1 /* locale.c
2 *
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
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 /*
12 * A Elbereth Gilthoniel,
13 * silivren penna míriel
14 * o menel aglar elenath!
15 * Na-chaered palan-díriel
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"]
21 */
22
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
25 *
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
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
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'.
35 *
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
44 */
45
46 #include "EXTERN.h"
47 #define PERL_IN_LOCALE_C
48 #include "perl_langinfo.h"
49 #include "perl.h"
50
51 #include "reentr.h"
52
53 #ifdef I_WCHAR
54 # include <wchar.h>
55 #endif
56 #ifdef I_WCTYPE
57 # include <wctype.h>
58 #endif
59
60 /* If the environment says to, we can output debugging information during
61 * initialization. This is done before option parsing, and before any thread
62 * creation, so can be a file-level static */
63 #if ! defined(DEBUGGING)
64 # define debug_initialization 0
65 # define DEBUG_INITIALIZATION_set(v)
66 #else
67 static bool debug_initialization = FALSE;
68 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
69 #endif
70
71
72 /* Returns the Unix errno portion; ignoring any others. This is a macro here
73 * instead of putting it into perl.h, because unclear to khw what should be
74 * done generally. */
75 #define GET_ERRNO saved_errno
76
77 /* strlen() of a literal string constant. We might want this more general,
78 * but using it in just this file for now. A problem with more generality is
79 * the compiler warnings about comparing unlike signs */
80 #define STRLENs(s) (sizeof("" s "") - 1)
81
82 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
83 * return of setlocale(), then this is extremely likely to be the C or POSIX
84 * locale. However, the output of setlocale() is documented to be opaque, but
85 * the odds are extremely small that it would return these two strings for some
86 * other locale. Note that VMS in these two locales includes many non-ASCII
87 * characters as controls and punctuation (below are hex bytes):
88 * cntrl: 84-97 9B-9F
89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
90 * Oddly, none there are listed as alphas, though some represent alphabetics
91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
92 #define isNAME_C_OR_POSIX(name) \
93 ( (name) != NULL \
94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
95 || strEQ((name), "POSIX")))
96
97 #ifdef USE_LOCALE
98
99 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
100 * looked up. This is in the form of a C string: */
101
102 #define UTF8NESS_SEP "\v"
103 #define UTF8NESS_PREFIX "\f"
104
105 /* So, the string looks like:
106 *
107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
108 *
109 * where the digit 0 after the \a indicates that the locale starting just
110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
111
112 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
113 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
114
115 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
117
118 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
119 * kept there always. The remining portion of the cache is LRU, with the
120 * oldest looked-up locale at the tail end */
121
122 STATIC char *
S_stdize_locale(pTHX_ char * locs)123 S_stdize_locale(pTHX_ char *locs)
124 {
125 /* Standardize the locale name from a string returned by 'setlocale',
126 * possibly modifying that string.
127 *
128 * The typical return value of setlocale() is either
129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
131 * (the space-separated values represent the various sublocales,
132 * in some unspecified order). This is not handled by this function.
133 *
134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
135 * which is harmful for further use of the string in setlocale(). This
136 * function removes the trailing new line and everything up through the '='
137 * */
138
139 const char * const s = strchr(locs, '=');
140 bool okay = TRUE;
141
142 PERL_ARGS_ASSERT_STDIZE_LOCALE;
143
144 if (s) {
145 const char * const t = strchr(s, '.');
146 okay = FALSE;
147 if (t) {
148 const char * const u = strchr(t, '\n');
149 if (u && (u[1] == 0)) {
150 const STRLEN len = u - s;
151 Move(s + 1, locs, len, char);
152 locs[len] = 0;
153 okay = TRUE;
154 }
155 }
156 }
157
158 if (!okay)
159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
160
161 return locs;
162 }
163
164 /* Two parallel arrays; first the locale categories Perl uses on this system;
165 * the second array is their names. These arrays are in mostly arbitrary
166 * order. */
167
168 const int categories[] = {
169
170 # ifdef USE_LOCALE_NUMERIC
171 LC_NUMERIC,
172 # endif
173 # ifdef USE_LOCALE_CTYPE
174 LC_CTYPE,
175 # endif
176 # ifdef USE_LOCALE_COLLATE
177 LC_COLLATE,
178 # endif
179 # ifdef USE_LOCALE_TIME
180 LC_TIME,
181 # endif
182 # ifdef USE_LOCALE_MESSAGES
183 LC_MESSAGES,
184 # endif
185 # ifdef USE_LOCALE_MONETARY
186 LC_MONETARY,
187 # endif
188 # ifdef USE_LOCALE_ADDRESS
189 LC_ADDRESS,
190 # endif
191 # ifdef USE_LOCALE_IDENTIFICATION
192 LC_IDENTIFICATION,
193 # endif
194 # ifdef USE_LOCALE_MEASUREMENT
195 LC_MEASUREMENT,
196 # endif
197 # ifdef USE_LOCALE_PAPER
198 LC_PAPER,
199 # endif
200 # ifdef USE_LOCALE_TELEPHONE
201 LC_TELEPHONE,
202 # endif
203 # ifdef USE_LOCALE_SYNTAX
204 LC_SYNTAX,
205 # endif
206 # ifdef USE_LOCALE_TOD
207 LC_TOD,
208 # endif
209 # ifdef LC_ALL
210 LC_ALL,
211 # endif
212 -1 /* Placeholder because C doesn't allow a
213 trailing comma, and it would get complicated
214 with all the #ifdef's */
215 };
216
217 /* The top-most real element is LC_ALL */
218
219 const char * const category_names[] = {
220
221 # ifdef USE_LOCALE_NUMERIC
222 "LC_NUMERIC",
223 # endif
224 # ifdef USE_LOCALE_CTYPE
225 "LC_CTYPE",
226 # endif
227 # ifdef USE_LOCALE_COLLATE
228 "LC_COLLATE",
229 # endif
230 # ifdef USE_LOCALE_TIME
231 "LC_TIME",
232 # endif
233 # ifdef USE_LOCALE_MESSAGES
234 "LC_MESSAGES",
235 # endif
236 # ifdef USE_LOCALE_MONETARY
237 "LC_MONETARY",
238 # endif
239 # ifdef USE_LOCALE_ADDRESS
240 "LC_ADDRESS",
241 # endif
242 # ifdef USE_LOCALE_IDENTIFICATION
243 "LC_IDENTIFICATION",
244 # endif
245 # ifdef USE_LOCALE_MEASUREMENT
246 "LC_MEASUREMENT",
247 # endif
248 # ifdef USE_LOCALE_PAPER
249 "LC_PAPER",
250 # endif
251 # ifdef USE_LOCALE_TELEPHONE
252 "LC_TELEPHONE",
253 # endif
254 # ifdef USE_LOCALE_SYNTAX
255 "LC_SYNTAX",
256 # endif
257 # ifdef USE_LOCALE_TOD
258 "LC_TOD",
259 # endif
260 # ifdef LC_ALL
261 "LC_ALL",
262 # endif
263 NULL /* Placeholder */
264 };
265
266 # ifdef LC_ALL
267
268 /* On systems with LC_ALL, it is kept in the highest index position. (-2
269 * to account for the final unused placeholder element.) */
270 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
271
272 # else
273
274 /* On systems without LC_ALL, we pretend it is there, one beyond the real
275 * top element, hence in the unused placeholder element. */
276 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
277
278 # endif
279
280 /* Pretending there is an LC_ALL element just above allows us to avoid most
281 * special cases. Most loops through these arrays in the code below are
282 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
283 * on either type of system. But the code must be written to not access the
284 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
285 * checked for at compile time by using the #define LC_ALL_INDEX which is only
286 * defined if we do have LC_ALL. */
287
288 STATIC const char *
S_category_name(const int category)289 S_category_name(const int category)
290 {
291 unsigned int i;
292
293 #ifdef LC_ALL
294
295 if (category == LC_ALL) {
296 return "LC_ALL";
297 }
298
299 #endif
300
301 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
302 if (category == categories[i]) {
303 return category_names[i];
304 }
305 }
306
307 {
308 const char suffix[] = " (unknown)";
309 int temp = category;
310 Size_t length = sizeof(suffix) + 1;
311 char * unknown;
312 dTHX;
313
314 if (temp < 0) {
315 length++;
316 temp = - temp;
317 }
318
319 /* Calculate the number of digits */
320 while (temp >= 10) {
321 temp /= 10;
322 length++;
323 }
324
325 Newx(unknown, length, char);
326 my_snprintf(unknown, length, "%d%s", category, suffix);
327 SAVEFREEPV(unknown);
328 return unknown;
329 }
330 }
331
332 /* Now create LC_foo_INDEX #defines for just those categories on this system */
333 # ifdef USE_LOCALE_NUMERIC
334 # define LC_NUMERIC_INDEX 0
335 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
336 # else
337 # define _DUMMY_NUMERIC -1
338 # endif
339 # ifdef USE_LOCALE_CTYPE
340 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
341 # define _DUMMY_CTYPE LC_CTYPE_INDEX
342 # else
343 # define _DUMMY_CTYPE _DUMMY_NUMERIC
344 # endif
345 # ifdef USE_LOCALE_COLLATE
346 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
347 # define _DUMMY_COLLATE LC_COLLATE_INDEX
348 # else
349 # define _DUMMY_COLLATE _DUMMY_CTYPE
350 # endif
351 # ifdef USE_LOCALE_TIME
352 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
353 # define _DUMMY_TIME LC_TIME_INDEX
354 # else
355 # define _DUMMY_TIME _DUMMY_COLLATE
356 # endif
357 # ifdef USE_LOCALE_MESSAGES
358 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
359 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
360 # else
361 # define _DUMMY_MESSAGES _DUMMY_TIME
362 # endif
363 # ifdef USE_LOCALE_MONETARY
364 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
365 # define _DUMMY_MONETARY LC_MONETARY_INDEX
366 # else
367 # define _DUMMY_MONETARY _DUMMY_MESSAGES
368 # endif
369 # ifdef USE_LOCALE_ADDRESS
370 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
371 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
372 # else
373 # define _DUMMY_ADDRESS _DUMMY_MONETARY
374 # endif
375 # ifdef USE_LOCALE_IDENTIFICATION
376 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
377 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
378 # else
379 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
380 # endif
381 # ifdef USE_LOCALE_MEASUREMENT
382 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
383 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
384 # else
385 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
386 # endif
387 # ifdef USE_LOCALE_PAPER
388 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
389 # define _DUMMY_PAPER LC_PAPER_INDEX
390 # else
391 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
392 # endif
393 # ifdef USE_LOCALE_TELEPHONE
394 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
395 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
396 # else
397 # define _DUMMY_TELEPHONE _DUMMY_PAPER
398 # endif
399 # ifdef USE_LOCALE_SYNTAX
400 # define LC_SYNTAX_INDEX _DUMMY_TELEPHONE + 1
401 # define _DUMMY_SYNTAX LC_SYNTAX_INDEX
402 # else
403 # define _DUMMY_SYNTAX _DUMMY_TELEPHONE
404 # endif
405 # ifdef USE_LOCALE_TOD
406 # define LC_TOD_INDEX _DUMMY_SYNTAX + 1
407 # define _DUMMY_TOD LC_TOD_INDEX
408 # else
409 # define _DUMMY_TOD _DUMMY_SYNTAX
410 # endif
411 # ifdef LC_ALL
412 # define LC_ALL_INDEX _DUMMY_TOD + 1
413 # endif
414 #endif /* ifdef USE_LOCALE */
415
416 /* Windows requres a customized base-level setlocale() */
417 #ifdef WIN32
418 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
419 #else
420 # define my_setlocale(cat, locale) setlocale(cat, locale)
421 #endif
422
423 #ifndef USE_POSIX_2008_LOCALE
424
425 /* "do_setlocale_c" is intended to be called when the category is a constant
426 * known at compile time; "do_setlocale_r", not known until run time */
427 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
428 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
429 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
430
431 #else /* Below uses POSIX 2008 */
432
433 /* We emulate setlocale with our own function. LC_foo is not valid for the
434 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
435 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
436 * proper offset into the array 'category_masks[]'. At runtime, we have to
437 * search through the array (as the actual numbers may not be small contiguous
438 * positive integers which would lend themselves to array lookup). */
439 # define do_setlocale_c(cat, locale) \
440 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
441 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
442
443 # if ! defined(__GLIBC__) || ! defined(USE_LOCALE_MESSAGES)
444
445 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
446
447 # else /* Invalidate glibc cache of loaded translations, see [perl #134264] */
448
449 # include <libintl.h>
450 # define FIX_GLIBC_LC_MESSAGES_BUG(i) \
451 STMT_START { \
452 if ((i) == LC_MESSAGES_INDEX) { \
453 textdomain(textdomain(NULL)); \
454 } \
455 } STMT_END
456
457 # endif
458
459 /* A third array, parallel to the ones above to map from category to its
460 * equivalent mask */
461 const int category_masks[] = {
462 # ifdef USE_LOCALE_NUMERIC
463 LC_NUMERIC_MASK,
464 # endif
465 # ifdef USE_LOCALE_CTYPE
466 LC_CTYPE_MASK,
467 # endif
468 # ifdef USE_LOCALE_COLLATE
469 LC_COLLATE_MASK,
470 # endif
471 # ifdef USE_LOCALE_TIME
472 LC_TIME_MASK,
473 # endif
474 # ifdef USE_LOCALE_MESSAGES
475 LC_MESSAGES_MASK,
476 # endif
477 # ifdef USE_LOCALE_MONETARY
478 LC_MONETARY_MASK,
479 # endif
480 # ifdef USE_LOCALE_ADDRESS
481 LC_ADDRESS_MASK,
482 # endif
483 # ifdef USE_LOCALE_IDENTIFICATION
484 LC_IDENTIFICATION_MASK,
485 # endif
486 # ifdef USE_LOCALE_MEASUREMENT
487 LC_MEASUREMENT_MASK,
488 # endif
489 # ifdef USE_LOCALE_PAPER
490 LC_PAPER_MASK,
491 # endif
492 # ifdef USE_LOCALE_TELEPHONE
493 LC_TELEPHONE_MASK,
494 # endif
495 # ifdef USE_LOCALE_SYNTAX
496 LC_SYNTAX_MASK,
497 # endif
498 # ifdef USE_LOCALE_TOD
499 LC_TOD_MASK,
500 # endif
501 /* LC_ALL can't be turned off by a Configure
502 * option, and in Posix 2008, should always be
503 * here, so compile it in unconditionally.
504 * This could catch some glitches at compile
505 * time */
506 LC_ALL_MASK
507 };
508
509 STATIC const char *
S_emulate_setlocale(const int category,const char * locale,unsigned int index,const bool is_index_valid)510 S_emulate_setlocale(const int category,
511 const char * locale,
512 unsigned int index,
513 const bool is_index_valid
514 )
515 {
516 /* This function effectively performs a setlocale() on just the current
517 * thread; thus it is thread-safe. It does this by using the POSIX 2008
518 * locale functions to emulate the behavior of setlocale(). Similar to
519 * regular setlocale(), the return from this function points to memory that
520 * can be overwritten by other system calls, so needs to be copied
521 * immediately if you need to retain it. The difference here is that
522 * system calls besides another setlocale() can overwrite it.
523 *
524 * By doing this, most locale-sensitive functions become thread-safe. The
525 * exceptions are mostly those that return a pointer to static memory.
526 *
527 * This function takes the same parameters, 'category' and 'locale', that
528 * the regular setlocale() function does, but it also takes two additional
529 * ones. This is because the 2008 functions don't use a category; instead
530 * they use a corresponding mask. Because this function operates in both
531 * worlds, it may need one or the other or both. This function can
532 * calculate the mask from the input category, but to avoid this
533 * calculation, if the caller knows at compile time what the mask is, it
534 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
535 * parameter is ignored.
536 *
537 * POSIX 2008, for some sick reason, chose not to provide a method to find
538 * the category name of a locale. Some vendors have created a
539 * querylocale() function to do just that. This function is a lot simpler
540 * to implement on systems that have this. Otherwise, we have to keep
541 * track of what the locale has been set to, so that we can return its
542 * name to emulate setlocale(). It's also possible for C code in some
543 * library to change the locale without us knowing it, though as of
544 * September 2017, there are no occurrences in CPAN of uselocale(). Some
545 * libraries do use setlocale(), but that changes the global locale, and
546 * threads using per-thread locales will just ignore those changes.
547 * Another problem is that without querylocale(), we have to guess at what
548 * was meant by setting a locale of "". We handle this by not actually
549 * ever setting to "" (unless querylocale exists), but to emulate what we
550 * think should happen for "".
551 */
552
553 int mask;
554 locale_t old_obj;
555 locale_t new_obj;
556 dTHX;
557
558 # ifdef DEBUGGING
559
560 if (DEBUG_Lv_TEST || debug_initialization) {
561 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
562 }
563
564 # endif
565
566 /* If the input mask might be incorrect, calculate the correct one */
567 if (! is_index_valid) {
568 unsigned int i;
569
570 # ifdef DEBUGGING
571
572 if (DEBUG_Lv_TEST || debug_initialization) {
573 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
574 }
575
576 # endif
577
578 for (i = 0; i <= LC_ALL_INDEX; i++) {
579 if (category == categories[i]) {
580 index = i;
581 goto found_index;
582 }
583 }
584
585 /* Here, we don't know about this category, so can't handle it.
586 * Fallback to the early POSIX usages */
587 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
588 "Unknown locale category %d; can't set it to %s\n",
589 category, locale);
590 return NULL;
591
592 found_index: ;
593
594 # ifdef DEBUGGING
595
596 if (DEBUG_Lv_TEST || debug_initialization) {
597 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
598 }
599
600 # endif
601
602 }
603
604 mask = category_masks[index];
605
606 # ifdef DEBUGGING
607
608 if (DEBUG_Lv_TEST || debug_initialization) {
609 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
610 }
611
612 # endif
613
614 /* If just querying what the existing locale is ... */
615 if (locale == NULL) {
616 locale_t cur_obj = uselocale((locale_t) 0);
617
618 # ifdef DEBUGGING
619
620 if (DEBUG_Lv_TEST || debug_initialization) {
621 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
622 }
623
624 # endif
625
626 if (cur_obj == LC_GLOBAL_LOCALE) {
627 return my_setlocale(category, NULL);
628 }
629
630 # ifdef HAS_QUERYLOCALE
631
632 return (char *) querylocale(mask, cur_obj);
633
634 # else
635
636 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
637 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
638
639 # if defined(_NL_LOCALE_NAME) \
640 && defined(DEBUGGING) \
641 /* On systems that accept any locale name, the real underlying \
642 * locale is often returned by this internal function, so we \
643 * can't use it */ \
644 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
645 {
646 /* Internal glibc for querylocale(), but doesn't handle
647 * empty-string ("") locale properly; who knows what other
648 * glitches. Check for it now, under debug. */
649
650 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
651 uselocale((locale_t) 0));
652 /*
653 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
654 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
655 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
656 */
657 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
658 if ( strNE(PL_curlocales[index], temp_name)
659 && ! ( isNAME_C_OR_POSIX(temp_name)
660 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
661
662 # ifdef USE_C_BACKTRACE
663
664 dump_c_backtrace(Perl_debug_log, 20, 1);
665
666 # endif
667
668 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
669 " (%s) and what internal glibc thinks"
670 " (%s)\n", category_names[index],
671 PL_curlocales[index], temp_name);
672 }
673
674 return temp_name;
675 }
676 }
677
678 # endif
679
680 /* Without querylocale(), we have to use our record-keeping we've
681 * done. */
682
683 if (category != LC_ALL) {
684
685 # ifdef DEBUGGING
686
687 if (DEBUG_Lv_TEST || debug_initialization) {
688 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
689 }
690
691 # endif
692
693 return PL_curlocales[index];
694 }
695 else { /* For LC_ALL */
696 unsigned int i;
697 Size_t names_len = 0;
698 char * all_string;
699 bool are_all_categories_the_same_locale = TRUE;
700
701 /* If we have a valid LC_ALL value, just return it */
702 if (PL_curlocales[LC_ALL_INDEX]) {
703
704 # ifdef DEBUGGING
705
706 if (DEBUG_Lv_TEST || debug_initialization) {
707 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
708 }
709
710 # endif
711
712 return PL_curlocales[LC_ALL_INDEX];
713 }
714
715 /* Otherwise, we need to construct a string of name=value pairs.
716 * We use the glibc syntax, like
717 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
718 * First calculate the needed size. Along the way, check if all
719 * the locale names are the same */
720 for (i = 0; i < LC_ALL_INDEX; i++) {
721
722 # ifdef DEBUGGING
723
724 if (DEBUG_Lv_TEST || debug_initialization) {
725 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
726 }
727
728 # endif
729
730 names_len += strlen(category_names[i])
731 + 1 /* '=' */
732 + strlen(PL_curlocales[i])
733 + 1; /* ';' */
734
735 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
736 are_all_categories_the_same_locale = FALSE;
737 }
738 }
739
740 /* If they are the same, we don't actually have to construct the
741 * string; we just make the entry in LC_ALL_INDEX valid, and be
742 * that single name */
743 if (are_all_categories_the_same_locale) {
744 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
745 return PL_curlocales[LC_ALL_INDEX];
746 }
747
748 names_len++; /* Trailing '\0' */
749 SAVEFREEPV(Newx(all_string, names_len, char));
750 *all_string = '\0';
751
752 /* Then fill in the string */
753 for (i = 0; i < LC_ALL_INDEX; i++) {
754
755 # ifdef DEBUGGING
756
757 if (DEBUG_Lv_TEST || debug_initialization) {
758 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
759 }
760
761 # endif
762
763 my_strlcat(all_string, category_names[i], names_len);
764 my_strlcat(all_string, "=", names_len);
765 my_strlcat(all_string, PL_curlocales[i], names_len);
766 my_strlcat(all_string, ";", names_len);
767 }
768
769 # ifdef DEBUGGING
770
771 if (DEBUG_L_TEST || debug_initialization) {
772 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
773 }
774
775 #endif
776
777 return all_string;
778 }
779
780 # ifdef EINVAL
781
782 SETERRNO(EINVAL, LIB_INVARG);
783
784 # endif
785
786 return NULL;
787
788 # endif
789
790 } /* End of this being setlocale(LC_foo, NULL) */
791
792 /* Here, we are switching locales. */
793
794 # ifndef HAS_QUERYLOCALE
795
796 if (strEQ(locale, "")) {
797
798 /* For non-querylocale() systems, we do the setting of "" ourselves to
799 * be sure that we really know what's going on. We follow the Linux
800 * documented behavior (but if that differs from the actual behavior,
801 * this won't work exactly as the OS implements). We go out and
802 * examine the environment based on our understanding of how the system
803 * works, and use that to figure things out */
804
805 const char * const lc_all = PerlEnv_getenv("LC_ALL");
806
807 /* Use any "LC_ALL" environment variable, as it overrides everything
808 * else. */
809 if (lc_all && strNE(lc_all, "")) {
810 locale = lc_all;
811 }
812 else {
813
814 /* Otherwise, we need to dig deeper. Unless overridden, the
815 * default is the LANG environment variable; if it doesn't exist,
816 * then "C" */
817
818 const char * default_name;
819
820 default_name = PerlEnv_getenv("LANG");
821
822 if (! default_name || strEQ(default_name, "")) {
823 default_name = "C";
824 }
825
826 if (category != LC_ALL) {
827 const char * const name = PerlEnv_getenv(category_names[index]);
828
829 /* Here we are setting a single category. Assume will have the
830 * default name */
831 locale = default_name;
832
833 /* But then look for an overriding environment variable */
834 if (name && strNE(name, "")) {
835 locale = name;
836 }
837 }
838 else {
839 bool did_override = FALSE;
840 unsigned int i;
841
842 /* Here, we are getting LC_ALL. Any categories that don't have
843 * a corresponding environment variable set should be set to
844 * LANG, or to "C" if there is no LANG. If no individual
845 * categories differ from this, we can just set LC_ALL. This
846 * is buggy on systems that have extra categories that we don't
847 * know about. If there is an environment variable that sets
848 * that category, we won't know to look for it, and so our use
849 * of LANG or "C" improperly overrides it. On the other hand,
850 * if we don't do what is done here, and there is no
851 * environment variable, the category's locale should be set to
852 * LANG or "C". So there is no good solution. khw thinks the
853 * best is to look at systems to see what categories they have,
854 * and include them, and then to assume that we know the
855 * complete set */
856
857 for (i = 0; i < LC_ALL_INDEX; i++) {
858 const char * const env_override
859 = PerlEnv_getenv(category_names[i]);
860 const char * this_locale = ( env_override
861 && strNE(env_override, ""))
862 ? env_override
863 : default_name;
864 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
865 {
866 return NULL;
867 }
868
869 if (strNE(this_locale, default_name)) {
870 did_override = TRUE;
871 }
872 }
873
874 /* If all the categories are the same, we can set LC_ALL to
875 * that */
876 if (! did_override) {
877 locale = default_name;
878 }
879 else {
880
881 /* Here, LC_ALL is no longer valid, as some individual
882 * categories don't match it. We call ourselves
883 * recursively, as that will execute the code that
884 * generates the proper locale string for this situation.
885 * We don't do the remainder of this function, as that is
886 * to update our records, and we've just done that for the
887 * individual categories in the loop above, and doing so
888 * would cause LC_ALL to be done as well */
889 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
890 }
891 }
892 }
893 } /* End of this being setlocale(LC_foo, "") */
894 else if (strchr(locale, ';')) {
895
896 /* LC_ALL may actually incude a conglomeration of various categories.
897 * Without querylocale, this code uses the glibc (as of this writing)
898 * syntax for representing that, but that is not a stable API, and
899 * other platforms do it differently, so we have to handle all cases
900 * ourselves */
901
902 unsigned int i;
903 const char * s = locale;
904 const char * e = locale + strlen(locale);
905 const char * p = s;
906 const char * category_end;
907 const char * name_start;
908 const char * name_end;
909
910 /* If the string that gives what to set doesn't include all categories,
911 * the omitted ones get set to "C". To get this behavior, first set
912 * all the individual categories to "C", and override the furnished
913 * ones below */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
915 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
916 return NULL;
917 }
918 }
919
920 while (s < e) {
921
922 /* Parse through the category */
923 while (isWORDCHAR(*p)) {
924 p++;
925 }
926 category_end = p;
927
928 if (*p++ != '=') {
929 Perl_croak(aTHX_
930 "panic: %s: %d: Unexpected character in locale name '%02X",
931 __FILE__, __LINE__, *(p-1));
932 }
933
934 /* Parse through the locale name */
935 name_start = p;
936 while (p < e && *p != ';') {
937 if (! isGRAPH(*p)) {
938 Perl_croak(aTHX_
939 "panic: %s: %d: Unexpected character in locale name '%02X",
940 __FILE__, __LINE__, *(p-1));
941 }
942 p++;
943 }
944 name_end = p;
945
946 /* Space past the semi-colon */
947 if (p < e) {
948 p++;
949 }
950
951 /* Find the index of the category name in our lists */
952 for (i = 0; i < LC_ALL_INDEX; i++) {
953 char * individ_locale;
954
955 /* Keep going if this isn't the index. The strnNE() avoids a
956 * Perl_form(), but would fail if ever a category name could be
957 * a substring of another one, like if there were a
958 * "LC_TIME_DATE" */
959 if strnNE(s, category_names[i], category_end - s) {
960 continue;
961 }
962
963 /* If this index is for the single category we're changing, we
964 * have found the locale to set it to. */
965 if (category == categories[i]) {
966 locale = Perl_form(aTHX_ "%.*s",
967 (int) (name_end - name_start),
968 name_start);
969 goto ready_to_set;
970 }
971
972 assert(category == LC_ALL);
973 individ_locale = Perl_form(aTHX_ "%.*s",
974 (int) (name_end - name_start), name_start);
975 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
976 {
977 return NULL;
978 }
979 }
980
981 s = p;
982 }
983
984 /* Here we have set all the individual categories by recursive calls.
985 * These collectively should have fixed up LC_ALL, so can just query
986 * what that now is */
987 assert(category == LC_ALL);
988
989 return do_setlocale_c(LC_ALL, NULL);
990 } /* End of this being setlocale(LC_ALL,
991 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
992
993 ready_to_set: ;
994
995 /* Here at the end of having to deal with the absence of querylocale().
996 * Some cases have already been fully handled by recursive calls to this
997 * function. But at this point, we haven't dealt with those, but are now
998 * prepared to, knowing what the locale name to set this category to is.
999 * This would have come for free if this system had had querylocale() */
1000
1001 # endif /* end of ! querylocale */
1002
1003 assert(PL_C_locale_obj);
1004
1005 /* Switching locales generally entails freeing the current one's space (at
1006 * the C library's discretion). We need to stop using that locale before
1007 * the switch. So switch to a known locale object that we don't otherwise
1008 * mess with. This returns the locale object in effect at the time of the
1009 * switch. */
1010 old_obj = uselocale(PL_C_locale_obj);
1011
1012 # ifdef DEBUGGING
1013
1014 if (DEBUG_Lv_TEST || debug_initialization) {
1015 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
1016 }
1017
1018 # endif
1019
1020 if (! old_obj) {
1021
1022 # ifdef DEBUGGING
1023
1024 if (DEBUG_L_TEST || debug_initialization) {
1025 dSAVE_ERRNO;
1026 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1027 RESTORE_ERRNO;
1028 }
1029
1030 # endif
1031
1032 return NULL;
1033 }
1034
1035 # ifdef DEBUGGING
1036
1037 if (DEBUG_Lv_TEST || debug_initialization) {
1038 PerlIO_printf(Perl_debug_log,
1039 "%s:%d: emulate_setlocale now using %p\n",
1040 __FILE__, __LINE__, PL_C_locale_obj);
1041 }
1042
1043 # endif
1044
1045 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1046 * in fact, we already have switched to it (in preparation for what
1047 * normally is to come). But since we're already there, continue to use
1048 * it instead of trying to create a new locale */
1049 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1050
1051 # ifdef DEBUGGING
1052
1053 if (DEBUG_Lv_TEST || debug_initialization) {
1054 PerlIO_printf(Perl_debug_log,
1055 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1056 }
1057
1058 # endif
1059
1060 new_obj = PL_C_locale_obj;
1061
1062 /* We already had switched to the C locale in preparation for freeing
1063 * 'old_obj' */
1064 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1065 freelocale(old_obj);
1066 }
1067 }
1068 else {
1069 /* If we weren't in a thread safe locale, set so that newlocale() below
1070 * which uses 'old_obj', uses an empty one. Same for our reserved C
1071 * object. The latter is defensive coding, so that, even if there is
1072 * some bug, we will never end up trying to modify either of these, as
1073 * if passed to newlocale(), they can be. */
1074 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1075 old_obj = (locale_t) 0;
1076 }
1077
1078 /* Ready to create a new locale by modification of the exising one */
1079 new_obj = newlocale(mask, locale, old_obj);
1080
1081 if (! new_obj) {
1082 dSAVE_ERRNO;
1083
1084 # ifdef DEBUGGING
1085
1086 if (DEBUG_L_TEST || debug_initialization) {
1087 PerlIO_printf(Perl_debug_log,
1088 "%s:%d: emulate_setlocale creating new object"
1089 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1090 }
1091
1092 # endif
1093
1094 if (! uselocale(old_obj)) {
1095
1096 # ifdef DEBUGGING
1097
1098 if (DEBUG_L_TEST || debug_initialization) {
1099 PerlIO_printf(Perl_debug_log,
1100 "%s:%d: switching back failed: %d\n",
1101 __FILE__, __LINE__, GET_ERRNO);
1102 }
1103
1104 # endif
1105
1106 }
1107 RESTORE_ERRNO;
1108 return NULL;
1109 }
1110
1111 # ifdef DEBUGGING
1112
1113 if (DEBUG_Lv_TEST || debug_initialization) {
1114 PerlIO_printf(Perl_debug_log,
1115 "%s:%d: emulate_setlocale created %p",
1116 __FILE__, __LINE__, new_obj);
1117 if (old_obj) {
1118 PerlIO_printf(Perl_debug_log,
1119 "; should have freed %p", old_obj);
1120 }
1121 PerlIO_printf(Perl_debug_log, "\n");
1122 }
1123
1124 # endif
1125
1126 /* And switch into it */
1127 if (! uselocale(new_obj)) {
1128 dSAVE_ERRNO;
1129
1130 # ifdef DEBUGGING
1131
1132 if (DEBUG_L_TEST || debug_initialization) {
1133 PerlIO_printf(Perl_debug_log,
1134 "%s:%d: emulate_setlocale switching to new object"
1135 " failed\n", __FILE__, __LINE__);
1136 }
1137
1138 # endif
1139
1140 if (! uselocale(old_obj)) {
1141
1142 # ifdef DEBUGGING
1143
1144 if (DEBUG_L_TEST || debug_initialization) {
1145 PerlIO_printf(Perl_debug_log,
1146 "%s:%d: switching back failed: %d\n",
1147 __FILE__, __LINE__, GET_ERRNO);
1148 }
1149
1150 # endif
1151
1152 }
1153 freelocale(new_obj);
1154 RESTORE_ERRNO;
1155 return NULL;
1156 }
1157 }
1158
1159 # ifdef DEBUGGING
1160
1161 if (DEBUG_Lv_TEST || debug_initialization) {
1162 PerlIO_printf(Perl_debug_log,
1163 "%s:%d: emulate_setlocale now using %p\n",
1164 __FILE__, __LINE__, new_obj);
1165 }
1166
1167 # endif
1168
1169 /* We are done, except for updating our records (if the system doesn't keep
1170 * them) and in the case of locale "", we don't actually know what the
1171 * locale that got switched to is, as it came from the environment. So
1172 * have to find it */
1173
1174 # ifdef HAS_QUERYLOCALE
1175
1176 if (strEQ(locale, "")) {
1177 locale = querylocale(mask, new_obj);
1178 }
1179
1180 # else
1181
1182 /* Here, 'locale' is the return value */
1183
1184 /* Without querylocale(), we have to update our records */
1185
1186 if (category == LC_ALL) {
1187 unsigned int i;
1188
1189 /* For LC_ALL, we change all individual categories to correspond */
1190 /* PL_curlocales is a parallel array, so has same
1191 * length as 'categories' */
1192 for (i = 0; i <= LC_ALL_INDEX; i++) {
1193 Safefree(PL_curlocales[i]);
1194 PL_curlocales[i] = savepv(locale);
1195 }
1196
1197 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
1198 }
1199 else {
1200
1201 /* For a single category, if it's not the same as the one in LC_ALL, we
1202 * nullify LC_ALL */
1203
1204 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1205 Safefree(PL_curlocales[LC_ALL_INDEX]);
1206 PL_curlocales[LC_ALL_INDEX] = NULL;
1207 }
1208
1209 /* Then update the category's record */
1210 Safefree(PL_curlocales[index]);
1211 PL_curlocales[index] = savepv(locale);
1212
1213 FIX_GLIBC_LC_MESSAGES_BUG(index);
1214 }
1215
1216 # endif
1217
1218 return locale;
1219 }
1220
1221 #endif /* USE_POSIX_2008_LOCALE */
1222
1223 #ifdef USE_LOCALE
1224
1225 STATIC void
S_set_numeric_radix(pTHX_ const bool use_locale)1226 S_set_numeric_radix(pTHX_ const bool use_locale)
1227 {
1228 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1229 * TRUE, use the radix character derived from the current locale */
1230
1231 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1232 || defined(HAS_NL_LANGINFO))
1233
1234 const char * radix = (use_locale)
1235 ? my_nl_langinfo(RADIXCHAR, FALSE)
1236 /* FALSE => already in dest locale */
1237 : ".";
1238
1239 sv_setpv(PL_numeric_radix_sv, radix);
1240
1241 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1242 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1243 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1244 SvCUR(PL_numeric_radix_sv))
1245 && _is_cur_LC_category_utf8(LC_NUMERIC))
1246 {
1247 SvUTF8_on(PL_numeric_radix_sv);
1248 }
1249
1250 # ifdef DEBUGGING
1251
1252 if (DEBUG_L_TEST || debug_initialization) {
1253 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1254 SvPVX(PL_numeric_radix_sv),
1255 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1256 }
1257
1258 # endif
1259 #else
1260
1261 PERL_UNUSED_ARG(use_locale);
1262
1263 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1264
1265 }
1266
1267 STATIC void
S_new_numeric(pTHX_ const char * newnum)1268 S_new_numeric(pTHX_ const char *newnum)
1269 {
1270
1271 #ifndef USE_LOCALE_NUMERIC
1272
1273 PERL_UNUSED_ARG(newnum);
1274
1275 #else
1276
1277 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1278 * core Perl this and that 'newnum' is the name of the new locale.
1279 * It installs this locale as the current underlying default.
1280 *
1281 * The default locale and the C locale can be toggled between by use of the
1282 * set_numeric_underlying() and set_numeric_standard() functions, which
1283 * should probably not be called directly, but only via macros like
1284 * SET_NUMERIC_STANDARD() in perl.h.
1285 *
1286 * The toggling is necessary mainly so that a non-dot radix decimal point
1287 * character can be output, while allowing internal calculations to use a
1288 * dot.
1289 *
1290 * This sets several interpreter-level variables:
1291 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1292 * PL_numeric_underlying A boolean indicating if the toggled state is such
1293 * that the current locale is the program's underlying
1294 * locale
1295 * PL_numeric_standard An int indicating if the toggled state is such
1296 * that the current locale is the C locale or
1297 * indistinguishable from the C locale. If non-zero, it
1298 * is in C; if > 1, it means it may not be toggled away
1299 * from C.
1300 * PL_numeric_underlying_is_standard A bool kept by this function
1301 * indicating that the underlying locale and the standard
1302 * C locale are indistinguishable for the purposes of
1303 * LC_NUMERIC. This happens when both of the above two
1304 * variables are true at the same time. (Toggling is a
1305 * no-op under these circumstances.) This variable is
1306 * used to avoid having to recalculate.
1307 */
1308
1309 char *save_newnum;
1310
1311 if (! newnum) {
1312 Safefree(PL_numeric_name);
1313 PL_numeric_name = NULL;
1314 PL_numeric_standard = TRUE;
1315 PL_numeric_underlying = TRUE;
1316 PL_numeric_underlying_is_standard = TRUE;
1317 return;
1318 }
1319
1320 save_newnum = stdize_locale(savepv(newnum));
1321 PL_numeric_underlying = TRUE;
1322 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1323
1324 #ifndef TS_W32_BROKEN_LOCALECONV
1325
1326 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1327 * them. But on broken Windows systems calling my_nl_langinfo() for
1328 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1329 * and just always change the locale if not C nor POSIX on those systems */
1330 if (! PL_numeric_standard) {
1331 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1332 FALSE /* Don't toggle locale */ ))
1333 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1334 }
1335
1336 #endif
1337
1338 /* Save the new name if it isn't the same as the previous one, if any */
1339 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1340 Safefree(PL_numeric_name);
1341 PL_numeric_name = save_newnum;
1342 }
1343 else {
1344 Safefree(save_newnum);
1345 }
1346
1347 PL_numeric_underlying_is_standard = PL_numeric_standard;
1348
1349 # ifdef HAS_POSIX_2008_LOCALE
1350
1351 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1352 PL_numeric_name,
1353 PL_underlying_numeric_obj);
1354
1355 #endif
1356
1357 if (DEBUG_L_TEST || debug_initialization) {
1358 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1359 }
1360
1361 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1362 * have to worry about the radix being a non-dot. (Core operations that
1363 * need the underlying locale change to it temporarily). */
1364 if (PL_numeric_standard) {
1365 set_numeric_radix(0);
1366 }
1367 else {
1368 set_numeric_standard();
1369 }
1370
1371 #endif /* USE_LOCALE_NUMERIC */
1372
1373 }
1374
1375 void
Perl_set_numeric_standard(pTHX)1376 Perl_set_numeric_standard(pTHX)
1377 {
1378
1379 #ifdef USE_LOCALE_NUMERIC
1380
1381 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1382 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1383 * macro avoids calling this routine if toggling isn't necessary according
1384 * to our records (which could be wrong if some XS code has changed the
1385 * locale behind our back) */
1386
1387 # ifdef DEBUGGING
1388
1389 if (DEBUG_L_TEST || debug_initialization) {
1390 PerlIO_printf(Perl_debug_log,
1391 "Setting LC_NUMERIC locale to standard C\n");
1392 }
1393
1394 # endif
1395
1396 do_setlocale_c(LC_NUMERIC, "C");
1397 PL_numeric_standard = TRUE;
1398 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1399 set_numeric_radix(0);
1400
1401 #endif /* USE_LOCALE_NUMERIC */
1402
1403 }
1404
1405 void
Perl_set_numeric_underlying(pTHX)1406 Perl_set_numeric_underlying(pTHX)
1407 {
1408
1409 #ifdef USE_LOCALE_NUMERIC
1410
1411 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1412 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1413 * instead of calling this directly. The macro avoids calling this routine
1414 * if toggling isn't necessary according to our records (which could be
1415 * wrong if some XS code has changed the locale behind our back) */
1416
1417 # ifdef DEBUGGING
1418
1419 if (DEBUG_L_TEST || debug_initialization) {
1420 PerlIO_printf(Perl_debug_log,
1421 "Setting LC_NUMERIC locale to %s\n",
1422 PL_numeric_name);
1423 }
1424
1425 # endif
1426
1427 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1428 PL_numeric_standard = PL_numeric_underlying_is_standard;
1429 PL_numeric_underlying = TRUE;
1430 set_numeric_radix(! PL_numeric_standard);
1431
1432 #endif /* USE_LOCALE_NUMERIC */
1433
1434 }
1435
1436 /*
1437 * Set up for a new ctype locale.
1438 */
1439 STATIC void
S_new_ctype(pTHX_ const char * newctype)1440 S_new_ctype(pTHX_ const char *newctype)
1441 {
1442
1443 #ifndef USE_LOCALE_CTYPE
1444
1445 PERL_UNUSED_ARG(newctype);
1446 PERL_UNUSED_CONTEXT;
1447
1448 #else
1449
1450 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1451 * core Perl this and that 'newctype' is the name of the new locale.
1452 *
1453 * This function sets up the folding arrays for all 256 bytes, assuming
1454 * that tofold() is tolc() since fold case is not a concept in POSIX,
1455 *
1456 * Any code changing the locale (outside this file) should use
1457 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1458 * this function should be called directly only from this file and from
1459 * POSIX::setlocale() */
1460
1461 unsigned int i;
1462
1463 /* Don't check for problems if we are suppressing the warnings */
1464 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1465 bool maybe_utf8_turkic = FALSE;
1466
1467 PERL_ARGS_ASSERT_NEW_CTYPE;
1468
1469 /* We will replace any bad locale warning with 1) nothing if the new one is
1470 * ok; or 2) a new warning for the bad new locale */
1471 if (PL_warn_locale) {
1472 SvREFCNT_dec_NN(PL_warn_locale);
1473 PL_warn_locale = NULL;
1474 }
1475
1476 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1477
1478 /* A UTF-8 locale gets standard rules. But note that code still has to
1479 * handle this specially because of the three problematic code points */
1480 if (PL_in_utf8_CTYPE_locale) {
1481 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1482
1483 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1484 * Turkic. Make sure these two are the only anomalies. (We don't use
1485 * towupper and towlower because they aren't in C89.) */
1486
1487 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1488
1489 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1490
1491 #else
1492
1493 if (toupper('i') == 'i' && tolower('I') == 'I') {
1494
1495 #endif
1496 check_for_problems = TRUE;
1497 maybe_utf8_turkic = TRUE;
1498 }
1499 }
1500
1501 /* We don't populate the other lists if a UTF-8 locale, but do check that
1502 * everything works as expected, unless checking turned off */
1503 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1504 /* Assume enough space for every character being bad. 4 spaces each
1505 * for the 94 printable characters that are output like "'x' "; and 5
1506 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1507 * NUL */
1508 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1509 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1510 to start */
1511 unsigned int bad_count = 0; /* Count of bad characters */
1512
1513 for (i = 0; i < 256; i++) {
1514 if (! PL_in_utf8_CTYPE_locale) {
1515 if (isupper(i))
1516 PL_fold_locale[i] = (U8) tolower(i);
1517 else if (islower(i))
1518 PL_fold_locale[i] = (U8) toupper(i);
1519 else
1520 PL_fold_locale[i] = (U8) i;
1521 }
1522
1523 /* If checking for locale problems, see if the native ASCII-range
1524 * printables plus \n and \t are in their expected categories in
1525 * the new locale. If not, this could mean big trouble, upending
1526 * Perl's and most programs' assumptions, like having a
1527 * metacharacter with special meaning become a \w. Fortunately,
1528 * it's very rare to find locales that aren't supersets of ASCII
1529 * nowadays. It isn't a problem for most controls to be changed
1530 * into something else; we check only \n and \t, though perhaps \r
1531 * could be an issue as well. */
1532 if ( check_for_problems
1533 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1534 {
1535 bool is_bad = FALSE;
1536 char name[4] = { '\0' };
1537
1538 /* Convert the name into a string */
1539 if (isGRAPH_A(i)) {
1540 name[0] = i;
1541 name[1] = '\0';
1542 }
1543 else if (i == '\n') {
1544 my_strlcpy(name, "\\n", sizeof(name));
1545 }
1546 else if (i == '\t') {
1547 my_strlcpy(name, "\\t", sizeof(name));
1548 }
1549 else {
1550 assert(i == ' ');
1551 my_strlcpy(name, "' '", sizeof(name));
1552 }
1553
1554 /* Check each possibe class */
1555 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1556 is_bad = TRUE;
1557 DEBUG_L(PerlIO_printf(Perl_debug_log,
1558 "isalnum('%s') unexpectedly is %d\n",
1559 name, cBOOL(isalnum(i))));
1560 }
1561 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1562 is_bad = TRUE;
1563 DEBUG_L(PerlIO_printf(Perl_debug_log,
1564 "isalpha('%s') unexpectedly is %d\n",
1565 name, cBOOL(isalpha(i))));
1566 }
1567 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1568 is_bad = TRUE;
1569 DEBUG_L(PerlIO_printf(Perl_debug_log,
1570 "isdigit('%s') unexpectedly is %d\n",
1571 name, cBOOL(isdigit(i))));
1572 }
1573 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1574 is_bad = TRUE;
1575 DEBUG_L(PerlIO_printf(Perl_debug_log,
1576 "isgraph('%s') unexpectedly is %d\n",
1577 name, cBOOL(isgraph(i))));
1578 }
1579 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1580 is_bad = TRUE;
1581 DEBUG_L(PerlIO_printf(Perl_debug_log,
1582 "islower('%s') unexpectedly is %d\n",
1583 name, cBOOL(islower(i))));
1584 }
1585 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1586 is_bad = TRUE;
1587 DEBUG_L(PerlIO_printf(Perl_debug_log,
1588 "isprint('%s') unexpectedly is %d\n",
1589 name, cBOOL(isprint(i))));
1590 }
1591 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1592 is_bad = TRUE;
1593 DEBUG_L(PerlIO_printf(Perl_debug_log,
1594 "ispunct('%s') unexpectedly is %d\n",
1595 name, cBOOL(ispunct(i))));
1596 }
1597 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1598 is_bad = TRUE;
1599 DEBUG_L(PerlIO_printf(Perl_debug_log,
1600 "isspace('%s') unexpectedly is %d\n",
1601 name, cBOOL(isspace(i))));
1602 }
1603 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1604 is_bad = TRUE;
1605 DEBUG_L(PerlIO_printf(Perl_debug_log,
1606 "isupper('%s') unexpectedly is %d\n",
1607 name, cBOOL(isupper(i))));
1608 }
1609 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1610 is_bad = TRUE;
1611 DEBUG_L(PerlIO_printf(Perl_debug_log,
1612 "isxdigit('%s') unexpectedly is %d\n",
1613 name, cBOOL(isxdigit(i))));
1614 }
1615 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1616 is_bad = TRUE;
1617 DEBUG_L(PerlIO_printf(Perl_debug_log,
1618 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1619 name, tolower(i), (int) toLOWER_A(i)));
1620 }
1621 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1622 is_bad = TRUE;
1623 DEBUG_L(PerlIO_printf(Perl_debug_log,
1624 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1625 name, toupper(i), (int) toUPPER_A(i)));
1626 }
1627 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1628 is_bad = TRUE;
1629 DEBUG_L(PerlIO_printf(Perl_debug_log,
1630 "'\\n' (=%02X) is not a control\n", (int) i));
1631 }
1632
1633 /* Add to the list; Separate multiple entries with a blank */
1634 if (is_bad) {
1635 if (bad_count) {
1636 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1637 }
1638 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1639 bad_count++;
1640 }
1641 }
1642 }
1643
1644 if (bad_count == 2 && maybe_utf8_turkic) {
1645 bad_count = 0;
1646 *bad_chars_list = '\0';
1647 PL_fold_locale['I'] = 'I';
1648 PL_fold_locale['i'] = 'i';
1649 PL_in_utf8_turkic_locale = TRUE;
1650 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1651 __FILE__, __LINE__, newctype));
1652 }
1653 else {
1654 PL_in_utf8_turkic_locale = FALSE;
1655 }
1656
1657 # ifdef MB_CUR_MAX
1658
1659 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1660 * this locale requires more than one byte, there are going to be
1661 * problems. */
1662 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1663 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1664 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1665
1666 if ( check_for_problems && MB_CUR_MAX > 1
1667 && ! PL_in_utf8_CTYPE_locale
1668
1669 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1670 * locale. Just assume that the implementation for them (plus
1671 * for POSIX) is correct and the > 1 value is spurious. (Since
1672 * these are specially handled to never be considered UTF-8
1673 * locales, as long as this is the only problem, everything
1674 * should work fine */
1675 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1676 {
1677 multi_byte_locale = TRUE;
1678 }
1679
1680 # endif
1681
1682 /* If we found problems and we want them output, do so */
1683 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1684 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1685 {
1686 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1687 PL_warn_locale = Perl_newSVpvf(aTHX_
1688 "Locale '%s' contains (at least) the following characters"
1689 " which have\nunexpected meanings: %s\nThe Perl program"
1690 " will use the expected meanings",
1691 newctype, bad_chars_list);
1692 }
1693 else {
1694 PL_warn_locale = Perl_newSVpvf(aTHX_
1695 "Locale '%s' may not work well.%s%s%s\n",
1696 newctype,
1697 (multi_byte_locale)
1698 ? " Some characters in it are not recognized by"
1699 " Perl."
1700 : "",
1701 (bad_count)
1702 ? "\nThe following characters (and maybe others)"
1703 " may not have the same meaning as the Perl"
1704 " program expects:\n"
1705 : "",
1706 (bad_count)
1707 ? bad_chars_list
1708 : ""
1709 );
1710 }
1711
1712 # ifdef HAS_NL_LANGINFO
1713
1714 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1715 /* parameter FALSE is a don't care here */
1716 my_nl_langinfo(CODESET, FALSE));
1717
1718 # endif
1719
1720 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1721
1722 /* If we are actually in the scope of the locale or are debugging,
1723 * output the message now. If not in that scope, we save the
1724 * message to be output at the first operation using this locale,
1725 * if that actually happens. Most programs don't use locales, so
1726 * they are immune to bad ones. */
1727 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1728
1729 /* The '0' below suppresses a bogus gcc compiler warning */
1730 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1731
1732 if (IN_LC(LC_CTYPE)) {
1733 SvREFCNT_dec_NN(PL_warn_locale);
1734 PL_warn_locale = NULL;
1735 }
1736 }
1737 }
1738 }
1739
1740 #endif /* USE_LOCALE_CTYPE */
1741
1742 }
1743
1744 void
1745 Perl__warn_problematic_locale()
1746 {
1747
1748 #ifdef USE_LOCALE_CTYPE
1749
1750 dTHX;
1751
1752 /* Internal-to-core function that outputs the message in PL_warn_locale,
1753 * and then NULLS it. Should be called only through the macro
1754 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1755
1756 if (PL_warn_locale) {
1757 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1758 SvPVX(PL_warn_locale),
1759 0 /* dummy to avoid compiler warning */ );
1760 SvREFCNT_dec_NN(PL_warn_locale);
1761 PL_warn_locale = NULL;
1762 }
1763
1764 #endif
1765
1766 }
1767
1768 STATIC void
1769 S_new_collate(pTHX_ const char *newcoll)
1770 {
1771
1772 #ifndef USE_LOCALE_COLLATE
1773
1774 PERL_UNUSED_ARG(newcoll);
1775 PERL_UNUSED_CONTEXT;
1776
1777 #else
1778
1779 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1780 * core Perl this and that 'newcoll' is the name of the new locale.
1781 *
1782 * The design of locale collation is that every locale change is given an
1783 * index 'PL_collation_ix'. The first time a string particpates in an
1784 * operation that requires collation while locale collation is active, it
1785 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1786 * magic includes the collation index, and the transformation of the string
1787 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1788 * instead of the string itself. If a string changes, the magic is
1789 * cleared. The next time the locale changes, the index is incremented,
1790 * and so we know during a comparison that the transformation is not
1791 * necessarily still valid, and so is recomputed. Note that if the locale
1792 * changes enough times, the index could wrap (a U32), and it is possible
1793 * that a transformation would improperly be considered valid, leading to
1794 * an unlikely bug */
1795
1796 if (! newcoll) {
1797 if (PL_collation_name) {
1798 ++PL_collation_ix;
1799 Safefree(PL_collation_name);
1800 PL_collation_name = NULL;
1801 }
1802 PL_collation_standard = TRUE;
1803 is_standard_collation:
1804 PL_collxfrm_base = 0;
1805 PL_collxfrm_mult = 2;
1806 PL_in_utf8_COLLATE_locale = FALSE;
1807 PL_strxfrm_NUL_replacement = '\0';
1808 PL_strxfrm_max_cp = 0;
1809 return;
1810 }
1811
1812 /* If this is not the same locale as currently, set the new one up */
1813 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1814 ++PL_collation_ix;
1815 Safefree(PL_collation_name);
1816 PL_collation_name = stdize_locale(savepv(newcoll));
1817 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1818 if (PL_collation_standard) {
1819 goto is_standard_collation;
1820 }
1821
1822 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1823 PL_strxfrm_NUL_replacement = '\0';
1824 PL_strxfrm_max_cp = 0;
1825
1826 /* A locale collation definition includes primary, secondary, tertiary,
1827 * etc. weights for each character. To sort, the primary weights are
1828 * used, and only if they compare equal, then the secondary weights are
1829 * used, and only if they compare equal, then the tertiary, etc.
1830 *
1831 * strxfrm() works by taking the input string, say ABC, and creating an
1832 * output transformed string consisting of first the primary weights,
1833 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1834 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1835 * may not have weights at every level. In our example, let's say B
1836 * doesn't have a tertiary weight, and A doesn't have a secondary
1837 * weight. The constructed string is then going to be
1838 * A¹B¹C¹ B²C² A³C³ ....
1839 * This has the desired effect that strcmp() will look at the secondary
1840 * or tertiary weights only if the strings compare equal at all higher
1841 * priority weights. The spaces shown here, like in
1842 * "A¹B¹C¹ A²B²C² "
1843 * are not just for readability. In the general case, these must
1844 * actually be bytes, which we will call here 'separator weights'; and
1845 * they must be smaller than any other weight value, but since these
1846 * are C strings, only the terminating one can be a NUL (some
1847 * implementations may include a non-NUL separator weight just before
1848 * the NUL). Implementations tend to reserve 01 for the separator
1849 * weights. They are needed so that a shorter string's secondary
1850 * weights won't be misconstrued as primary weights of a longer string,
1851 * etc. By making them smaller than any other weight, the shorter
1852 * string will sort first. (Actually, if all secondary weights are
1853 * smaller than all primary ones, there is no need for a separator
1854 * weight between those two levels, etc.)
1855 *
1856 * The length of the transformed string is roughly a linear function of
1857 * the input string. It's not exactly linear because some characters
1858 * don't have weights at all levels. When we call strxfrm() we have to
1859 * allocate some memory to hold the transformed string. The
1860 * calculations below try to find coefficients 'm' and 'b' for this
1861 * locale so that m*x + b equals how much space we need, given the size
1862 * of the input string in 'x'. If we calculate too small, we increase
1863 * the size as needed, and call strxfrm() again, but it is better to
1864 * get it right the first time to avoid wasted expensive string
1865 * transformations. */
1866
1867 {
1868 /* We use the string below to find how long the tranformation of it
1869 * is. Almost all locales are supersets of ASCII, or at least the
1870 * ASCII letters. We use all of them, half upper half lower,
1871 * because if we used fewer, we might hit just the ones that are
1872 * outliers in a particular locale. Most of the strings being
1873 * collated will contain a preponderance of letters, and even if
1874 * they are above-ASCII, they are likely to have the same number of
1875 * weight levels as the ASCII ones. It turns out that digits tend
1876 * to have fewer levels, and some punctuation has more, but those
1877 * are relatively sparse in text, and khw believes this gives a
1878 * reasonable result, but it could be changed if experience so
1879 * dictates. */
1880 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1881 char * x_longer; /* Transformed 'longer' */
1882 Size_t x_len_longer; /* Length of 'x_longer' */
1883
1884 char * x_shorter; /* We also transform a substring of 'longer' */
1885 Size_t x_len_shorter;
1886
1887 /* _mem_collxfrm() is used get the transformation (though here we
1888 * are interested only in its length). It is used because it has
1889 * the intelligence to handle all cases, but to work, it needs some
1890 * values of 'm' and 'b' to get it started. For the purposes of
1891 * this calculation we use a very conservative estimate of 'm' and
1892 * 'b'. This assumes a weight can be multiple bytes, enough to
1893 * hold any UV on the platform, and there are 5 levels, 4 weight
1894 * bytes, and a trailing NUL. */
1895 PL_collxfrm_base = 5;
1896 PL_collxfrm_mult = 5 * sizeof(UV);
1897
1898 /* Find out how long the transformation really is */
1899 x_longer = _mem_collxfrm(longer,
1900 sizeof(longer) - 1,
1901 &x_len_longer,
1902
1903 /* We avoid converting to UTF-8 in the
1904 * called function by telling it the
1905 * string is in UTF-8 if the locale is a
1906 * UTF-8 one. Since the string passed
1907 * here is invariant under UTF-8, we can
1908 * claim it's UTF-8 even though it isn't.
1909 * */
1910 PL_in_utf8_COLLATE_locale);
1911 Safefree(x_longer);
1912
1913 /* Find out how long the transformation of a substring of 'longer'
1914 * is. Together the lengths of these transformations are
1915 * sufficient to calculate 'm' and 'b'. The substring is all of
1916 * 'longer' except the first character. This minimizes the chances
1917 * of being swayed by outliers */
1918 x_shorter = _mem_collxfrm(longer + 1,
1919 sizeof(longer) - 2,
1920 &x_len_shorter,
1921 PL_in_utf8_COLLATE_locale);
1922 Safefree(x_shorter);
1923
1924 /* If the results are nonsensical for this simple test, the whole
1925 * locale definition is suspect. Mark it so that locale collation
1926 * is not active at all for it. XXX Should we warn? */
1927 if ( x_len_shorter == 0
1928 || x_len_longer == 0
1929 || x_len_shorter >= x_len_longer)
1930 {
1931 PL_collxfrm_mult = 0;
1932 PL_collxfrm_base = 0;
1933 }
1934 else {
1935 SSize_t base; /* Temporary */
1936
1937 /* We have both: m * strlen(longer) + b = x_len_longer
1938 * m * strlen(shorter) + b = x_len_shorter;
1939 * subtracting yields:
1940 * m * (strlen(longer) - strlen(shorter))
1941 * = x_len_longer - x_len_shorter
1942 * But we have set things up so that 'shorter' is 1 byte smaller
1943 * than 'longer'. Hence:
1944 * m = x_len_longer - x_len_shorter
1945 *
1946 * But if something went wrong, make sure the multiplier is at
1947 * least 1.
1948 */
1949 if (x_len_longer > x_len_shorter) {
1950 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1951 }
1952 else {
1953 PL_collxfrm_mult = 1;
1954 }
1955
1956 /* mx + b = len
1957 * so: b = len - mx
1958 * but in case something has gone wrong, make sure it is
1959 * non-negative */
1960 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1961 if (base < 0) {
1962 base = 0;
1963 }
1964
1965 /* Add 1 for the trailing NUL */
1966 PL_collxfrm_base = base + 1;
1967 }
1968
1969 # ifdef DEBUGGING
1970
1971 if (DEBUG_L_TEST || debug_initialization) {
1972 PerlIO_printf(Perl_debug_log,
1973 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1974 "x_len_longer=%zu,"
1975 " collate multipler=%zu, collate base=%zu\n",
1976 __FILE__, __LINE__,
1977 PL_in_utf8_COLLATE_locale,
1978 x_len_shorter, x_len_longer,
1979 PL_collxfrm_mult, PL_collxfrm_base);
1980 }
1981 # endif
1982
1983 }
1984 }
1985
1986 #endif /* USE_LOCALE_COLLATE */
1987
1988 }
1989
1990 #endif
1991
1992 #ifdef WIN32
1993
1994 #define USE_WSETLOCALE
1995
1996 #ifdef USE_WSETLOCALE
1997
1998 STATIC char *
1999 S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2000 wchar_t *wlocale;
2001 wchar_t *wresult;
2002 char *result;
2003
2004 if (locale) {
2005 int req_size =
2006 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2007
2008 if (!req_size) {
2009 errno = EINVAL;
2010 return NULL;
2011 }
2012
2013 Newx(wlocale, req_size, wchar_t);
2014 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2015 Safefree(wlocale);
2016 errno = EINVAL;
2017 return NULL;
2018 }
2019 }
2020 else {
2021 wlocale = NULL;
2022 }
2023 wresult = _wsetlocale(category, wlocale);
2024 Safefree(wlocale);
2025 if (wresult) {
2026 int req_size =
2027 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2028 Newx(result, req_size, char);
2029 SAVEFREEPV(result); /* is there something better we can do here? */
2030 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2031 result, req_size, NULL, NULL)) {
2032 errno = EINVAL;
2033 return NULL;
2034 }
2035 }
2036 else {
2037 result = NULL;
2038 }
2039
2040 return result;
2041 }
2042
2043 #endif
2044
2045 STATIC char *
2046 S_win32_setlocale(pTHX_ int category, const char* locale)
2047 {
2048 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2049 * difference between the two unless the input locale is "", which normally
2050 * means on Windows to get the machine default, which is set via the
2051 * computer's "Regional and Language Options" (or its current equivalent).
2052 * In POSIX, it instead means to find the locale from the user's
2053 * environment. This routine changes the Windows behavior to first look in
2054 * the environment, and, if anything is found, use that instead of going to
2055 * the machine default. If there is no environment override, the machine
2056 * default is used, by calling the real setlocale() with "".
2057 *
2058 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2059 * use the particular category's variable if set; otherwise to use the LANG
2060 * variable. */
2061
2062 bool override_LC_ALL = FALSE;
2063 char * result;
2064 unsigned int i;
2065
2066 if (locale && strEQ(locale, "")) {
2067
2068 # ifdef LC_ALL
2069
2070 locale = PerlEnv_getenv("LC_ALL");
2071 if (! locale) {
2072 if (category == LC_ALL) {
2073 override_LC_ALL = TRUE;
2074 }
2075 else {
2076
2077 # endif
2078
2079 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2080 if (category == categories[i]) {
2081 locale = PerlEnv_getenv(category_names[i]);
2082 goto found_locale;
2083 }
2084 }
2085
2086 locale = PerlEnv_getenv("LANG");
2087 if (! locale) {
2088 locale = "";
2089 }
2090
2091 found_locale: ;
2092
2093 # ifdef LC_ALL
2094
2095 }
2096 }
2097
2098 # endif
2099
2100 }
2101
2102 #ifdef USE_WSETLOCALE
2103 result = S_wrap_wsetlocale(aTHX_ category, locale);
2104 #else
2105 result = setlocale(category, locale);
2106 #endif
2107 DEBUG_L(STMT_START {
2108 dSAVE_ERRNO;
2109 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2110 setlocale_debug_string(category, locale, result));
2111 RESTORE_ERRNO;
2112 } STMT_END);
2113
2114 if (! override_LC_ALL) {
2115 return result;
2116 }
2117
2118 /* Here the input category was LC_ALL, and we have set it to what is in the
2119 * LANG variable or the system default if there is no LANG. But these have
2120 * lower priority than the other LC_foo variables, so override it for each
2121 * one that is set. (If they are set to "", it means to use the same thing
2122 * we just set LC_ALL to, so can skip) */
2123
2124 for (i = 0; i < LC_ALL_INDEX; i++) {
2125 result = PerlEnv_getenv(category_names[i]);
2126 if (result && strNE(result, "")) {
2127 #ifdef USE_WSETLOCALE
2128 S_wrap_wsetlocale(aTHX_ categories[i], result);
2129 #else
2130 setlocale(categories[i], result);
2131 #endif
2132 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2133 __FILE__, __LINE__,
2134 setlocale_debug_string(categories[i], result, "not captured")));
2135 }
2136 }
2137
2138 result = setlocale(LC_ALL, NULL);
2139 DEBUG_L(STMT_START {
2140 dSAVE_ERRNO;
2141 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2142 __FILE__, __LINE__,
2143 setlocale_debug_string(LC_ALL, NULL, result));
2144 RESTORE_ERRNO;
2145 } STMT_END);
2146
2147 return result;
2148 }
2149
2150 #endif
2151
2152 /*
2153 =for apidoc Perl_setlocale
2154
2155 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2156 taking the same parameters, and returning the same information, except that it
2157 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2158 instead return C<C> if the underlying locale has a non-dot decimal point
2159 character, or a non-empty thousands separator for displaying floating point
2160 numbers. This is because perl keeps that locale category such that it has a
2161 dot and empty separator, changing the locale briefly during the operations
2162 where the underlying one is required. C<Perl_setlocale> knows about this, and
2163 compensates; regular C<setlocale> doesn't.
2164
2165 Another reason it isn't completely a drop-in replacement is that it is
2166 declared to return S<C<const char *>>, whereas the system setlocale omits the
2167 C<const> (presumably because its API was specified long ago, and can't be
2168 updated; it is illegal to change the information C<setlocale> returns; doing
2169 so leads to segfaults.)
2170
2171 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2172 C<setlocale> can be completely ineffective on some platforms under some
2173 configurations.
2174
2175 C<Perl_setlocale> should not be used to change the locale except on systems
2176 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2177 the system C<setlocale()> is ineffective, returning the wrong information, and
2178 failing to actually change the locale. C<Perl_setlocale>, however works
2179 properly in all circumstances.
2180
2181 The return points to a per-thread static buffer, which is overwritten the next
2182 time C<Perl_setlocale> is called from the same thread.
2183
2184 =cut
2185
2186 */
2187
2188 const char *
2189 Perl_setlocale(const int category, const char * locale)
2190 {
2191 /* This wraps POSIX::setlocale() */
2192
2193 #ifndef USE_LOCALE
2194
2195 PERL_UNUSED_ARG(category);
2196 PERL_UNUSED_ARG(locale);
2197
2198 return "C";
2199
2200 #else
2201
2202 const char * retval;
2203 const char * newlocale;
2204 dSAVEDERRNO;
2205 dTHX;
2206 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2207
2208 #ifdef USE_LOCALE_NUMERIC
2209
2210 /* A NULL locale means only query what the current one is. We have the
2211 * LC_NUMERIC name saved, because we are normally switched into the C
2212 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2213 * the correct results. All other categories don't require special
2214 * handling */
2215 if (locale == NULL) {
2216 if (category == LC_NUMERIC) {
2217
2218 /* We don't have to copy this return value, as it is a per-thread
2219 * variable, and won't change until a future setlocale */
2220 return PL_numeric_name;
2221 }
2222
2223 # ifdef LC_ALL
2224
2225 else if (category == LC_ALL) {
2226 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2227 }
2228
2229 # endif
2230
2231 }
2232
2233 #endif
2234
2235 retval = save_to_buffer(do_setlocale_r(category, locale),
2236 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2237 SAVE_ERRNO;
2238
2239 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2240
2241 if (locale == NULL && category == LC_ALL) {
2242 RESTORE_LC_NUMERIC();
2243 }
2244
2245 #endif
2246
2247 DEBUG_L(PerlIO_printf(Perl_debug_log,
2248 "%s:%d: %s\n", __FILE__, __LINE__,
2249 setlocale_debug_string(category, locale, retval)));
2250
2251 RESTORE_ERRNO;
2252
2253 if (! retval) {
2254 return NULL;
2255 }
2256
2257 /* If locale == NULL, we are just querying the state */
2258 if (locale == NULL) {
2259 return retval;
2260 }
2261
2262 /* Now that have switched locales, we have to update our records to
2263 * correspond. */
2264
2265 switch (category) {
2266
2267 #ifdef USE_LOCALE_CTYPE
2268
2269 case LC_CTYPE:
2270 new_ctype(retval);
2271 break;
2272
2273 #endif
2274 #ifdef USE_LOCALE_COLLATE
2275
2276 case LC_COLLATE:
2277 new_collate(retval);
2278 break;
2279
2280 #endif
2281 #ifdef USE_LOCALE_NUMERIC
2282
2283 case LC_NUMERIC:
2284 new_numeric(retval);
2285 break;
2286
2287 #endif
2288 #ifdef LC_ALL
2289
2290 case LC_ALL:
2291
2292 /* LC_ALL updates all the things we care about. The values may not
2293 * be the same as 'retval', as the locale "" may have set things
2294 * individually */
2295
2296 # ifdef USE_LOCALE_CTYPE
2297
2298 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2299 new_ctype(newlocale);
2300 Safefree(newlocale);
2301
2302 # endif /* USE_LOCALE_CTYPE */
2303 # ifdef USE_LOCALE_COLLATE
2304
2305 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2306 new_collate(newlocale);
2307 Safefree(newlocale);
2308
2309 # endif
2310 # ifdef USE_LOCALE_NUMERIC
2311
2312 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2313 new_numeric(newlocale);
2314 Safefree(newlocale);
2315
2316 # endif /* USE_LOCALE_NUMERIC */
2317 #endif /* LC_ALL */
2318
2319 default:
2320 break;
2321 }
2322
2323 return retval;
2324
2325 #endif
2326
2327 }
2328
2329 PERL_STATIC_INLINE const char *
2330 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2331 {
2332 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2333 * growing it if necessary */
2334
2335 Size_t string_size;
2336
2337 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2338
2339 if (! string) {
2340 return NULL;
2341 }
2342
2343 string_size = strlen(string) + offset + 1;
2344
2345 if (*buf_size == 0) {
2346 Newx(*buf, string_size, char);
2347 *buf_size = string_size;
2348 }
2349 else if (string_size > *buf_size) {
2350 Renew(*buf, string_size, char);
2351 *buf_size = string_size;
2352 }
2353
2354 Copy(string, *buf + offset, string_size - offset, char);
2355 return *buf;
2356 }
2357
2358 /*
2359
2360 =for apidoc Perl_langinfo
2361
2362 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2363 taking the same C<item> parameter values, and returning the same information.
2364 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2365 of Perl's locale handling from your code, and can be used on systems that lack
2366 a native C<nl_langinfo>.
2367
2368 Expanding on these:
2369
2370 =over
2371
2372 =item *
2373
2374 The reason it isn't quite a drop-in replacement is actually an advantage. The
2375 only difference is that it returns S<C<const char *>>, whereas plain
2376 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2377 forbidden to write into the buffer. By declaring this C<const>, the compiler
2378 enforces this restriction, so if it is violated, you know at compilation time,
2379 rather than getting segfaults at runtime.
2380
2381 =item *
2382
2383 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2384 without you having to write extra code. The reason for the extra code would be
2385 because these are from the C<LC_NUMERIC> locale category, which is normally
2386 kept set by Perl so that the radix is a dot, and the separator is the empty
2387 string, no matter what the underlying locale is supposed to be, and so to get
2388 the expected results, you have to temporarily toggle into the underlying
2389 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2390 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2391 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2392 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2393 (decimal point) character to be a dot.)
2394
2395 =item *
2396
2397 The system function it replaces can have its static return buffer trashed,
2398 not only by a subsequent call to that function, but by a C<freelocale>,
2399 C<setlocale>, or other locale change. The returned buffer of this function is
2400 not changed until the next call to it, so the buffer is never in a trashed
2401 state.
2402
2403 =item *
2404
2405 Its return buffer is per-thread, so it also is never overwritten by a call to
2406 this function from another thread; unlike the function it replaces.
2407
2408 =item *
2409
2410 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2411 as Windows, hence makes your code more portable. Of the fifty-some possible
2412 items specified by the POSIX 2008 standard,
2413 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2414 only one is completely unimplemented, though on non-Windows platforms, another
2415 significant one is also not implemented). It uses various techniques to
2416 recover the other items, including calling C<L<localeconv(3)>>, and
2417 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2418 available. Later C<strftime()> versions have additional capabilities; C<""> is
2419 returned for those not available on your system.
2420
2421 It is important to note that when called with an item that is recovered by
2422 using C<localeconv>, the buffer from any previous explicit call to
2423 C<localeconv> will be overwritten. This means you must save that buffer's
2424 contents if you need to access them after a call to this function. (But note
2425 that you might not want to be using C<localeconv()> directly anyway, because of
2426 issues like the ones listed in the second item of this list (above) for
2427 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2428 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2429 unpack).
2430
2431 The details for those items which may deviate from what this emulation returns
2432 and what a native C<nl_langinfo()> would return are specified in
2433 L<I18N::Langinfo>.
2434
2435 =back
2436
2437 When using C<Perl_langinfo> on systems that don't have a native
2438 C<nl_langinfo()>, you must
2439
2440 #include "perl_langinfo.h"
2441
2442 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2443 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2444 C<langinfo.h> would try to import into the namespace for code that doesn't need
2445 it.)
2446
2447 The original impetus for C<Perl_langinfo()> was so that code that needs to
2448 find out the current currency symbol, floating point radix character, or digit
2449 grouping separator can use, on all systems, the simpler and more
2450 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2451 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2452 is better to use the methods given in L<perlcall> to call
2453 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2454
2455 =cut
2456
2457 */
2458
2459 const char *
2460 #ifdef HAS_NL_LANGINFO
2461 Perl_langinfo(const nl_item item)
2462 #else
2463 Perl_langinfo(const int item)
2464 #endif
2465 {
2466 return my_nl_langinfo(item, TRUE);
2467 }
2468
2469 STATIC const char *
2470 #ifdef HAS_NL_LANGINFO
2471 S_my_nl_langinfo(const nl_item item, bool toggle)
2472 #else
2473 S_my_nl_langinfo(const int item, bool toggle)
2474 #endif
2475 {
2476 dTHX;
2477 const char * retval;
2478
2479 #ifdef USE_LOCALE_NUMERIC
2480
2481 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2482 * two items, and only if not already there */
2483 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2484 || PL_numeric_underlying))
2485
2486 #endif /* No toggling needed if not using LC_NUMERIC */
2487
2488 toggle = FALSE;
2489
2490 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2491 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2492 || ! defined(HAS_POSIX_2008_LOCALE)
2493
2494 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2495 * for those items dependent on it. This must be copied to a buffer before
2496 * switching back, as some systems destroy the buffer when setlocale() is
2497 * called */
2498
2499 {
2500 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2501
2502 if (toggle) {
2503 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2504 }
2505
2506 /* Prevent interference from another thread executing this code
2507 * section. */
2508 NL_LANGINFO_LOCK;
2509
2510 /* Copy to a per-thread buffer, which is also one that won't be
2511 * destroyed by a subsequent setlocale(), such as the
2512 * RESTORE_LC_NUMERIC may do just below. */
2513 retval = save_to_buffer(nl_langinfo(item),
2514 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2515 NL_LANGINFO_UNLOCK;
2516
2517 if (toggle) {
2518 RESTORE_LC_NUMERIC();
2519 }
2520 }
2521
2522 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2523
2524 {
2525 bool do_free = FALSE;
2526 locale_t cur = uselocale((locale_t) 0);
2527
2528 if (cur == LC_GLOBAL_LOCALE) {
2529 cur = duplocale(LC_GLOBAL_LOCALE);
2530 do_free = TRUE;
2531 }
2532
2533 # ifdef USE_LOCALE_NUMERIC
2534
2535 if (toggle) {
2536 if (PL_underlying_numeric_obj) {
2537 cur = PL_underlying_numeric_obj;
2538 }
2539 else {
2540 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2541 do_free = TRUE;
2542 }
2543 }
2544
2545 # endif
2546
2547 /* We have to save it to a buffer, because the freelocale() just below
2548 * can invalidate the internal one */
2549 retval = save_to_buffer(nl_langinfo_l(item, cur),
2550 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2551
2552 if (do_free) {
2553 freelocale(cur);
2554 }
2555 }
2556
2557 # endif
2558
2559 if (strEQ(retval, "")) {
2560 if (item == YESSTR) {
2561 return "yes";
2562 }
2563 if (item == NOSTR) {
2564 return "no";
2565 }
2566 }
2567
2568 return retval;
2569
2570 #else /* Below, emulate nl_langinfo as best we can */
2571
2572 {
2573
2574 # ifdef HAS_LOCALECONV
2575
2576 const struct lconv* lc;
2577 const char * temp;
2578 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2579
2580 # ifdef TS_W32_BROKEN_LOCALECONV
2581
2582 const char * save_global;
2583 const char * save_thread;
2584 int needed_size;
2585 char * ptr;
2586 char * e;
2587 char * item_start;
2588
2589 # endif
2590 # endif
2591 # ifdef HAS_STRFTIME
2592
2593 struct tm tm;
2594 bool return_format = FALSE; /* Return the %format, not the value */
2595 const char * format;
2596
2597 # endif
2598
2599 /* We copy the results to a per-thread buffer, even if not
2600 * multi-threaded. This is in part to simplify this code, and partly
2601 * because we need a buffer anyway for strftime(), and partly because a
2602 * call of localeconv() could otherwise wipe out the buffer, and the
2603 * programmer would not be expecting this, as this is a nl_langinfo()
2604 * substitute after all, so s/he might be thinking their localeconv()
2605 * is safe until another localeconv() call. */
2606
2607 switch (item) {
2608 Size_t len;
2609
2610 /* This is unimplemented */
2611 case ERA: /* For use with strftime() %E modifier */
2612
2613 default:
2614 return "";
2615
2616 /* We use only an English set, since we don't know any more */
2617 case YESEXPR: return "^[+1yY]";
2618 case YESSTR: return "yes";
2619 case NOEXPR: return "^[-0nN]";
2620 case NOSTR: return "no";
2621
2622 case CODESET:
2623
2624 # ifndef WIN32
2625
2626 /* On non-windows, this is unimplemented, in part because of
2627 * inconsistencies between vendors. The Darwin native
2628 * nl_langinfo() implementation simply looks at everything past
2629 * any dot in the name, but that doesn't work for other
2630 * vendors. Many Linux locales that don't have UTF-8 in their
2631 * names really are UTF-8, for example; z/OS locales that do
2632 * have UTF-8 in their names, aren't really UTF-8 */
2633 return "";
2634
2635 # else
2636
2637 { /* But on Windows, the name does seem to be consistent, so
2638 use that. */
2639 const char * p;
2640 const char * first;
2641 Size_t offset = 0;
2642 const char * name = my_setlocale(LC_CTYPE, NULL);
2643
2644 if (isNAME_C_OR_POSIX(name)) {
2645 return "ANSI_X3.4-1968";
2646 }
2647
2648 /* Find the dot in the locale name */
2649 first = (const char *) strchr(name, '.');
2650 if (! first) {
2651 first = name;
2652 goto has_nondigit;
2653 }
2654
2655 /* Look at everything past the dot */
2656 first++;
2657 p = first;
2658
2659 while (*p) {
2660 if (! isDIGIT(*p)) {
2661 goto has_nondigit;
2662 }
2663
2664 p++;
2665 }
2666
2667 /* Here everything past the dot is a digit. Treat it as a
2668 * code page */
2669 retval = save_to_buffer("CP", &PL_langinfo_buf,
2670 &PL_langinfo_bufsize, 0);
2671 offset = STRLENs("CP");
2672
2673 has_nondigit:
2674
2675 retval = save_to_buffer(first, &PL_langinfo_buf,
2676 &PL_langinfo_bufsize, offset);
2677 }
2678
2679 break;
2680
2681 # endif
2682 # ifdef HAS_LOCALECONV
2683
2684 case CRNCYSTR:
2685
2686 /* We don't bother with localeconv_l() because any system that
2687 * has it is likely to also have nl_langinfo() */
2688
2689 LOCALECONV_LOCK; /* Prevent interference with other threads
2690 using localeconv() */
2691
2692 # ifdef TS_W32_BROKEN_LOCALECONV
2693
2694 /* This is a workaround for a Windows bug prior to VS 15.
2695 * What we do here is, while locked, switch to the global
2696 * locale so localeconv() works; then switch back just before
2697 * the unlock. This can screw things up if some thread is
2698 * already using the global locale while assuming no other is.
2699 * A different workaround would be to call GetCurrencyFormat on
2700 * a known value, and parse it; patches welcome
2701 *
2702 * We have to use LC_ALL instead of LC_MONETARY because of
2703 * another bug in Windows */
2704
2705 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2706 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2707 save_global= savepv(my_setlocale(LC_ALL, NULL));
2708 my_setlocale(LC_ALL, save_thread);
2709
2710 # endif
2711
2712 lc = localeconv();
2713 if ( ! lc
2714 || ! lc->currency_symbol
2715 || strEQ("", lc->currency_symbol))
2716 {
2717 LOCALECONV_UNLOCK;
2718 return "";
2719 }
2720
2721 /* Leave the first spot empty to be filled in below */
2722 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2723 &PL_langinfo_bufsize, 1);
2724 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2725 { /* khw couldn't figure out how the localedef specifications
2726 would show that the $ should replace the radix; this is
2727 just a guess as to how it might work.*/
2728 PL_langinfo_buf[0] = '.';
2729 }
2730 else if (lc->p_cs_precedes) {
2731 PL_langinfo_buf[0] = '-';
2732 }
2733 else {
2734 PL_langinfo_buf[0] = '+';
2735 }
2736
2737 # ifdef TS_W32_BROKEN_LOCALECONV
2738
2739 my_setlocale(LC_ALL, save_global);
2740 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2741 my_setlocale(LC_ALL, save_thread);
2742 Safefree(save_global);
2743 Safefree(save_thread);
2744
2745 # endif
2746
2747 LOCALECONV_UNLOCK;
2748 break;
2749
2750 # ifdef TS_W32_BROKEN_LOCALECONV
2751
2752 case RADIXCHAR:
2753
2754 /* For this, we output a known simple floating point number to
2755 * a buffer, and parse it, looking for the radix */
2756
2757 if (toggle) {
2758 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2759 }
2760
2761 if (PL_langinfo_bufsize < 10) {
2762 PL_langinfo_bufsize = 10;
2763 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2764 }
2765
2766 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2767 "%.1f", 1.5);
2768 if (needed_size >= (int) PL_langinfo_bufsize) {
2769 PL_langinfo_bufsize = needed_size + 1;
2770 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2771 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2772 "%.1f", 1.5);
2773 assert(needed_size < (int) PL_langinfo_bufsize);
2774 }
2775
2776 ptr = PL_langinfo_buf;
2777 e = PL_langinfo_buf + PL_langinfo_bufsize;
2778
2779 /* Find the '1' */
2780 while (ptr < e && *ptr != '1') {
2781 ptr++;
2782 }
2783 ptr++;
2784
2785 /* Find the '5' */
2786 item_start = ptr;
2787 while (ptr < e && *ptr != '5') {
2788 ptr++;
2789 }
2790
2791 /* Everything in between is the radix string */
2792 if (ptr >= e) {
2793 PL_langinfo_buf[0] = '?';
2794 PL_langinfo_buf[1] = '\0';
2795 }
2796 else {
2797 *ptr = '\0';
2798 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2799 }
2800
2801 if (toggle) {
2802 RESTORE_LC_NUMERIC();
2803 }
2804
2805 retval = PL_langinfo_buf;
2806 break;
2807
2808 # else
2809
2810 case RADIXCHAR: /* No special handling needed */
2811
2812 # endif
2813
2814 case THOUSEP:
2815
2816 if (toggle) {
2817 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2818 }
2819
2820 LOCALECONV_LOCK; /* Prevent interference with other threads
2821 using localeconv() */
2822
2823 # ifdef TS_W32_BROKEN_LOCALECONV
2824
2825 /* This should only be for the thousands separator. A
2826 * different work around would be to use GetNumberFormat on a
2827 * known value and parse the result to find the separator */
2828 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2829 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2830 save_global = savepv(my_setlocale(LC_ALL, NULL));
2831 my_setlocale(LC_ALL, save_thread);
2832 # if 0
2833 /* This is the start of code that for broken Windows replaces
2834 * the above and below code, and instead calls
2835 * GetNumberFormat() and then would parse that to find the
2836 * thousands separator. It needs to handle UTF-16 vs -8
2837 * issues. */
2838
2839 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2840 DEBUG_L(PerlIO_printf(Perl_debug_log,
2841 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2842 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2843
2844 # endif
2845 # endif
2846
2847 lc = localeconv();
2848 if (! lc) {
2849 temp = "";
2850 }
2851 else {
2852 temp = (item == RADIXCHAR)
2853 ? lc->decimal_point
2854 : lc->thousands_sep;
2855 if (! temp) {
2856 temp = "";
2857 }
2858 }
2859
2860 retval = save_to_buffer(temp, &PL_langinfo_buf,
2861 &PL_langinfo_bufsize, 0);
2862
2863 # ifdef TS_W32_BROKEN_LOCALECONV
2864
2865 my_setlocale(LC_ALL, save_global);
2866 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2867 my_setlocale(LC_ALL, save_thread);
2868 Safefree(save_global);
2869 Safefree(save_thread);
2870
2871 # endif
2872
2873 LOCALECONV_UNLOCK;
2874
2875 if (toggle) {
2876 RESTORE_LC_NUMERIC();
2877 }
2878
2879 break;
2880
2881 # endif
2882 # ifdef HAS_STRFTIME
2883
2884 /* These are defined by C89, so we assume that strftime supports
2885 * them, and so are returned unconditionally; they may not be what
2886 * the locale actually says, but should give good enough results
2887 * for someone using them as formats (as opposed to trying to parse
2888 * them to figure out what the locale says). The other format
2889 * items are actually tested to verify they work on the platform */
2890 case D_FMT: return "%x";
2891 case T_FMT: return "%X";
2892 case D_T_FMT: return "%c";
2893
2894 /* These formats are only available in later strfmtime's */
2895 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2896
2897 /* The rest can be gotten from most versions of strftime(). */
2898 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2899 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2900 case ALT_DIGITS:
2901 case AM_STR: case PM_STR:
2902 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2903 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2904 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2905 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2906 case DAY_5: case DAY_6: case DAY_7:
2907 case MON_1: case MON_2: case MON_3: case MON_4:
2908 case MON_5: case MON_6: case MON_7: case MON_8:
2909 case MON_9: case MON_10: case MON_11: case MON_12:
2910
2911 init_tm(&tm); /* Precaution against core dumps */
2912 tm.tm_sec = 30;
2913 tm.tm_min = 30;
2914 tm.tm_hour = 6;
2915 tm.tm_year = 2017 - 1900;
2916 tm.tm_wday = 0;
2917 tm.tm_mon = 0;
2918
2919 GCC_DIAG_IGNORE_STMT(-Wimplicit-fallthrough);
2920
2921 switch (item) {
2922 default:
2923 Perl_croak(aTHX_
2924 "panic: %s: %d: switch case: %d problem",
2925 __FILE__, __LINE__, item);
2926 NOT_REACHED; /* NOTREACHED */
2927
2928 case PM_STR: tm.tm_hour = 18;
2929 case AM_STR:
2930 format = "%p";
2931 break;
2932
2933 case ABDAY_7: tm.tm_wday++;
2934 case ABDAY_6: tm.tm_wday++;
2935 case ABDAY_5: tm.tm_wday++;
2936 case ABDAY_4: tm.tm_wday++;
2937 case ABDAY_3: tm.tm_wday++;
2938 case ABDAY_2: tm.tm_wday++;
2939 case ABDAY_1:
2940 format = "%a";
2941 break;
2942
2943 case DAY_7: tm.tm_wday++;
2944 case DAY_6: tm.tm_wday++;
2945 case DAY_5: tm.tm_wday++;
2946 case DAY_4: tm.tm_wday++;
2947 case DAY_3: tm.tm_wday++;
2948 case DAY_2: tm.tm_wday++;
2949 case DAY_1:
2950 format = "%A";
2951 break;
2952
2953 case ABMON_12: tm.tm_mon++;
2954 case ABMON_11: tm.tm_mon++;
2955 case ABMON_10: tm.tm_mon++;
2956 case ABMON_9: tm.tm_mon++;
2957 case ABMON_8: tm.tm_mon++;
2958 case ABMON_7: tm.tm_mon++;
2959 case ABMON_6: tm.tm_mon++;
2960 case ABMON_5: tm.tm_mon++;
2961 case ABMON_4: tm.tm_mon++;
2962 case ABMON_3: tm.tm_mon++;
2963 case ABMON_2: tm.tm_mon++;
2964 case ABMON_1:
2965 format = "%b";
2966 break;
2967
2968 case MON_12: tm.tm_mon++;
2969 case MON_11: tm.tm_mon++;
2970 case MON_10: tm.tm_mon++;
2971 case MON_9: tm.tm_mon++;
2972 case MON_8: tm.tm_mon++;
2973 case MON_7: tm.tm_mon++;
2974 case MON_6: tm.tm_mon++;
2975 case MON_5: tm.tm_mon++;
2976 case MON_4: tm.tm_mon++;
2977 case MON_3: tm.tm_mon++;
2978 case MON_2: tm.tm_mon++;
2979 case MON_1:
2980 format = "%B";
2981 break;
2982
2983 case T_FMT_AMPM:
2984 format = "%r";
2985 return_format = TRUE;
2986 break;
2987
2988 case ERA_D_FMT:
2989 format = "%Ex";
2990 return_format = TRUE;
2991 break;
2992
2993 case ERA_T_FMT:
2994 format = "%EX";
2995 return_format = TRUE;
2996 break;
2997
2998 case ERA_D_T_FMT:
2999 format = "%Ec";
3000 return_format = TRUE;
3001 break;
3002
3003 case ALT_DIGITS:
3004 tm.tm_wday = 0;
3005 format = "%Ow"; /* Find the alternate digit for 0 */
3006 break;
3007 }
3008
3009 GCC_DIAG_RESTORE_STMT;
3010
3011 /* We can't use my_strftime() because it doesn't look at
3012 * tm_wday */
3013 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3014 format, &tm))
3015 {
3016 /* A zero return means one of:
3017 * a) there wasn't enough space in PL_langinfo_buf
3018 * b) the format, like a plain %p, returns empty
3019 * c) it was an illegal format, though some
3020 * implementations of strftime will just return the
3021 * illegal format as a plain character sequence.
3022 *
3023 * To quickly test for case 'b)', try again but precede
3024 * the format with a plain character. If that result is
3025 * still empty, the problem is either 'a)' or 'c)' */
3026
3027 Size_t format_size = strlen(format) + 1;
3028 Size_t mod_size = format_size + 1;
3029 char * mod_format;
3030 char * temp_result;
3031
3032 Newx(mod_format, mod_size, char);
3033 Newx(temp_result, PL_langinfo_bufsize, char);
3034 *mod_format = ' ';
3035 my_strlcpy(mod_format + 1, format, mod_size);
3036 len = strftime(temp_result,
3037 PL_langinfo_bufsize,
3038 mod_format, &tm);
3039 Safefree(mod_format);
3040 Safefree(temp_result);
3041
3042 /* If 'len' is non-zero, it means that we had a case like
3043 * %p which means the current locale doesn't use a.m. or
3044 * p.m., and that is valid */
3045 if (len == 0) {
3046
3047 /* Here, still didn't work. If we get well beyond a
3048 * reasonable size, bail out to prevent an infinite
3049 * loop. */
3050
3051 if (PL_langinfo_bufsize > 100 * format_size) {
3052 *PL_langinfo_buf = '\0';
3053 }
3054 else {
3055 /* Double the buffer size to retry; Add 1 in case
3056 * original was 0, so we aren't stuck at 0. */
3057 PL_langinfo_bufsize *= 2;
3058 PL_langinfo_bufsize++;
3059 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3060 continue;
3061 }
3062 }
3063
3064 break;
3065 }
3066
3067 /* Here, we got a result.
3068 *
3069 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3070 * alternate format for wday 0. If the value is the same as
3071 * the normal 0, there isn't an alternate, so clear the buffer.
3072 * */
3073 if ( item == ALT_DIGITS
3074 && strEQ(PL_langinfo_buf, "0"))
3075 {
3076 *PL_langinfo_buf = '\0';
3077 }
3078
3079 /* ALT_DIGITS is problematic. Experiments on it showed that
3080 * strftime() did not always work properly when going from
3081 * alt-9 to alt-10. Only a few locales have this item defined,
3082 * and in all of them on Linux that khw was able to find,
3083 * nl_langinfo() merely returned the alt-0 character, possibly
3084 * doubled. Most Unicode digits are in blocks of 10
3085 * consecutive code points, so that is sufficient information
3086 * for those scripts, as we can infer alt-1, alt-2, .... But
3087 * for a Japanese locale, a CJK ideographic 0 is returned, and
3088 * the CJK digits are not in code point order, so you can't
3089 * really infer anything. The localedef for this locale did
3090 * specify the succeeding digits, so that strftime() works
3091 * properly on them, without needing to infer anything. But
3092 * the nl_langinfo() return did not give sufficient information
3093 * for the caller to understand what's going on. So until
3094 * there is evidence that it should work differently, this
3095 * returns the alt-0 string for ALT_DIGITS.
3096 *
3097 * wday was chosen because its range is all a single digit.
3098 * Things like tm_sec have two digits as the minimum: '00' */
3099
3100 retval = PL_langinfo_buf;
3101
3102 /* If to return the format, not the value, overwrite the buffer
3103 * with it. But some strftime()s will keep the original format
3104 * if illegal, so change those to "" */
3105 if (return_format) {
3106 if (strEQ(PL_langinfo_buf, format)) {
3107 *PL_langinfo_buf = '\0';
3108 }
3109 else {
3110 retval = save_to_buffer(format, &PL_langinfo_buf,
3111 &PL_langinfo_bufsize, 0);
3112 }
3113 }
3114
3115 break;
3116
3117 # endif
3118
3119 }
3120 }
3121
3122 return retval;
3123
3124 #endif
3125
3126 }
3127
3128 /*
3129 * Initialize locale awareness.
3130 */
3131 int
3132 Perl_init_i18nl10n(pTHX_ int printwarn)
3133 {
3134 /* printwarn is
3135 *
3136 * 0 if not to output warning when setup locale is bad
3137 * 1 if to output warning based on value of PERL_BADLANG
3138 * >1 if to output regardless of PERL_BADLANG
3139 *
3140 * returns
3141 * 1 = set ok or not applicable,
3142 * 0 = fallback to a locale of lower priority
3143 * -1 = fallback to all locales failed, not even to the C locale
3144 *
3145 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3146 * set, debugging information is output.
3147 *
3148 * This looks more complicated than it is, mainly due to the #ifdefs and
3149 * error handling.
3150 *
3151 * Besides some asserts, data structure initialization, and specific
3152 * platform complications, this routine is effectively just two things.
3153 *
3154 * a) setlocale(LC_ALL, "");
3155 *
3156 * which sets LC_ALL to the values in the current environment.
3157 *
3158 * And for each individual category 'foo' whose value we care about:
3159 *
3160 * b) save_foo = setlocale(LC_foo, NULL); handle_foo(save_foo);
3161 *
3162 * (We don't tend to care about categories like LC_PAPER, for example.)
3163 *
3164 * But there are complications. On systems without LC_ALL, it emulates
3165 * step a) by looping through all the categories, and doing
3166 *
3167 * setlocale(LC_foo, "");
3168 *
3169 * on each.
3170 *
3171 * And it has to deal with if this is an embedded perl, whose locale
3172 * doesn't come from the environment, but has been set up by the caller.
3173 * This is pretty simply handled: the "" in the setlocale calls is not a
3174 * string constant, but a variable which is set to NULL in the embedded
3175 * case.
3176 *
3177 * But the major complication is handling failure and doing fallback.
3178 * There is an array, trial_locales, the elements of which are looped over
3179 * until the locale is successfully set. The array is initialized with
3180 * just one element, for
3181 * setlocale(LC_ALL, $NULL_or_empty)
3182 * If that works, as it almost always does, there's no more elements and
3183 * the loop iterates just the once. Otherwise elements are added for each
3184 * of the environment variables that POSIX dictates should control the
3185 * program, in priority order, with a final one being "C". The loop is
3186 * repeated until the first one succeeds. If all fail, we limp along with
3187 * whatever state we got to. If there is no LC_ALL, an inner loop is run
3188 * through all categories (making things look complex).
3189 *
3190 * A further complication is that Windows has an additional fallback, the
3191 * user-default ANSI code page obtained from the operating system. This is
3192 * added as yet another loop iteration, just before the final "C"
3193 *
3194 * On Ultrix, the locale MUST come from the environment, so there is
3195 * preliminary code to set it. I (khw) am not sure that it is necessary,
3196 * and that this couldn't be folded into the loop, but barring any real
3197 * platforms to test on, it's staying as-is
3198 */
3199
3200 int ok = 1;
3201
3202 #ifndef USE_LOCALE
3203
3204 PERL_UNUSED_ARG(printwarn);
3205
3206 #else /* USE_LOCALE */
3207 # ifdef __GLIBC__
3208
3209 const char * const language = PerlEnv_getenv("LANGUAGE");
3210
3211 # endif
3212
3213 /* NULL uses the existing already set up locale */
3214 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3215 ? NULL
3216 : "";
3217 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3218 unsigned int trial_locales_count;
3219 const char * const lc_all = PerlEnv_getenv("LC_ALL");
3220 const char * const lang = PerlEnv_getenv("LANG");
3221 bool setlocale_failure = FALSE;
3222 unsigned int i;
3223
3224 /* A later getenv() could zap this, so only use here */
3225 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3226
3227 const bool locwarn = (printwarn > 1
3228 || ( printwarn
3229 && ( ! bad_lang_use_once
3230 || (
3231 /* disallow with "" or "0" */
3232 *bad_lang_use_once
3233 && strNE("0", bad_lang_use_once)))));
3234
3235 /* setlocale() return vals; not copied so must be looked at immediately */
3236 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3237
3238 /* current locale for given category; should have been copied so aren't
3239 * volatile */
3240 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3241
3242 # ifdef WIN32
3243
3244 /* In some systems you can find out the system default locale
3245 * and use that as the fallback locale. */
3246 # define SYSTEM_DEFAULT_LOCALE
3247 # endif
3248 # ifdef SYSTEM_DEFAULT_LOCALE
3249
3250 const char *system_default_locale = NULL;
3251
3252 # endif
3253
3254 # ifndef DEBUGGING
3255 # define DEBUG_LOCALE_INIT(a,b,c)
3256 # else
3257
3258 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3259
3260 # define DEBUG_LOCALE_INIT(category, locale, result) \
3261 STMT_START { \
3262 if (debug_initialization) { \
3263 PerlIO_printf(Perl_debug_log, \
3264 "%s:%d: %s\n", \
3265 __FILE__, __LINE__, \
3266 setlocale_debug_string(category, \
3267 locale, \
3268 result)); \
3269 } \
3270 } STMT_END
3271
3272 /* Make sure the parallel arrays are properly set up */
3273 # ifdef USE_LOCALE_NUMERIC
3274 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3275 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3276 # ifdef USE_POSIX_2008_LOCALE
3277 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3278 # endif
3279 # endif
3280 # ifdef USE_LOCALE_CTYPE
3281 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3282 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3283 # ifdef USE_POSIX_2008_LOCALE
3284 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3285 # endif
3286 # endif
3287 # ifdef USE_LOCALE_COLLATE
3288 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3289 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3290 # ifdef USE_POSIX_2008_LOCALE
3291 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3292 # endif
3293 # endif
3294 # ifdef USE_LOCALE_TIME
3295 assert(categories[LC_TIME_INDEX] == LC_TIME);
3296 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3297 # ifdef USE_POSIX_2008_LOCALE
3298 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3299 # endif
3300 # endif
3301 # ifdef USE_LOCALE_MESSAGES
3302 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3303 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3304 # ifdef USE_POSIX_2008_LOCALE
3305 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3306 # endif
3307 # endif
3308 # ifdef USE_LOCALE_MONETARY
3309 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3310 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3311 # ifdef USE_POSIX_2008_LOCALE
3312 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3313 # endif
3314 # endif
3315 # ifdef USE_LOCALE_ADDRESS
3316 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3317 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3318 # ifdef USE_POSIX_2008_LOCALE
3319 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3320 # endif
3321 # endif
3322 # ifdef USE_LOCALE_IDENTIFICATION
3323 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3324 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3325 # ifdef USE_POSIX_2008_LOCALE
3326 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3327 # endif
3328 # endif
3329 # ifdef USE_LOCALE_MEASUREMENT
3330 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3331 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3332 # ifdef USE_POSIX_2008_LOCALE
3333 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3334 # endif
3335 # endif
3336 # ifdef USE_LOCALE_PAPER
3337 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3338 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3339 # ifdef USE_POSIX_2008_LOCALE
3340 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3341 # endif
3342 # endif
3343 # ifdef USE_LOCALE_TELEPHONE
3344 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3345 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3346 # ifdef USE_POSIX_2008_LOCALE
3347 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3348 # endif
3349 # endif
3350 # ifdef USE_LOCALE_SYNTAX
3351 assert(categories[LC_SYNTAX_INDEX] == LC_SYNTAX);
3352 assert(strEQ(category_names[LC_SYNTAX_INDEX], "LC_SYNTAX"));
3353 # ifdef USE_POSIX_2008_LOCALE
3354 assert(category_masks[LC_SYNTAX_INDEX] == LC_SYNTAX_MASK);
3355 # endif
3356 # endif
3357 # ifdef USE_LOCALE_TOD
3358 assert(categories[LC_TOD_INDEX] == LC_TOD);
3359 assert(strEQ(category_names[LC_TOD_INDEX], "LC_TOD"));
3360 # ifdef USE_POSIX_2008_LOCALE
3361 assert(category_masks[LC_TOD_INDEX] == LC_TOD_MASK);
3362 # endif
3363 # endif
3364 # ifdef LC_ALL
3365 assert(categories[LC_ALL_INDEX] == LC_ALL);
3366 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3367 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3368 # ifdef USE_POSIX_2008_LOCALE
3369 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3370 # endif
3371 # endif
3372 # endif /* DEBUGGING */
3373
3374 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
3375 * why these particular incantations are used. */
3376 #ifdef HAS_MBRLEN
3377 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
3378 #endif
3379 #ifdef HAS_MBRTOWC
3380 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
3381 #endif
3382 #ifdef HAS_WCTOMBR
3383 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
3384 #endif
3385
3386 /* Initialize the cache of the program's UTF-8ness for the always known
3387 * locales C and POSIX */
3388 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3389 sizeof(PL_locale_utf8ness));
3390
3391 # ifdef USE_THREAD_SAFE_LOCALE
3392 # ifdef WIN32
3393
3394 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3395
3396 # endif
3397 # endif
3398 # ifdef USE_POSIX_2008_LOCALE
3399
3400 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3401 if (! PL_C_locale_obj) {
3402 Perl_croak_nocontext(
3403 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3404 }
3405 if (DEBUG_Lv_TEST || debug_initialization) {
3406 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3407 }
3408
3409 # endif
3410
3411 # ifdef USE_LOCALE_NUMERIC
3412
3413 PL_numeric_radix_sv = newSVpvs(".");
3414
3415 # endif
3416
3417 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3418
3419 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3420 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3421
3422 # endif
3423 # ifdef LOCALE_ENVIRON_REQUIRED
3424
3425 /*
3426 * Ultrix setlocale(..., "") fails if there are no environment
3427 * variables from which to get a locale name.
3428 */
3429
3430 # ifndef LC_ALL
3431 # error Ultrix without LC_ALL not implemented
3432 # else
3433
3434 {
3435 bool done = FALSE;
3436 if (lang) {
3437 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3438 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3439 if (sl_result[LC_ALL_INDEX])
3440 done = TRUE;
3441 else
3442 setlocale_failure = TRUE;
3443 }
3444 if (! setlocale_failure) {
3445 const char * locale_param;
3446 for (i = 0; i < LC_ALL_INDEX; i++) {
3447 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3448 ? setlocale_init
3449 : NULL;
3450 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3451 if (! sl_result[i]) {
3452 setlocale_failure = TRUE;
3453 }
3454 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3455 }
3456 }
3457 }
3458
3459 # endif /* LC_ALL */
3460 # endif /* LOCALE_ENVIRON_REQUIRED */
3461
3462 /* We try each locale in the list until we get one that works, or exhaust
3463 * the list. Normally the loop is executed just once. But if setting the
3464 * locale fails, inside the loop we add fallback trials to the array and so
3465 * will execute the loop multiple times */
3466 trial_locales[0] = setlocale_init;
3467 trial_locales_count = 1;
3468
3469 for (i= 0; i < trial_locales_count; i++) {
3470 const char * trial_locale = trial_locales[i];
3471
3472 if (i > 0) {
3473
3474 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3475 * when i==0, but I (khw) don't think that behavior makes much
3476 * sense */
3477 setlocale_failure = FALSE;
3478
3479 # ifdef SYSTEM_DEFAULT_LOCALE
3480 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3481
3482 /* On Windows machines, an entry of "" after the 0th means to use
3483 * the system default locale, which we now proceed to get. */
3484 if (strEQ(trial_locale, "")) {
3485 unsigned int j;
3486
3487 /* Note that this may change the locale, but we are going to do
3488 * that anyway just below */
3489 system_default_locale = do_setlocale_c(LC_ALL, "");
3490 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3491
3492 /* Skip if invalid or if it's already on the list of locales to
3493 * try */
3494 if (! system_default_locale) {
3495 goto next_iteration;
3496 }
3497 for (j = 0; j < trial_locales_count; j++) {
3498 if (strEQ(system_default_locale, trial_locales[j])) {
3499 goto next_iteration;
3500 }
3501 }
3502
3503 trial_locale = system_default_locale;
3504 }
3505 # else
3506 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3507 # endif
3508 # endif /* SYSTEM_DEFAULT_LOCALE */
3509
3510 } /* For i > 0 */
3511
3512 # ifdef LC_ALL
3513
3514 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3515 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3516 if (! sl_result[LC_ALL_INDEX]) {
3517 setlocale_failure = TRUE;
3518 }
3519 else {
3520 /* Since LC_ALL succeeded, it should have changed all the other
3521 * categories it can to its value; so we massage things so that the
3522 * setlocales below just return their category's current values.
3523 * This adequately handles the case in NetBSD where LC_COLLATE may
3524 * not be defined for a locale, and setting it individually will
3525 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3526 * the POSIX locale. */
3527 trial_locale = NULL;
3528 }
3529
3530 # endif /* LC_ALL */
3531
3532 if (! setlocale_failure) {
3533 unsigned int j;
3534 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3535 curlocales[j]
3536 = savepv(do_setlocale_r(categories[j], trial_locale));
3537 if (! curlocales[j]) {
3538 setlocale_failure = TRUE;
3539 }
3540 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3541 }
3542
3543 if (LIKELY(! setlocale_failure)) { /* All succeeded */
3544 break; /* Exit trial_locales loop */
3545 }
3546 }
3547
3548 /* Here, something failed; will need to try a fallback. */
3549 ok = 0;
3550
3551 if (i == 0) {
3552 unsigned int j;
3553
3554 if (locwarn) { /* Output failure info only on the first one */
3555
3556 # ifdef LC_ALL
3557
3558 PerlIO_printf(Perl_error_log,
3559 "perl: warning: Setting locale failed.\n");
3560
3561 # else /* !LC_ALL */
3562
3563 PerlIO_printf(Perl_error_log,
3564 "perl: warning: Setting locale failed for the categories:\n\t");
3565
3566 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3567 if (! curlocales[j]) {
3568 PerlIO_printf(Perl_error_log, category_names[j]);
3569 }
3570 else {
3571 Safefree(curlocales[j]);
3572 }
3573 }
3574
3575 # endif /* LC_ALL */
3576
3577 PerlIO_printf(Perl_error_log,
3578 "perl: warning: Please check that your locale settings:\n");
3579
3580 # ifdef __GLIBC__
3581
3582 PerlIO_printf(Perl_error_log,
3583 "\tLANGUAGE = %c%s%c,\n",
3584 language ? '"' : '(',
3585 language ? language : "unset",
3586 language ? '"' : ')');
3587 # endif
3588
3589 PerlIO_printf(Perl_error_log,
3590 "\tLC_ALL = %c%s%c,\n",
3591 lc_all ? '"' : '(',
3592 lc_all ? lc_all : "unset",
3593 lc_all ? '"' : ')');
3594
3595 # if defined(USE_ENVIRON_ARRAY)
3596
3597 {
3598 char **e;
3599
3600 /* Look through the environment for any variables of the
3601 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3602 * already handled above. These are assumed to be locale
3603 * settings. Output them and their values. */
3604 for (e = environ; *e; e++) {
3605 const STRLEN prefix_len = sizeof("LC_") - 1;
3606 STRLEN uppers_len;
3607
3608 if ( strBEGINs(*e, "LC_")
3609 && ! strBEGINs(*e, "LC_ALL=")
3610 && (uppers_len = strspn(*e + prefix_len,
3611 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3612 && ((*e)[prefix_len + uppers_len] == '='))
3613 {
3614 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3615 (int) (prefix_len + uppers_len), *e,
3616 *e + prefix_len + uppers_len + 1);
3617 }
3618 }
3619 }
3620
3621 # else
3622
3623 PerlIO_printf(Perl_error_log,
3624 "\t(possibly more locale environment variables)\n");
3625
3626 # endif
3627
3628 PerlIO_printf(Perl_error_log,
3629 "\tLANG = %c%s%c\n",
3630 lang ? '"' : '(',
3631 lang ? lang : "unset",
3632 lang ? '"' : ')');
3633
3634 PerlIO_printf(Perl_error_log,
3635 " are supported and installed on your system.\n");
3636 }
3637
3638 /* Calculate what fallback locales to try. We have avoided this
3639 * until we have to, because failure is quite unlikely. This will
3640 * usually change the upper bound of the loop we are in.
3641 *
3642 * Since the system's default way of setting the locale has not
3643 * found one that works, We use Perl's defined ordering: LC_ALL,
3644 * LANG, and the C locale. We don't try the same locale twice, so
3645 * don't add to the list if already there. (On POSIX systems, the
3646 * LC_ALL element will likely be a repeat of the 0th element "",
3647 * but there's no harm done by doing it explicitly.
3648 *
3649 * Note that this tries the LC_ALL environment variable even on
3650 * systems which have no LC_ALL locale setting. This may or may
3651 * not have been originally intentional, but there's no real need
3652 * to change the behavior. */
3653 if (lc_all) {
3654 for (j = 0; j < trial_locales_count; j++) {
3655 if (strEQ(lc_all, trial_locales[j])) {
3656 goto done_lc_all;
3657 }
3658 }
3659 trial_locales[trial_locales_count++] = lc_all;
3660 }
3661 done_lc_all:
3662
3663 if (lang) {
3664 for (j = 0; j < trial_locales_count; j++) {
3665 if (strEQ(lang, trial_locales[j])) {
3666 goto done_lang;
3667 }
3668 }
3669 trial_locales[trial_locales_count++] = lang;
3670 }
3671 done_lang:
3672
3673 # if defined(WIN32) && defined(LC_ALL)
3674
3675 /* For Windows, we also try the system default locale before "C".
3676 * (If there exists a Windows without LC_ALL we skip this because
3677 * it gets too complicated. For those, the "C" is the next
3678 * fallback possibility). The "" is the same as the 0th element of
3679 * the array, but the code at the loop above knows to treat it
3680 * differently when not the 0th */
3681 trial_locales[trial_locales_count++] = "";
3682
3683 # endif
3684
3685 for (j = 0; j < trial_locales_count; j++) {
3686 if (strEQ("C", trial_locales[j])) {
3687 goto done_C;
3688 }
3689 }
3690 trial_locales[trial_locales_count++] = "C";
3691
3692 done_C: ;
3693 } /* end of first time through the loop */
3694
3695 # ifdef WIN32
3696
3697 next_iteration: ;
3698
3699 # endif
3700
3701 } /* end of looping through the trial locales */
3702
3703 if (ok < 1) { /* If we tried to fallback */
3704 const char* msg;
3705 if (! setlocale_failure) { /* fallback succeeded */
3706 msg = "Falling back to";
3707 }
3708 else { /* fallback failed */
3709 unsigned int j;
3710
3711 /* We dropped off the end of the loop, so have to decrement i to
3712 * get back to the value the last time through */
3713 i--;
3714
3715 ok = -1;
3716 msg = "Failed to fall back to";
3717
3718 /* To continue, we should use whatever values we've got */
3719
3720 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3721 Safefree(curlocales[j]);
3722 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3723 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3724 }
3725 }
3726
3727 if (locwarn) {
3728 const char * description;
3729 const char * name = "";
3730 if (strEQ(trial_locales[i], "C")) {
3731 description = "the standard locale";
3732 name = "C";
3733 }
3734
3735 # ifdef SYSTEM_DEFAULT_LOCALE
3736
3737 else if (strEQ(trial_locales[i], "")) {
3738 description = "the system default locale";
3739 if (system_default_locale) {
3740 name = system_default_locale;
3741 }
3742 }
3743
3744 # endif /* SYSTEM_DEFAULT_LOCALE */
3745
3746 else {
3747 description = "a fallback locale";
3748 name = trial_locales[i];
3749 }
3750 if (name && strNE(name, "")) {
3751 PerlIO_printf(Perl_error_log,
3752 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3753 }
3754 else {
3755 PerlIO_printf(Perl_error_log,
3756 "perl: warning: %s %s.\n", msg, description);
3757 }
3758 }
3759 } /* End of tried to fallback */
3760
3761 /* Done with finding the locales; update our records */
3762
3763 # ifdef USE_LOCALE_CTYPE
3764
3765 new_ctype(curlocales[LC_CTYPE_INDEX]);
3766
3767 # endif
3768 # ifdef USE_LOCALE_COLLATE
3769
3770 new_collate(curlocales[LC_COLLATE_INDEX]);
3771
3772 # endif
3773 # ifdef USE_LOCALE_NUMERIC
3774
3775 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3776
3777 # endif
3778
3779 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3780
3781 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3782
3783 /* This caches whether each category's locale is UTF-8 or not. This
3784 * may involve changing the locale. It is ok to do this at
3785 * initialization time before any threads have started, but not later
3786 * unless thread-safe operations are used.
3787 * Caching means that if the program heeds our dictate not to change
3788 * locales in threaded applications, this data will remain valid, and
3789 * it may get queried without having to change locales. If the
3790 * environment is such that all categories have the same locale, this
3791 * isn't needed, as the code will not change the locale; but this
3792 * handles the uncommon case where the environment has disparate
3793 * locales for the categories */
3794 (void) _is_cur_LC_category_utf8(categories[i]);
3795
3796 # endif
3797
3798 Safefree(curlocales[i]);
3799 }
3800
3801 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3802
3803 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3804 * locale is UTF-8. The call to new_ctype() just above has already
3805 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3806 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3807 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3808 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3809 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3810
3811 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3812 This is an alternative to using the -C command line switch
3813 (the -C if present will override this). */
3814 {
3815 const char *p = PerlEnv_getenv("PERL_UNICODE");
3816 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3817 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3818 PL_utf8cache = -1;
3819 }
3820
3821 # endif
3822 #endif /* USE_LOCALE */
3823 #ifdef DEBUGGING
3824
3825 /* So won't continue to output stuff */
3826 DEBUG_INITIALIZATION_set(FALSE);
3827
3828 #endif
3829
3830 return ok;
3831 }
3832
3833 #ifdef USE_LOCALE_COLLATE
3834
3835 char *
3836 Perl__mem_collxfrm(pTHX_ const char *input_string,
3837 STRLEN len, /* Length of 'input_string' */
3838 STRLEN *xlen, /* Set to length of returned string
3839 (not including the collation index
3840 prefix) */
3841 bool utf8 /* Is the input in UTF-8? */
3842 )
3843 {
3844
3845 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3846 * differences. First, it handles embedded NULs. Second, it allocates a bit
3847 * more memory than needed for the transformed data itself. The real
3848 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3849 * the length of that, and doesn't include the collation index size.
3850 * Please see sv_collxfrm() to see how this is used. */
3851
3852 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3853
3854 char * s = (char *) input_string;
3855 STRLEN s_strlen = strlen(input_string);
3856 char *xbuf = NULL;
3857 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3858 STRLEN length_in_chars;
3859 bool first_time = TRUE; /* Cleared after first loop iteration */
3860
3861 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3862
3863 /* Must be NUL-terminated */
3864 assert(*(input_string + len) == '\0');
3865
3866 /* If this locale has defective collation, skip */
3867 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3868 DEBUG_L(PerlIO_printf(Perl_debug_log,
3869 "_mem_collxfrm: locale's collation is defective\n"));
3870 goto bad;
3871 }
3872
3873 /* Replace any embedded NULs with the control that sorts before any others.
3874 * This will give as good as possible results on strings that don't
3875 * otherwise contain that character, but otherwise there may be
3876 * less-than-perfect results with that character and NUL. This is
3877 * unavoidable unless we replace strxfrm with our own implementation. */
3878 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3879 NUL */
3880 char * e = s + len;
3881 char * sans_nuls;
3882 STRLEN sans_nuls_len;
3883 int try_non_controls;
3884 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3885 making sure 2nd byte is NUL.
3886 */
3887 STRLEN this_replacement_len;
3888
3889 /* If we don't know what non-NUL control character sorts lowest for
3890 * this locale, find it */
3891 if (PL_strxfrm_NUL_replacement == '\0') {
3892 int j;
3893 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3894 includes the collation index
3895 prefixed. */
3896
3897 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3898
3899 /* Unlikely, but it may be that no control will work to replace
3900 * NUL, in which case we instead look for any character. Controls
3901 * are preferred because collation order is, in general, context
3902 * sensitive, with adjoining characters affecting the order, and
3903 * controls are less likely to have such interactions, allowing the
3904 * NUL-replacement to stand on its own. (Another way to look at it
3905 * is to imagine what would happen if the NUL were replaced by a
3906 * combining character; it wouldn't work out all that well.) */
3907 for (try_non_controls = 0;
3908 try_non_controls < 2;
3909 try_non_controls++)
3910 {
3911 /* Look through all legal code points (NUL isn't) */
3912 for (j = 1; j < 256; j++) {
3913 char * x; /* j's xfrm plus collation index */
3914 STRLEN x_len; /* length of 'x' */
3915 STRLEN trial_len = 1;
3916 char cur_source[] = { '\0', '\0' };
3917
3918 /* Skip non-controls the first time through the loop. The
3919 * controls in a UTF-8 locale are the L1 ones */
3920 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3921 ? ! isCNTRL_L1(j)
3922 : ! isCNTRL_LC(j))
3923 {
3924 continue;
3925 }
3926
3927 /* Create a 1-char string of the current code point */
3928 cur_source[0] = (char) j;
3929
3930 /* Then transform it */
3931 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3932 0 /* The string is not in UTF-8 */);
3933
3934 /* Ignore any character that didn't successfully transform.
3935 * */
3936 if (! x) {
3937 continue;
3938 }
3939
3940 /* If this character's transformation is lower than
3941 * the current lowest, this one becomes the lowest */
3942 if ( cur_min_x == NULL
3943 || strLT(x + COLLXFRM_HDR_LEN,
3944 cur_min_x + COLLXFRM_HDR_LEN))
3945 {
3946 PL_strxfrm_NUL_replacement = j;
3947 Safefree(cur_min_x);
3948 cur_min_x = x;
3949 }
3950 else {
3951 Safefree(x);
3952 }
3953 } /* end of loop through all 255 characters */
3954
3955 /* Stop looking if found */
3956 if (cur_min_x) {
3957 break;
3958 }
3959
3960 /* Unlikely, but possible, if there aren't any controls that
3961 * work in the locale, repeat the loop, looking for any
3962 * character that works */
3963 DEBUG_L(PerlIO_printf(Perl_debug_log,
3964 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3965 } /* End of loop to try first the controls, then any char */
3966
3967 if (! cur_min_x) {
3968 DEBUG_L(PerlIO_printf(Perl_debug_log,
3969 "_mem_collxfrm: Couldn't find any character to replace"
3970 " embedded NULs in locale %s with", PL_collation_name));
3971 goto bad;
3972 }
3973
3974 DEBUG_L(PerlIO_printf(Perl_debug_log,
3975 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3976 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3977
3978 Safefree(cur_min_x);
3979 } /* End of determining the character that is to replace NULs */
3980
3981 /* If the replacement is variant under UTF-8, it must match the
3982 * UTF8-ness of the original */
3983 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3984 this_replacement_char[0] =
3985 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3986 this_replacement_char[1] =
3987 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3988 this_replacement_len = 2;
3989 }
3990 else {
3991 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3992 /* this_replacement_char[1] = '\0' was done at initialization */
3993 this_replacement_len = 1;
3994 }
3995
3996 /* The worst case length for the replaced string would be if every
3997 * character in it is NUL. Multiply that by the length of each
3998 * replacement, and allow for a trailing NUL */
3999 sans_nuls_len = (len * this_replacement_len) + 1;
4000 Newx(sans_nuls, sans_nuls_len, char);
4001 *sans_nuls = '\0';
4002
4003 /* Replace each NUL with the lowest collating control. Loop until have
4004 * exhausted all the NULs */
4005 while (s + s_strlen < e) {
4006 my_strlcat(sans_nuls, s, sans_nuls_len);
4007
4008 /* Do the actual replacement */
4009 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
4010
4011 /* Move past the input NUL */
4012 s += s_strlen + 1;
4013 s_strlen = strlen(s);
4014 }
4015
4016 /* And add anything that trails the final NUL */
4017 my_strlcat(sans_nuls, s, sans_nuls_len);
4018
4019 /* Switch so below we transform this modified string */
4020 s = sans_nuls;
4021 len = strlen(s);
4022 } /* End of replacing NULs */
4023
4024 /* Make sure the UTF8ness of the string and locale match */
4025 if (utf8 != PL_in_utf8_COLLATE_locale) {
4026 /* XXX convert above Unicode to 10FFFF? */
4027 const char * const t = s; /* Temporary so we can later find where the
4028 input was */
4029
4030 /* Here they don't match. Change the string's to be what the locale is
4031 * expecting */
4032
4033 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
4034 s = (char *) bytes_to_utf8((const U8 *) s, &len);
4035 utf8 = TRUE;
4036 }
4037 else { /* locale is not UTF-8; but input is; downgrade the input */
4038
4039 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
4040
4041 /* If the downgrade was successful we are done, but if the input
4042 * contains things that require UTF-8 to represent, have to do
4043 * damage control ... */
4044 if (UNLIKELY(utf8)) {
4045
4046 /* What we do is construct a non-UTF-8 string with
4047 * 1) the characters representable by a single byte converted
4048 * to be so (if necessary);
4049 * 2) and the rest converted to collate the same as the
4050 * highest collating representable character. That makes
4051 * them collate at the end. This is similar to how we
4052 * handle embedded NULs, but we use the highest collating
4053 * code point instead of the smallest. Like the NUL case,
4054 * this isn't perfect, but is the best we can reasonably
4055 * do. Every above-255 code point will sort the same as
4056 * the highest-sorting 0-255 code point. If that code
4057 * point can combine in a sequence with some other code
4058 * points for weight calculations, us changing something to
4059 * be it can adversely affect the results. But in most
4060 * cases, it should work reasonably. And note that this is
4061 * really an illegal situation: using code points above 255
4062 * on a locale where only 0-255 are valid. If two strings
4063 * sort entirely equal, then the sort order for the
4064 * above-255 code points will be in code point order. */
4065
4066 utf8 = FALSE;
4067
4068 /* If we haven't calculated the code point with the maximum
4069 * collating order for this locale, do so now */
4070 if (! PL_strxfrm_max_cp) {
4071 int j;
4072
4073 /* The current transformed string that collates the
4074 * highest (except it also includes the prefixed collation
4075 * index. */
4076 char * cur_max_x = NULL;
4077
4078 /* Look through all legal code points (NUL isn't) */
4079 for (j = 1; j < 256; j++) {
4080 char * x;
4081 STRLEN x_len;
4082 char cur_source[] = { '\0', '\0' };
4083
4084 /* Create a 1-char string of the current code point */
4085 cur_source[0] = (char) j;
4086
4087 /* Then transform it */
4088 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4089
4090 /* If something went wrong (which it shouldn't), just
4091 * ignore this code point */
4092 if (! x) {
4093 continue;
4094 }
4095
4096 /* If this character's transformation is higher than
4097 * the current highest, this one becomes the highest */
4098 if ( cur_max_x == NULL
4099 || strGT(x + COLLXFRM_HDR_LEN,
4100 cur_max_x + COLLXFRM_HDR_LEN))
4101 {
4102 PL_strxfrm_max_cp = j;
4103 Safefree(cur_max_x);
4104 cur_max_x = x;
4105 }
4106 else {
4107 Safefree(x);
4108 }
4109 }
4110
4111 if (! cur_max_x) {
4112 DEBUG_L(PerlIO_printf(Perl_debug_log,
4113 "_mem_collxfrm: Couldn't find any character to"
4114 " replace above-Latin1 chars in locale %s with",
4115 PL_collation_name));
4116 goto bad;
4117 }
4118
4119 DEBUG_L(PerlIO_printf(Perl_debug_log,
4120 "_mem_collxfrm: highest 1-byte collating character"
4121 " in locale %s is 0x%02X\n",
4122 PL_collation_name,
4123 PL_strxfrm_max_cp));
4124
4125 Safefree(cur_max_x);
4126 }
4127
4128 /* Here we know which legal code point collates the highest.
4129 * We are ready to construct the non-UTF-8 string. The length
4130 * will be at least 1 byte smaller than the input string
4131 * (because we changed at least one 2-byte character into a
4132 * single byte), but that is eaten up by the trailing NUL */
4133 Newx(s, len, char);
4134
4135 {
4136 STRLEN i;
4137 STRLEN d= 0;
4138 char * e = (char *) t + len;
4139
4140 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4141 U8 cur_char = t[i];
4142 if (UTF8_IS_INVARIANT(cur_char)) {
4143 s[d++] = cur_char;
4144 }
4145 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4146 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4147 }
4148 else { /* Replace illegal cp with highest collating
4149 one */
4150 s[d++] = PL_strxfrm_max_cp;
4151 }
4152 }
4153 s[d++] = '\0';
4154 Renew(s, d, char); /* Free up unused space */
4155 }
4156 }
4157 }
4158
4159 /* Here, we have constructed a modified version of the input. It could
4160 * be that we already had a modified copy before we did this version.
4161 * If so, that copy is no longer needed */
4162 if (t != input_string) {
4163 Safefree(t);
4164 }
4165 }
4166
4167 length_in_chars = (utf8)
4168 ? utf8_length((U8 *) s, (U8 *) s + len)
4169 : len;
4170
4171 /* The first element in the output is the collation id, used by
4172 * sv_collxfrm(); then comes the space for the transformed string. The
4173 * equation should give us a good estimate as to how much is needed */
4174 xAlloc = COLLXFRM_HDR_LEN
4175 + PL_collxfrm_base
4176 + (PL_collxfrm_mult * length_in_chars);
4177 Newx(xbuf, xAlloc, char);
4178 if (UNLIKELY(! xbuf)) {
4179 DEBUG_L(PerlIO_printf(Perl_debug_log,
4180 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4181 goto bad;
4182 }
4183
4184 /* Store the collation id */
4185 *(U32*)xbuf = PL_collation_ix;
4186
4187 /* Then the transformation of the input. We loop until successful, or we
4188 * give up */
4189 for (;;) {
4190
4191 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4192
4193 /* If the transformed string occupies less space than we told strxfrm()
4194 * was available, it means it successfully transformed the whole
4195 * string. */
4196 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4197
4198 /* Some systems include a trailing NUL in the returned length.
4199 * Ignore it, using a loop in case multiple trailing NULs are
4200 * returned. */
4201 while ( (*xlen) > 0
4202 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4203 {
4204 (*xlen)--;
4205 }
4206
4207 /* If the first try didn't get it, it means our prediction was low.
4208 * Modify the coefficients so that we predict a larger value in any
4209 * future transformations */
4210 if (! first_time) {
4211 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4212 STRLEN computed_guess = PL_collxfrm_base
4213 + (PL_collxfrm_mult * length_in_chars);
4214
4215 /* On zero-length input, just keep current slope instead of
4216 * dividing by 0 */
4217 const STRLEN new_m = (length_in_chars != 0)
4218 ? needed / length_in_chars
4219 : PL_collxfrm_mult;
4220
4221 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4222 "%s: %d: initial size of %zu bytes for a length "
4223 "%zu string was insufficient, %zu needed\n",
4224 __FILE__, __LINE__,
4225 computed_guess, length_in_chars, needed));
4226
4227 /* If slope increased, use it, but discard this result for
4228 * length 1 strings, as we can't be sure that it's a real slope
4229 * change */
4230 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4231
4232 # ifdef DEBUGGING
4233
4234 STRLEN old_m = PL_collxfrm_mult;
4235 STRLEN old_b = PL_collxfrm_base;
4236
4237 # endif
4238
4239 PL_collxfrm_mult = new_m;
4240 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4241 computed_guess = PL_collxfrm_base
4242 + (PL_collxfrm_mult * length_in_chars);
4243 if (computed_guess < needed) {
4244 PL_collxfrm_base += needed - computed_guess;
4245 }
4246
4247 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4248 "%s: %d: slope is now %zu; was %zu, base "
4249 "is now %zu; was %zu\n",
4250 __FILE__, __LINE__,
4251 PL_collxfrm_mult, old_m,
4252 PL_collxfrm_base, old_b));
4253 }
4254 else { /* Slope didn't change, but 'b' did */
4255 const STRLEN new_b = needed
4256 - computed_guess
4257 + PL_collxfrm_base;
4258 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4259 "%s: %d: base is now %zu; was %zu\n",
4260 __FILE__, __LINE__,
4261 new_b, PL_collxfrm_base));
4262 PL_collxfrm_base = new_b;
4263 }
4264 }
4265
4266 break;
4267 }
4268
4269 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4270 DEBUG_L(PerlIO_printf(Perl_debug_log,
4271 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4272 *xlen, PERL_INT_MAX));
4273 goto bad;
4274 }
4275
4276 /* A well-behaved strxfrm() returns exactly how much space it needs
4277 * (usually not including the trailing NUL) when it fails due to not
4278 * enough space being provided. Assume that this is the case unless
4279 * it's been proven otherwise */
4280 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4281 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4282 }
4283 else { /* Here, either:
4284 * 1) The strxfrm() has previously shown bad behavior; or
4285 * 2) It isn't the first time through the loop, which means
4286 * that the strxfrm() is now showing bad behavior, because
4287 * we gave it what it said was needed in the previous
4288 * iteration, and it came back saying it needed still more.
4289 * (Many versions of cygwin fit this. When the buffer size
4290 * isn't sufficient, they return the input size instead of
4291 * how much is needed.)
4292 * Increase the buffer size by a fixed percentage and try again.
4293 * */
4294 xAlloc += (xAlloc / 4) + 1;
4295 PL_strxfrm_is_behaved = FALSE;
4296
4297 # ifdef DEBUGGING
4298
4299 if (DEBUG_Lv_TEST || debug_initialization) {
4300 PerlIO_printf(Perl_debug_log,
4301 "_mem_collxfrm required more space than previously calculated"
4302 " for locale %s, trying again with new guess=%zu+%zu\n",
4303 PL_collation_name, COLLXFRM_HDR_LEN,
4304 xAlloc - COLLXFRM_HDR_LEN);
4305 }
4306
4307 # endif
4308
4309 }
4310
4311 Renew(xbuf, xAlloc, char);
4312 if (UNLIKELY(! xbuf)) {
4313 DEBUG_L(PerlIO_printf(Perl_debug_log,
4314 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4315 goto bad;
4316 }
4317
4318 first_time = FALSE;
4319 }
4320
4321
4322 # ifdef DEBUGGING
4323
4324 if (DEBUG_Lv_TEST || debug_initialization) {
4325
4326 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4327 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4328 PerlIO_printf(Perl_debug_log, "%s\n",
4329 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4330 *xlen, 1));
4331 }
4332
4333 # endif
4334
4335 /* Free up unneeded space; retain ehough for trailing NUL */
4336 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4337
4338 if (s != input_string) {
4339 Safefree(s);
4340 }
4341
4342 return xbuf;
4343
4344 bad:
4345
4346 # ifdef DEBUGGING
4347
4348 if (DEBUG_Lv_TEST || debug_initialization) {
4349 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4350 }
4351
4352 # endif
4353
4354 Safefree(xbuf);
4355 if (s != input_string) {
4356 Safefree(s);
4357 }
4358 *xlen = 0;
4359
4360 return NULL;
4361 }
4362
4363 # ifdef DEBUGGING
4364
4365 STATIC void
4366 S_print_collxfrm_input_and_return(pTHX_
4367 const char * const s,
4368 const char * const e,
4369 const STRLEN * const xlen,
4370 const bool is_utf8)
4371 {
4372
4373 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4374
4375 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4376 (UV)PL_collation_ix);
4377 if (xlen) {
4378 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4379 }
4380 else {
4381 PerlIO_printf(Perl_debug_log, "NULL");
4382 }
4383 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4384 PL_collation_name);
4385 print_bytes_for_locale(s, e, is_utf8);
4386
4387 PerlIO_printf(Perl_debug_log, "'\n");
4388 }
4389
4390 # endif /* DEBUGGING */
4391 #endif /* USE_LOCALE_COLLATE */
4392 #ifdef USE_LOCALE
4393 # ifdef DEBUGGING
4394
4395 STATIC void
4396 S_print_bytes_for_locale(pTHX_
4397 const char * const s,
4398 const char * const e,
4399 const bool is_utf8)
4400 {
4401 const char * t = s;
4402 bool prev_was_printable = TRUE;
4403 bool first_time = TRUE;
4404
4405 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4406
4407 while (t < e) {
4408 UV cp = (is_utf8)
4409 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4410 : * (U8 *) t;
4411 if (isPRINT(cp)) {
4412 if (! prev_was_printable) {
4413 PerlIO_printf(Perl_debug_log, " ");
4414 }
4415 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4416 prev_was_printable = TRUE;
4417 }
4418 else {
4419 if (! first_time) {
4420 PerlIO_printf(Perl_debug_log, " ");
4421 }
4422 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4423 prev_was_printable = FALSE;
4424 }
4425 t += (is_utf8) ? UTF8SKIP(t) : 1;
4426 first_time = FALSE;
4427 }
4428 }
4429
4430 # endif /* #ifdef DEBUGGING */
4431
4432 STATIC const char *
4433 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4434 {
4435 /* Changes the locale for LC_'switch_category" to that of
4436 * LC_'template_category', if they aren't already the same. If not NULL,
4437 * 'template_locale' is the locale that 'template_category' is in.
4438 *
4439 * Returns a copy of the name of the original locale for 'switch_category'
4440 * so can be switched back to with the companion function
4441 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4442
4443 char * restore_to_locale = NULL;
4444
4445 if (switch_category == template_category) { /* No changes needed */
4446 return NULL;
4447 }
4448
4449 /* Find the original locale of the category we may need to change, so that
4450 * it can be restored to later */
4451 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4452 NULL)));
4453 if (! restore_to_locale) {
4454 Perl_croak(aTHX_
4455 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4456 __FILE__, __LINE__, category_name(switch_category), errno);
4457 }
4458
4459 /* If the locale of the template category wasn't passed in, find it now */
4460 if (template_locale == NULL) {
4461 template_locale = do_setlocale_r(template_category, NULL);
4462 if (! template_locale) {
4463 Perl_croak(aTHX_
4464 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4465 __FILE__, __LINE__, category_name(template_category), errno);
4466 }
4467 }
4468
4469 /* It the locales are the same, there's nothing to do */
4470 if (strEQ(restore_to_locale, template_locale)) {
4471 Safefree(restore_to_locale);
4472
4473 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4474 category_name(switch_category), template_locale));
4475
4476 return NULL;
4477 }
4478
4479 /* Finally, change the locale to the template one */
4480 if (! do_setlocale_r(switch_category, template_locale)) {
4481 Perl_croak(aTHX_
4482 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4483 __FILE__, __LINE__, category_name(switch_category),
4484 template_locale, errno);
4485 }
4486
4487 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4488 category_name(switch_category), template_locale));
4489
4490 return restore_to_locale;
4491 }
4492
4493 STATIC void
4494 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4495 {
4496 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4497 * copy that will be freed by this function), or do nothing if the latter
4498 * parameter is NULL */
4499
4500 if (original_locale == NULL) {
4501 return;
4502 }
4503
4504 if (! do_setlocale_r(category, original_locale)) {
4505 Perl_croak(aTHX_
4506 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4507 __FILE__, __LINE__,
4508 category_name(category), original_locale, errno);
4509 }
4510
4511 Safefree(original_locale);
4512 }
4513
4514 /* is_cur_LC_category_utf8 uses a small char buffer to avoid malloc/free */
4515 #define CUR_LC_BUFFER_SIZE 64
4516
4517 bool
4518 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4519 {
4520 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4521 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4522 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4523 * could give the wrong result. The result will very likely be correct for
4524 * languages that have commonly used non-ASCII characters, but for notably
4525 * English, it comes down to if the locale's name ends in something like
4526 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4527 *
4528 * If the platform is early C89, not containing mbtowc(), or we are
4529 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4530 * These work very well for non-Latin locales or those whose currency
4531 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4532 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4533 * on the name containing UTF-8 or not. */
4534
4535 /* Name of current locale corresponding to the input category */
4536 const char *save_input_locale = NULL;
4537
4538 bool is_utf8 = FALSE; /* The return value */
4539
4540 /* The variables below are for the cache of previous lookups using this
4541 * function. The cache is a C string, described at the definition for
4542 * 'C_and_POSIX_utf8ness'.
4543 *
4544 * The first part of the cache is fixed, for the C and POSIX locales. The
4545 * varying part starts just after them. */
4546 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4547
4548 Size_t utf8ness_cache_size; /* Size of the varying portion */
4549 Size_t input_name_len; /* Length in bytes of save_input_locale */
4550 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4551 the name in the cache */
4552 char * delimited; /* The name plus the delimiters used to store
4553 it in the cache */
4554 char buffer[CUR_LC_BUFFER_SIZE]; /* small buffer */
4555 char * name_pos; /* position of 'delimited' in the cache, or 0
4556 if not there */
4557
4558
4559 # ifdef LC_ALL
4560
4561 assert(category != LC_ALL);
4562
4563 # endif
4564
4565 /* Get the desired category's locale */
4566 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4567 if (! save_input_locale) {
4568 Perl_croak(aTHX_
4569 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4570 __FILE__, __LINE__, category_name(category), errno);
4571 }
4572
4573 DEBUG_L(PerlIO_printf(Perl_debug_log,
4574 "Current locale for %s is %s\n",
4575 category_name(category), save_input_locale));
4576
4577 input_name_len = strlen(save_input_locale);
4578
4579 /* In our cache, each name is accompanied by two delimiters and a single
4580 * utf8ness digit */
4581 input_name_len_with_overhead = input_name_len + 3;
4582
4583 if ( input_name_len_with_overhead <= CUR_LC_BUFFER_SIZE ) {
4584 /* we can use the buffer, avoid a malloc */
4585 delimited = buffer;
4586 } else { /* need a malloc */
4587 /* Allocate and populate space for a copy of the name surrounded by the
4588 * delimiters */
4589 Newx(delimited, input_name_len_with_overhead, char);
4590 }
4591
4592 delimited[0] = UTF8NESS_SEP[0];
4593 Copy(save_input_locale, delimited + 1, input_name_len, char);
4594 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4595 delimited[input_name_len+2] = '\0';
4596
4597 /* And see if that is in the cache */
4598 name_pos = instr(PL_locale_utf8ness, delimited);
4599 if (name_pos) {
4600 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4601
4602 # ifdef DEBUGGING
4603
4604 if (DEBUG_Lv_TEST || debug_initialization) {
4605 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4606 save_input_locale, is_utf8);
4607 }
4608
4609 # endif
4610
4611 /* And, if not already in that position, move it to the beginning of
4612 * the non-constant portion of the list, since it is the most recently
4613 * used. (We don't have to worry about overflow, since just moving
4614 * existing names around) */
4615 if (name_pos > utf8ness_cache) {
4616 Move(utf8ness_cache,
4617 utf8ness_cache + input_name_len_with_overhead,
4618 name_pos - utf8ness_cache, char);
4619 Copy(delimited,
4620 utf8ness_cache,
4621 input_name_len_with_overhead - 1, char);
4622 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4623 }
4624
4625 /* free only when not using the buffer */
4626 if ( delimited != buffer ) Safefree(delimited);
4627 Safefree(save_input_locale);
4628 return is_utf8;
4629 }
4630
4631 /* Here we don't have stored the utf8ness for the input locale. We have to
4632 * calculate it */
4633
4634 # if defined(USE_LOCALE_CTYPE) \
4635 && ( defined(HAS_NL_LANGINFO) \
4636 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4637
4638 {
4639 const char *original_ctype_locale
4640 = switch_category_locale_to_template(LC_CTYPE,
4641 category,
4642 save_input_locale);
4643
4644 /* Here the current LC_CTYPE is set to the locale of the category whose
4645 * information is desired. This means that nl_langinfo() and mbtowc()
4646 * should give the correct results */
4647
4648 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4649 calling the functions if we have this */
4650
4651 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4652 * Unicode code point. */
4653
4654 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4655 __FILE__, __LINE__, (int) MB_CUR_MAX));
4656 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
4657 is_utf8 = FALSE;
4658 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4659 goto finish_and_return;
4660 }
4661
4662 # endif
4663 # if defined(HAS_NL_LANGINFO)
4664
4665 { /* The task is easiest if the platform has this POSIX 2001 function.
4666 Except on some platforms it can wrongly return "", so have to have
4667 a fallback. And it can return that it's UTF-8, even if there are
4668 variances from that. For example, Turkish locales may use the
4669 alternate dotted I rules, and sometimes it appears to be a
4670 defective locale definition. XXX We should probably check for
4671 these in the Latin1 range and warn (but on glibc, requires
4672 iswalnum() etc. due to their not handling 80-FF correctly */
4673 const char *codeset = my_nl_langinfo(CODESET, FALSE);
4674 /* FALSE => already in dest locale */
4675
4676 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4677 "\tnllanginfo returned CODESET '%s'\n", codeset));
4678
4679 if (codeset && strNE(codeset, "")) {
4680
4681 /* If the implementation of foldEQ() somehow were
4682 * to change to not go byte-by-byte, this could
4683 * read past end of string, as only one length is
4684 * checked. But currently, a premature NUL will
4685 * compare false, and it will stop there */
4686 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
4687 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
4688
4689 DEBUG_L(PerlIO_printf(Perl_debug_log,
4690 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
4691 codeset, is_utf8));
4692 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4693 goto finish_and_return;
4694 }
4695 }
4696
4697 # endif
4698 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
4699 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
4700 * late adder to C89, so very likely to have it. However, testing has
4701 * shown that, like nl_langinfo() above, there are locales that are not
4702 * strictly UTF-8 that this will return that they are */
4703
4704 {
4705 wchar_t wc;
4706 int len;
4707 dSAVEDERRNO;
4708
4709 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4710
4711 mbstate_t ps;
4712
4713 # endif
4714
4715 /* mbrtowc() and mbtowc() convert a byte string to a wide
4716 * character. Feed a byte string to one of them and check that the
4717 * result is the expected Unicode code point */
4718
4719 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4720 /* Prefer this function if available, as it's reentrant */
4721
4722 memzero(&ps, sizeof(ps));;
4723 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
4724 state */
4725 SETERRNO(0, 0);
4726 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
4727 SAVE_ERRNO;
4728
4729 # else
4730
4731 MBTOWC_LOCK;
4732 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
4733 SETERRNO(0, 0);
4734 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
4735 SAVE_ERRNO;
4736 MBTOWC_UNLOCK;
4737
4738 # endif
4739
4740 RESTORE_ERRNO;
4741 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4742 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
4743 len, (unsigned int) wc, GET_ERRNO));
4744
4745 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
4746 && wc == (wchar_t) UNICODE_REPLACEMENT);
4747 }
4748
4749 # endif
4750
4751 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4752 goto finish_and_return;
4753 }
4754
4755 # else
4756
4757 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
4758 * try looking at the currency symbol to see if it disambiguates
4759 * things. Often that will be in the native script, and if the symbol
4760 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
4761 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
4762 * string being valid UTF-8 are quite small */
4763
4764 # ifdef USE_LOCALE_MONETARY
4765
4766 /* If have LC_MONETARY, we can look at the currency symbol. Often that
4767 * will be in the native script. We do this one first because there is
4768 * just one string to examine, so potentially avoids work */
4769
4770 {
4771 const char *original_monetary_locale
4772 = switch_category_locale_to_template(LC_MONETARY,
4773 category,
4774 save_input_locale);
4775 bool only_ascii = FALSE;
4776 const U8 * currency_string
4777 = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE);
4778 /* 2nd param not relevant for this item */
4779 const U8 * first_variant;
4780
4781 assert( *currency_string == '-'
4782 || *currency_string == '+'
4783 || *currency_string == '.');
4784
4785 currency_string++;
4786
4787 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
4788 {
4789 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));
4790 only_ascii = TRUE;
4791 }
4792 else {
4793 is_utf8 = is_strict_utf8_string(first_variant, 0);
4794 }
4795
4796 restore_switched_locale(LC_MONETARY, original_monetary_locale);
4797
4798 if (! only_ascii) {
4799
4800 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
4801 * otherwise assume the locale is UTF-8 if and only if the symbol
4802 * is non-ascii UTF-8. */
4803 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
4804 save_input_locale, is_utf8));
4805 goto finish_and_return;
4806 }
4807 }
4808
4809 # endif /* USE_LOCALE_MONETARY */
4810 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
4811
4812 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
4813 * the names of the months and weekdays, timezone, and am/pm indicator */
4814 {
4815 const char *original_time_locale
4816 = switch_category_locale_to_template(LC_TIME,
4817 category,
4818 save_input_locale);
4819 int hour = 10;
4820 bool is_dst = FALSE;
4821 int dom = 1;
4822 int month = 0;
4823 int i;
4824 char * formatted_time;
4825
4826 /* Here the current LC_TIME is set to the locale of the category
4827 * whose information is desired. Look at all the days of the week and
4828 * month names, and the timezone and am/pm indicator for UTF-8 variant
4829 * characters. The first such a one found will tell us if the locale
4830 * is UTF-8 or not */
4831
4832 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
4833 formatted_time = my_strftime("%A %B %Z %p",
4834 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
4835 if ( ! formatted_time
4836 || is_utf8_invariant_string((U8 *) formatted_time, 0))
4837 {
4838
4839 /* Here, we didn't find a non-ASCII. Try the next time through
4840 * with the complemented dst and am/pm, and try with the next
4841 * weekday. After we have gotten all weekdays, try the next
4842 * month */
4843 is_dst = ! is_dst;
4844 hour = (hour + 12) % 24;
4845 dom++;
4846 if (i > 6) {
4847 month++;
4848 }
4849 continue;
4850 }
4851
4852 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
4853 * false otherwise. But first, restore LC_TIME to its original
4854 * locale if we changed it */
4855 restore_switched_locale(LC_TIME, original_time_locale);
4856
4857 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
4858 save_input_locale,
4859 is_utf8_string((U8 *) formatted_time, 0)));
4860 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
4861 goto finish_and_return;
4862 }
4863
4864 /* Falling off the end of the loop indicates all the names were just
4865 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
4866 * to its original locale */
4867 restore_switched_locale(LC_TIME, original_time_locale);
4868 DEBUG_Lv(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));
4869 }
4870
4871 # endif
4872
4873 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
4874
4875 /* This code is ifdefd out because it was found to not be necessary in testing
4876 * on our dromedary test machine, which has over 700 locales. There, this
4877 * added no value to looking at the currency symbol and the time strings. I
4878 * left it in so as to avoid rewriting it if real-world experience indicates
4879 * that dromedary is an outlier. Essentially, instead of returning abpve if we
4880 * haven't found illegal utf8, we continue on and examine all the strerror()
4881 * messages on the platform for utf8ness. If all are ASCII, we still don't
4882 * know the answer; but otherwise we have a pretty good indication of the
4883 * utf8ness. The reason this doesn't help much is that the messages may not
4884 * have been translated into the locale. The currency symbol and time strings
4885 * are much more likely to have been translated. */
4886 {
4887 int e;
4888 bool non_ascii = FALSE;
4889 const char *original_messages_locale
4890 = switch_category_locale_to_template(LC_MESSAGES,
4891 category,
4892 save_input_locale);
4893 const char * errmsg = NULL;
4894
4895 /* Here the current LC_MESSAGES is set to the locale of the category
4896 * whose information is desired. Look through all the messages. We
4897 * can't use Strerror() here because it may expand to code that
4898 * segfaults in miniperl */
4899
4900 for (e = 0; e <= sys_nerr; e++) {
4901 errno = 0;
4902 errmsg = sys_errlist[e];
4903 if (errno || !errmsg) {
4904 break;
4905 }
4906 errmsg = savepv(errmsg);
4907 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
4908 non_ascii = TRUE;
4909 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
4910 break;
4911 }
4912 }
4913 Safefree(errmsg);
4914
4915 restore_switched_locale(LC_MESSAGES, original_messages_locale);
4916
4917 if (non_ascii) {
4918
4919 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
4920 * any non-ascii means it is one; otherwise we assume it isn't */
4921 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
4922 save_input_locale,
4923 is_utf8));
4924 goto finish_and_return;
4925 }
4926
4927 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));
4928 }
4929
4930 # endif
4931 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
4932 UTF-8 locale */
4933
4934 /* As a last resort, look at the locale name to see if it matches
4935 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
4936 * return of setlocale(), is actually defined to be opaque, so we can't
4937 * really rely on the absence of various substrings in the name to indicate
4938 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
4939 * be a UTF-8 locale. Similarly for the other common names */
4940
4941 {
4942 const Size_t final_pos = strlen(save_input_locale) - 1;
4943
4944 if (final_pos >= 3) {
4945 const char *name = save_input_locale;
4946
4947 /* Find next 'U' or 'u' and look from there */
4948 while ((name += strcspn(name, "Uu") + 1)
4949 <= save_input_locale + final_pos - 2)
4950 {
4951 if ( isALPHA_FOLD_NE(*name, 't')
4952 || isALPHA_FOLD_NE(*(name + 1), 'f'))
4953 {
4954 continue;
4955 }
4956 name += 2;
4957 if (*(name) == '-') {
4958 if ((name > save_input_locale + final_pos - 1)) {
4959 break;
4960 }
4961 name++;
4962 }
4963 if (*(name) == '8') {
4964 DEBUG_L(PerlIO_printf(Perl_debug_log,
4965 "Locale %s ends with UTF-8 in name\n",
4966 save_input_locale));
4967 is_utf8 = TRUE;
4968 goto finish_and_return;
4969 }
4970 }
4971 DEBUG_L(PerlIO_printf(Perl_debug_log,
4972 "Locale %s doesn't end with UTF-8 in name\n",
4973 save_input_locale));
4974 }
4975
4976 # ifdef WIN32
4977
4978 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
4979 if (memENDs(save_input_locale, final_pos, "65001")) {
4980 DEBUG_L(PerlIO_printf(Perl_debug_log,
4981 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
4982 save_input_locale));
4983 is_utf8 = TRUE;
4984 goto finish_and_return;
4985 }
4986
4987 # endif
4988 }
4989 # endif
4990
4991 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
4992 * since we are about to return FALSE anyway, there is no point in doing
4993 * this extra work */
4994
4995 # if 0
4996 if (instr(save_input_locale, "8859")) {
4997 DEBUG_L(PerlIO_printf(Perl_debug_log,
4998 "Locale %s has 8859 in name, not UTF-8 locale\n",
4999 save_input_locale));
5000 is_utf8 = FALSE;
5001 goto finish_and_return;
5002 }
5003 # endif
5004
5005 DEBUG_L(PerlIO_printf(Perl_debug_log,
5006 "Assuming locale %s is not a UTF-8 locale\n",
5007 save_input_locale));
5008 is_utf8 = FALSE;
5009
5010 # endif /* the code that is compiled when no modern LC_CTYPE */
5011
5012 finish_and_return:
5013
5014 /* Cache this result so we don't have to go through all this next time. */
5015 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
5016 - (utf8ness_cache - PL_locale_utf8ness);
5017
5018 /* But we can't save it if it is too large for the total space available */
5019 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
5020 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
5021
5022 /* Here it can fit, but we may need to clear out the oldest cached
5023 * result(s) to do so. Check */
5024 if (utf8ness_cache_len + input_name_len_with_overhead
5025 >= utf8ness_cache_size)
5026 {
5027 /* Here we have to clear something out to make room for this.
5028 * Start looking at the rightmost place where it could fit and find
5029 * the beginning of the entry that extends past that. */
5030 char * cutoff = (char *) my_memrchr(utf8ness_cache,
5031 UTF8NESS_SEP[0],
5032 utf8ness_cache_size
5033 - input_name_len_with_overhead);
5034
5035 assert(cutoff);
5036 assert(cutoff >= utf8ness_cache);
5037
5038 /* This and all subsequent entries must be removed */
5039 *cutoff = '\0';
5040 utf8ness_cache_len = strlen(utf8ness_cache);
5041 }
5042
5043 /* Make space for the new entry */
5044 Move(utf8ness_cache,
5045 utf8ness_cache + input_name_len_with_overhead,
5046 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
5047
5048 /* And insert it */
5049 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
5050 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
5051
5052 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1] & ~1) != '0') {
5053 Perl_croak(aTHX_
5054 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu,"
5055 " inserted_name=%s, its_len=%zu\n",
5056 __FILE__, __LINE__,
5057 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
5058 delimited, input_name_len_with_overhead);
5059 }
5060 }
5061
5062 # ifdef DEBUGGING
5063
5064 if (DEBUG_Lv_TEST) {
5065 const char * s = PL_locale_utf8ness;
5066
5067 /* Audit the structure */
5068 while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) {
5069 const char *e;
5070
5071 if (*s != UTF8NESS_SEP[0]) {
5072 Perl_croak(aTHX_
5073 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5074 " separator %.*s<-- HERE %s\n",
5075 __FILE__, __LINE__,
5076 (int) (s - PL_locale_utf8ness), PL_locale_utf8ness,
5077 s);
5078 }
5079 s++;
5080 e = strchr(s, UTF8NESS_PREFIX[0]);
5081 if (! e) {
5082 e = PL_locale_utf8ness + strlen(PL_locale_utf8ness);
5083 Perl_croak(aTHX_
5084 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5085 " separator %.*s<-- HERE %s\n",
5086 __FILE__, __LINE__,
5087 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5088 e);
5089 }
5090 e++;
5091 if (*e != '0' && *e != '1') {
5092 Perl_croak(aTHX_
5093 "panic: %s: %d: Corrupt utf8ness_cache: utf8ness"
5094 " must be [01] %.*s<-- HERE %s\n",
5095 __FILE__, __LINE__,
5096 (int) (e + 1 - PL_locale_utf8ness),
5097 PL_locale_utf8ness, e + 1);
5098 }
5099 if (ninstr(PL_locale_utf8ness, s, s-1, e)) {
5100 Perl_croak(aTHX_
5101 "panic: %s: %d: Corrupt utf8ness_cache: entry"
5102 " has duplicate %.*s<-- HERE %s\n",
5103 __FILE__, __LINE__,
5104 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5105 e);
5106 }
5107 s = e + 1;
5108 }
5109 }
5110
5111 if (DEBUG_Lv_TEST || debug_initialization) {
5112
5113 PerlIO_printf(Perl_debug_log,
5114 "PL_locale_utf8ness is now %s; returning %d\n",
5115 PL_locale_utf8ness, is_utf8);
5116 }
5117
5118 # endif
5119
5120 /* free only when not using the buffer */
5121 if ( delimited != buffer ) Safefree(delimited);
5122 Safefree(save_input_locale);
5123 return is_utf8;
5124 }
5125
5126 #endif
5127
5128 bool
5129 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
5130 {
5131 /* Internal function which returns if we are in the scope of a pragma that
5132 * enables the locale category 'category'. 'compiling' should indicate if
5133 * this is during the compilation phase (TRUE) or not (FALSE). */
5134
5135 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
5136
5137 SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0);
5138 if (! these_categories || these_categories == &PL_sv_placeholder) {
5139 return FALSE;
5140 }
5141
5142 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
5143 * a valid unsigned */
5144 assert(category >= -1);
5145 return cBOOL(SvUV(these_categories) & (1U << (category + 1)));
5146 }
5147
5148 char *
5149 Perl_my_strerror(pTHX_ const int errnum)
5150 {
5151 /* Returns a mortalized copy of the text of the error message associated
5152 * with 'errnum'. It uses the current locale's text unless the platform
5153 * doesn't have the LC_MESSAGES category or we are not being called from
5154 * within the scope of 'use locale'. In the former case, it uses whatever
5155 * strerror returns; in the latter case it uses the text from the C locale.
5156 *
5157 * The function just calls strerror(), but temporarily switches, if needed,
5158 * to the C locale */
5159
5160 char *errstr;
5161
5162 #ifndef USE_LOCALE_MESSAGES
5163
5164 /* If platform doesn't have messages category, we don't do any switching to
5165 * the C locale; we just use whatever strerror() returns */
5166
5167 errstr = savepv(Strerror(errnum));
5168
5169 #else /* Has locale messages */
5170
5171 const bool within_locale_scope = IN_LC(LC_MESSAGES);
5172
5173 # ifndef USE_ITHREADS
5174
5175 /* This function is trivial without threads. */
5176 if (within_locale_scope) {
5177 errstr = savepv(strerror(errnum));
5178 }
5179 else {
5180 const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL));
5181
5182 do_setlocale_c(LC_MESSAGES, "C");
5183 errstr = savepv(strerror(errnum));
5184 do_setlocale_c(LC_MESSAGES, save_locale);
5185 Safefree(save_locale);
5186 }
5187
5188 # elif defined(USE_POSIX_2008_LOCALE) \
5189 && defined(HAS_STRERROR_L)
5190
5191 /* This function is also trivial if we don't have to worry about thread
5192 * safety and have strerror_l(), as it handles the switch of locales so we
5193 * don't have to deal with that. We don't have to worry about thread
5194 * safety if strerror_r() is also available. Both it and strerror_l() are
5195 * thread-safe. Plain strerror() isn't thread safe. But on threaded
5196 * builds when strerror_r() is available, the apparent call to strerror()
5197 * below is actually a macro that behind-the-scenes calls strerror_r(). */
5198
5199 # ifdef HAS_STRERROR_R
5200
5201 if (within_locale_scope) {
5202 errstr = savepv(strerror(errnum));
5203 }
5204 else {
5205 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
5206 }
5207
5208 # else
5209
5210 /* Here we have strerror_l(), but not strerror_r() and we are on a
5211 * threaded-build. We use strerror_l() for everything, constructing a
5212 * locale to pass to it if necessary */
5213
5214 bool do_free = FALSE;
5215 locale_t locale_to_use;
5216
5217 if (within_locale_scope) {
5218 locale_to_use = uselocale((locale_t) 0);
5219 if (locale_to_use == LC_GLOBAL_LOCALE) {
5220 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
5221 do_free = TRUE;
5222 }
5223 }
5224 else { /* Use C locale if not within 'use locale' scope */
5225 locale_to_use = PL_C_locale_obj;
5226 }
5227
5228 errstr = savepv(strerror_l(errnum, locale_to_use));
5229
5230 if (do_free) {
5231 freelocale(locale_to_use);
5232 }
5233
5234 # endif
5235 # else /* Doesn't have strerror_l() */
5236
5237 const char * save_locale = NULL;
5238 bool locale_is_C = FALSE;
5239
5240 /* We have a critical section to prevent another thread from executing this
5241 * same code at the same time. (On thread-safe perls, the LOCK is a
5242 * no-op.) Since this is the only place in core that changes LC_MESSAGES
5243 * (unless the user has called setlocale(), this works to prevent races. */
5244 SETLOCALE_LOCK;
5245
5246 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5247 "my_strerror called with errnum %d\n", errnum));
5248 if (! within_locale_scope) {
5249 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
5250 if (! save_locale) {
5251 SETLOCALE_UNLOCK;
5252 Perl_croak(aTHX_
5253 "panic: %s: %d: Could not find current LC_MESSAGES locale,"
5254 " errno=%d\n", __FILE__, __LINE__, errno);
5255 }
5256 else {
5257 locale_is_C = isNAME_C_OR_POSIX(save_locale);
5258
5259 /* Switch to the C locale if not already in it */
5260 if (! locale_is_C) {
5261
5262 /* The setlocale() just below likely will zap 'save_locale', so
5263 * create a copy. */
5264 save_locale = savepv(save_locale);
5265 if (! do_setlocale_c(LC_MESSAGES, "C")) {
5266
5267 /* If, for some reason, the locale change failed, we
5268 * soldier on as best as possible under the circumstances,
5269 * using the current locale, and clear save_locale, so we
5270 * don't try to change back. On z/0S, all setlocale()
5271 * calls fail after you've created a thread. This is their
5272 * way of making sure the entire process is always a single
5273 * locale. This means that 'use locale' is always in place
5274 * for messages under these circumstances. */
5275 Safefree(save_locale);
5276 save_locale = NULL;
5277 }
5278 }
5279 }
5280 } /* end of ! within_locale_scope */
5281 else {
5282 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
5283 __FILE__, __LINE__));
5284 }
5285
5286 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5287 "Any locale change has been done; about to call Strerror\n"));
5288 errstr = savepv(Strerror(errnum));
5289
5290 if (! within_locale_scope) {
5291 if (save_locale && ! locale_is_C) {
5292 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
5293 SETLOCALE_UNLOCK;
5294 Perl_croak(aTHX_
5295 "panic: %s: %d: setlocale restore to '%s' failed, errno=%d\n",
5296 __FILE__, __LINE__, save_locale, errno);
5297 }
5298 Safefree(save_locale);
5299 }
5300 }
5301
5302 SETLOCALE_UNLOCK;
5303
5304 # endif /* End of doesn't have strerror_l */
5305 # ifdef DEBUGGING
5306
5307 if (DEBUG_Lv_TEST) {
5308 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
5309 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
5310 PerlIO_printf(Perl_debug_log, "'\n");
5311 }
5312
5313 # endif
5314 #endif /* End of does have locale messages */
5315
5316 SAVEFREEPV(errstr);
5317 return errstr;
5318 }
5319
5320 /*
5321
5322 =for apidoc switch_to_global_locale
5323
5324 On systems without locale support, or on typical single-threaded builds, or on
5325 platforms that do not support per-thread locale operations, this function does
5326 nothing. On such systems that do have locale support, only a locale global to
5327 the whole program is available.
5328
5329 On multi-threaded builds on systems that do have per-thread locale operations,
5330 this function converts the thread it is running in to use the global locale.
5331 This is for code that has not yet or cannot be updated to handle multi-threaded
5332 locale operation. As long as only a single thread is so-converted, everything
5333 works fine, as all the other threads continue to ignore the global one, so only
5334 this thread looks at it.
5335
5336 However, on Windows systems this isn't quite true prior to Visual Studio 15,
5337 at which point Microsoft fixed a bug. A race can occur if you use the
5338 following operations on earlier Windows platforms:
5339
5340 =over
5341
5342 =item L<POSIX::localeconv|POSIX/localeconv>
5343
5344 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5345
5346 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5347
5348 =back
5349
5350 The first item is not fixable (except by upgrading to a later Visual Studio
5351 release), but it would be possible to work around the latter two items by using
5352 the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches
5353 welcome.
5354
5355 Without this function call, threads that use the L<C<setlocale(3)>> system
5356 function will not work properly, as all the locale-sensitive functions will
5357 look at the per-thread locale, and C<setlocale> will have no effect on this
5358 thread.
5359
5360 Perl code should convert to either call
5361 L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system
5362 C<setlocale>) or use the methods given in L<perlcall> to call
5363 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly
5364 handle all cases of single- vs multi-thread, POSIX 2008-supported or not.
5365
5366 Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can
5367 continue to work if this function is called before transferring control to the
5368 library.
5369
5370 Upon return from the code that needs to use the global locale,
5371 L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe
5372 multi-thread operation.
5373
5374 =cut
5375 */
5376
5377 void
5378 Perl_switch_to_global_locale()
5379 {
5380
5381 #ifdef USE_THREAD_SAFE_LOCALE
5382 # ifdef WIN32
5383
5384 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
5385
5386 # else
5387 # ifdef HAS_QUERYLOCALE
5388
5389 setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0)));
5390
5391 # else
5392
5393 {
5394 unsigned int i;
5395
5396 for (i = 0; i < LC_ALL_INDEX; i++) {
5397 setlocale(categories[i], do_setlocale_r(categories[i], NULL));
5398 }
5399 }
5400
5401 # endif
5402
5403 uselocale(LC_GLOBAL_LOCALE);
5404
5405 # endif
5406 #endif
5407
5408 }
5409
5410 /*
5411
5412 =for apidoc sync_locale
5413
5414 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or
5415 change the locale (though changing the locale is antisocial and dangerous on
5416 multi-threaded systems that don't have multi-thread safe locale operations.
5417 (See L<perllocale/Multi-threaded operation>). Using the system
5418 L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries
5419 called from XS, such as C<Gtk> do so, and this can't be changed. When the
5420 locale is changed by XS code that didn't use
5421 L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the
5422 locale has changed. Use this function to do so, before returning to Perl.
5423
5424 The return value is a boolean: TRUE if the global locale at the time of call
5425 was in effect; and FALSE if a per-thread locale was in effect. This can be
5426 used by the caller that needs to restore things as-they-were to decide whether
5427 or not to call
5428 L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>.
5429
5430 =cut
5431 */
5432
5433 bool
5434 Perl_sync_locale()
5435 {
5436
5437 #ifndef USE_LOCALE
5438
5439 return TRUE;
5440
5441 #else
5442
5443 const char * newlocale;
5444 dTHX;
5445
5446 # ifdef USE_POSIX_2008_LOCALE
5447
5448 bool was_in_global_locale = FALSE;
5449 locale_t cur_obj = uselocale((locale_t) 0);
5450
5451 /* On Windows, unless the foreign code has turned off the thread-safe
5452 * locale setting, any plain setlocale() will have affected what we see, so
5453 * no need to worry. Otherwise, If the foreign code has done a plain
5454 * setlocale(), it will only affect the global locale on POSIX systems, but
5455 * will affect the */
5456 if (cur_obj == LC_GLOBAL_LOCALE) {
5457
5458 # ifdef HAS_QUERY_LOCALE
5459
5460 do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL));
5461
5462 # else
5463
5464 unsigned int i;
5465
5466 /* We can't trust that we can read the LC_ALL format on the
5467 * platform, so do them individually */
5468 for (i = 0; i < LC_ALL_INDEX; i++) {
5469 do_setlocale_r(categories[i], setlocale(categories[i], NULL));
5470 }
5471
5472 # endif
5473
5474 was_in_global_locale = TRUE;
5475 }
5476
5477 # else
5478
5479 bool was_in_global_locale = TRUE;
5480
5481 # endif
5482 # ifdef USE_LOCALE_CTYPE
5483
5484 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
5485 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5486 "%s:%d: %s\n", __FILE__, __LINE__,
5487 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
5488 new_ctype(newlocale);
5489 Safefree(newlocale);
5490
5491 # endif /* USE_LOCALE_CTYPE */
5492 # ifdef USE_LOCALE_COLLATE
5493
5494 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
5495 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5496 "%s:%d: %s\n", __FILE__, __LINE__,
5497 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
5498 new_collate(newlocale);
5499 Safefree(newlocale);
5500
5501 # endif
5502 # ifdef USE_LOCALE_NUMERIC
5503
5504 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
5505 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5506 "%s:%d: %s\n", __FILE__, __LINE__,
5507 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
5508 new_numeric(newlocale);
5509 Safefree(newlocale);
5510
5511 # endif /* USE_LOCALE_NUMERIC */
5512
5513 return was_in_global_locale;
5514
5515 #endif
5516
5517 }
5518
5519 #if defined(DEBUGGING) && defined(USE_LOCALE)
5520
5521 STATIC char *
5522 S_setlocale_debug_string(const int category, /* category number,
5523 like LC_ALL */
5524 const char* const locale, /* locale name */
5525
5526 /* return value from setlocale() when attempting to
5527 * set 'category' to 'locale' */
5528 const char* const retval)
5529 {
5530 /* Returns a pointer to a NUL-terminated string in static storage with
5531 * added text about the info passed in. This is not thread safe and will
5532 * be overwritten by the next call, so this should be used just to
5533 * formulate a string to immediately print or savepv() on. */
5534
5535 static char ret[256];
5536
5537 my_strlcpy(ret, "setlocale(", sizeof(ret));
5538 my_strlcat(ret, category_name(category), sizeof(ret));
5539 my_strlcat(ret, ", ", sizeof(ret));
5540
5541 if (locale) {
5542 my_strlcat(ret, "\"", sizeof(ret));
5543 my_strlcat(ret, locale, sizeof(ret));
5544 my_strlcat(ret, "\"", sizeof(ret));
5545 }
5546 else {
5547 my_strlcat(ret, "NULL", sizeof(ret));
5548 }
5549
5550 my_strlcat(ret, ") returned ", sizeof(ret));
5551
5552 if (retval) {
5553 my_strlcat(ret, "\"", sizeof(ret));
5554 my_strlcat(ret, retval, sizeof(ret));
5555 my_strlcat(ret, "\"", sizeof(ret));
5556 }
5557 else {
5558 my_strlcat(ret, "NULL", sizeof(ret));
5559 }
5560
5561 assert(strlen(ret) < sizeof(ret));
5562
5563 return ret;
5564 }
5565
5566 #endif
5567
5568 void
5569 Perl_thread_locale_init()
5570 {
5571 /* Called from a thread on startup*/
5572
5573 #ifdef USE_THREAD_SAFE_LOCALE
5574
5575 dTHX_DEBUGGING;
5576
5577 /* C starts the new thread in the global C locale. If we are thread-safe,
5578 * we want to not be in the global locale */
5579
5580 DEBUG_L(PerlIO_printf(Perl_debug_log,
5581 "%s:%d: new thread, initial locale is %s; calling setlocale\n",
5582 __FILE__, __LINE__, setlocale(LC_ALL, NULL)));
5583
5584 # ifdef WIN32
5585
5586 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
5587
5588 # else
5589
5590 Perl_setlocale(LC_ALL, "C");
5591
5592 # endif
5593 #endif
5594
5595 }
5596
5597 void
5598 Perl_thread_locale_term()
5599 {
5600 /* Called from a thread as it gets ready to terminate */
5601
5602 #ifdef USE_THREAD_SAFE_LOCALE
5603
5604 /* C starts the new thread in the global C locale. If we are thread-safe,
5605 * we want to not be in the global locale */
5606
5607 # ifndef WIN32
5608
5609 { /* Free up */
5610 locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE);
5611 if (cur_obj != LC_GLOBAL_LOCALE && cur_obj != PL_C_locale_obj) {
5612 freelocale(cur_obj);
5613 }
5614 }
5615
5616 # endif
5617 #endif
5618
5619 }
5620
5621 /*
5622 * ex: set ts=8 sts=4 sw=4 et:
5623 */
5624