale_t locale_to_use;
3482
3483 if (within_locale_scope) {
3484 locale_to_use = uselocale((locale_t) 0);
3485 if (locale_to_use == LC_GLOBAL_LOCALE) {
3486 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
3487 do_free = TRUE;
3488 }
3489 }
3490 else { /* Use C locale if not within 'use locale' scope */
3491 locale_to_use = PL_C_locale_obj;
3492 }
3493
3494 errstr = savepv(strerror_l(errnum, locale_to_use));
3495
3496 if (do_free) {
3497 freelocale(locale_to_use);
3498 }
3499
3500# endif
3501# else /* Doesn't have strerror_l() */
7aaa36b1
KW
3502
3503# ifdef USE_POSIX_2008_LOCALE
3504
fcd0e682 3505 locale_t save_locale = NULL;
7aaa36b1
KW
3506
3507# else
3508
8de4332b 3509 const char * save_locale = NULL;
c9dda6da 3510 bool locale_is_C = FALSE;
2c6ee1a7 3511
6ebbc862
KW
3512 /* We have a critical section to prevent another thread from changing the
3513 * locale out from under us (or zapping the buffer returned from
3514 * setlocale() ) */
3515 LOCALE_LOCK;
3516
7aaa36b1 3517# endif
6ebbc862 3518
9c8a6dc2
KW
3519 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3520 "my_strerror called with errnum %d\n", errnum));
6ebbc862 3521 if (! within_locale_scope) {
c9dda6da 3522 errno = 0;
a0b53297 3523
f1d2176b 3524# ifdef USE_POSIX_2008_LOCALE /* Use the thread-safe locale functions */
6ebbc862 3525
9c8a6dc2
KW
3526 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3527 "Not within locale scope, about to call"
3528 " uselocale(0x%p)\n", PL_C_locale_obj));
6ebbc862 3529 save_locale = uselocale(PL_C_locale_obj);
c9dda6da
KW
3530 if (! save_locale) {
3531 DEBUG_L(PerlIO_printf(Perl_debug_log,
9c8a6dc2
KW
3532 "uselocale failed, errno=%d\n", errno));
3533 }
3534 else {
3535 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3536 "uselocale returned 0x%p\n", save_locale));
c9dda6da 3537 }
6ebbc862 3538
7aaa36b1 3539# else /* Not thread-safe build */
a0b53297 3540
837ce802 3541 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
c9dda6da
KW
3542 if (! save_locale) {
3543 DEBUG_L(PerlIO_printf(Perl_debug_log,
3544 "setlocale failed, errno=%d\n", errno));
3545 }
3546 else {
3547 locale_is_C = isNAME_C_OR_POSIX(save_locale);
2c6ee1a7 3548
c9dda6da
KW
3549 /* Switch to the C locale if not already in it */
3550 if (! locale_is_C) {
2c6ee1a7 3551
c9dda6da
KW
3552 /* The setlocale() just below likely will zap 'save_locale', so
3553 * create a copy. */
3554 save_locale = savepv(save_locale);
837ce802 3555 do_setlocale_c(LC_MESSAGES, "C");
c9dda6da 3556 }
6ebbc862 3557 }
2c6ee1a7 3558
7aaa36b1 3559# endif
2c6ee1a7 3560
6ebbc862 3561 } /* end of ! within_locale_scope */
9c8a6dc2
KW
3562 else {
3563 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
3564 __FILE__, __LINE__));
3565 }
a0b53297 3566
9c8a6dc2
KW
3567 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3568 "Any locale change has been done; about to call Strerror\n"));
52770946 3569 errstr = savepv(Strerror(errnum));
6ebbc862
KW
3570
3571 if (! within_locale_scope) {
c9dda6da 3572 errno = 0;
a0b53297 3573
f1d2176b 3574# ifdef USE_POSIX_2008_LOCALE
6ebbc862 3575
9c8a6dc2
KW
3576 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3577 "%s: %d: not within locale scope, restoring the locale\n",
3578 __FILE__, __LINE__));
c9dda6da
KW
3579 if (save_locale && ! uselocale(save_locale)) {
3580 DEBUG_L(PerlIO_printf(Perl_debug_log,
3581 "uselocale restore failed, errno=%d\n", errno));
3582 }
2c6ee1a7 3583 }
6ebbc862 3584
7aaa36b1 3585# else
6ebbc862 3586
c9dda6da 3587 if (save_locale && ! locale_is_C) {
837ce802 3588 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
c9dda6da
KW
3589 DEBUG_L(PerlIO_printf(Perl_debug_log,
3590 "setlocale restore failed, errno=%d\n", errno));
3591 }
6ebbc862
KW
3592 Safefree(save_locale);
3593 }
3594 }
3595
3596 LOCALE_UNLOCK;
3597
7aaa36b1
KW
3598# endif
3599# endif /* End of doesn't have strerror_l */
52770946 3600#endif /* End of does have locale messages */
6affbbf0
KW
3601
3602#ifdef DEBUGGING
3603
3604 if (DEBUG_Lv_TEST) {
3605 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
3606 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
3607 PerlIO_printf(Perl_debug_log, "'\n");
3608 }
3609
2c6ee1a7
KW
3610#endif
3611
52770946 3612 SAVEFREEPV(errstr);
6ebbc862 3613 return errstr;
2c6ee1a7
KW
3614}
3615
66610fdd 3616/*
747c467a 3617
747c467a
KW
3618=for apidoc sync_locale
3619
3620Changing the program's locale should be avoided by XS code. Nevertheless,
3621certain non-Perl libraries called from XS, such as C<Gtk> do so. When this
3622happens, Perl needs to be told that the locale has changed. Use this function
3623to do so, before returning to Perl.
3624
3625=cut
3626*/
3627
3628void
3629Perl_sync_locale(pTHX)
3630{
9f82ea3e 3631 char * newlocale;
747c467a
KW
3632
3633#ifdef USE_LOCALE_CTYPE
7d4bcc4a 3634
9f82ea3e
KW
3635 newlocale = do_setlocale_c(LC_CTYPE, NULL);
3636 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3637 "%s:%d: %s\n", __FILE__, __LINE__,
3638 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
3639 new_ctype(newlocale);
747c467a 3640
7d4bcc4a 3641#endif /* USE_LOCALE_CTYPE */
747c467a 3642#ifdef USE_LOCALE_COLLATE
7d4bcc4a 3643
9f82ea3e
KW
3644 newlocale = do_setlocale_c(LC_COLLATE, NULL);
3645 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3646 "%s:%d: %s\n", __FILE__, __LINE__,
3647 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
3648 new_collate(newlocale);
747c467a 3649
7d4bcc4a 3650#endif
747c467a 3651#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 3652
9f82ea3e
KW
3653 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
3654 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3655 "%s:%d: %s\n", __FILE__, __LINE__,
3656 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
3657 new_numeric(newlocale);
7d4bcc4a 3658
747c467a
KW
3659#endif /* USE_LOCALE_NUMERIC */
3660
3661}
3662
5d1187d1
KW
3663#if defined(DEBUGGING) && defined(USE_LOCALE)
3664
a4f00dcc
KW
3665STATIC char *
3666S_setlocale_debug_string(const int category, /* category number,
5d1187d1
KW
3667 like LC_ALL */
3668 const char* const locale, /* locale name */
3669
3670 /* return value from setlocale() when attempting to
3671 * set 'category' to 'locale' */
3672 const char* const retval)
3673{
3674 /* Returns a pointer to a NUL-terminated string in static storage with
3675 * added text about the info passed in. This is not thread safe and will
3676 * be overwritten by the next call, so this should be used just to
fa07b8e5 3677 * formulate a string to immediately print or savepv() on. */
5d1187d1 3678
398a990f
DM
3679 /* initialise to a non-null value to keep it out of BSS and so keep
3680 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
60b45a7d
KW
3681 static char ret[128] = "If you can read this, thank your buggy C"
3682 " library strlcpy(), and change your hints file"
3683 " to undef it";
7d4bcc4a 3684
e5f10d49 3685 my_strlcpy(ret, "setlocale(", sizeof(ret));
b09aaf40 3686 my_strlcat(ret, category_name(category), sizeof(ret));
fa07b8e5 3687 my_strlcat(ret, ", ", sizeof(ret));
5d1187d1
KW
3688
3689 if (locale) {
fa07b8e5
KW
3690 my_strlcat(ret, "\"", sizeof(ret));
3691 my_strlcat(ret, locale, sizeof(ret));
3692 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
3693 }
3694 else {
fa07b8e5 3695 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
3696 }
3697
fa07b8e5 3698 my_strlcat(ret, ") returned ", sizeof(ret));
5d1187d1
KW
3699
3700 if (retval) {
fa07b8e5
KW
3701 my_strlcat(ret, "\"", sizeof(ret));
3702 my_strlcat(ret, retval, sizeof(ret));
3703 my_strlcat(ret, "\"", sizeof(ret));
5d1187d1
KW
3704 }
3705 else {
fa07b8e5 3706 my_strlcat(ret, "NULL", sizeof(ret));
5d1187d1
KW
3707 }
3708
3709 assert(strlen(ret) < sizeof(ret));
3710
3711 return ret;
3712}
3713
3714#endif
747c467a
KW
3715
3716
3717/*
14d04a33 3718 * ex: set ts=8 sts=4 sw=4 et:
37442d52 3719 */