Commit | Line | Data |
---|---|---|
98994639 HS |
1 | /* locale.c |
2 | * | |
1129b882 NC |
3 | * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, |
4 | * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others | |
98994639 HS |
5 | * |
6 | * You may distribute under the terms of either the GNU General Public | |
7 | * License or the Artistic License, as specified in the README file. | |
8 | * | |
9 | */ | |
10 | ||
11 | /* | |
4ac71550 | 12 | * A Elbereth Gilthoniel, |
cdad3b53 | 13 | * silivren penna míriel |
4ac71550 | 14 | * o menel aglar elenath! |
cdad3b53 | 15 | * Na-chaered palan-díriel |
4ac71550 TC |
16 | * o galadhremmin ennorath, |
17 | * Fanuilos, le linnathon | |
18 | * nef aear, si nef aearon! | |
19 | * | |
20 | * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"] | |
98994639 HS |
21 | */ |
22 | ||
166f8a29 DM |
23 | /* utility functions for handling locale-specific stuff like what |
24 | * character represents the decimal point. | |
0d071d52 | 25 | * |
7d4bcc4a KW |
26 | * All C programs have an underlying locale. Perl code generally doesn't pay |
27 | * any attention to it except within the scope of a 'use locale'. For most | |
0d071d52 KW |
28 | * categories, it accomplishes this by just using different operations if it is |
29 | * in such scope than if not. However, various libc functions called by Perl | |
30 | * are affected by the LC_NUMERIC category, so there are macros in perl.h that | |
31 | * are used to toggle between the current locale and the C locale depending on | |
a9ad02a8 KW |
32 | * the desired behavior of those functions at the moment. And, LC_MESSAGES is |
33 | * switched to the C locale for outputting the message unless within the scope | |
34 | * of 'use locale'. | |
e9bc6d6b KW |
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. | |
166f8a29 DM |
44 | */ |
45 | ||
98994639 HS |
46 | #include "EXTERN.h" |
47 | #define PERL_IN_LOCALE_C | |
f7416781 | 48 | #include "perl_langinfo.h" |
98994639 HS |
49 | #include "perl.h" |
50 | ||
a4af207c JH |
51 | #include "reentr.h" |
52 | ||
0dec74cd KW |
53 | #ifdef I_WCHAR |
54 | # include <wchar.h> | |
55 | #endif | |
5b64f24c KW |
56 | #ifdef I_WCTYPE |
57 | # include <wctype.h> | |
58 | #endif | |
0dec74cd | 59 | |
2fcc0ca9 KW |
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 */ | |
5a4b0634 KW |
63 | #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT) |
64 | # define debug_initialization 0 | |
65 | # define DEBUG_INITIALIZATION_set(v) | |
66 | #else | |
2fcc0ca9 | 67 | static bool debug_initialization = FALSE; |
5a4b0634 | 68 | # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v) |
2fcc0ca9 KW |
69 | #endif |
70 | ||
48015184 KW |
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 | ||
291a84fb KW |
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 */ | |
ff1b739b KW |
80 | #define STRLENs(s) (sizeof("" s "") - 1) |
81 | ||
63e5b0d7 KW |
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 | ||
8ef6e574 KW |
97 | #ifdef USE_LOCALE |
98 | ||
47280b20 KW |
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: | |
98994639 | 106 | * |
47280b20 | 107 | * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0 |
98994639 | 108 | * |
47280b20 KW |
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 | ||
98994639 HS |
122 | STATIC char * |
123 | S_stdize_locale(pTHX_ char *locs) | |
124 | { | |
47280b20 KW |
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 | ||
7452cf6a | 139 | const char * const s = strchr(locs, '='); |
98994639 HS |
140 | bool okay = TRUE; |
141 | ||
7918f24d NC |
142 | PERL_ARGS_ASSERT_STDIZE_LOCALE; |
143 | ||
8772537c AL |
144 | if (s) { |
145 | const char * const t = strchr(s, '.'); | |
98994639 | 146 | okay = FALSE; |
8772537c AL |
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; | |
98994639 HS |
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 | ||
e5f10d49 KW |
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 | |
9821811f KW |
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 | |
e5f10d49 KW |
203 | # ifdef LC_ALL |
204 | LC_ALL, | |
205 | # endif | |
206 | -1 /* Placeholder because C doesn't allow a | |
207 | trailing comma, and it would get complicated | |
208 | with all the #ifdef's */ | |
209 | }; | |
210 | ||
211 | /* The top-most real element is LC_ALL */ | |
212 | ||
4ef8bdf9 | 213 | const char * const category_names[] = { |
e5f10d49 KW |
214 | |
215 | # ifdef USE_LOCALE_NUMERIC | |
216 | "LC_NUMERIC", | |
217 | # endif | |
218 | # ifdef USE_LOCALE_CTYPE | |
219 | "LC_CTYPE", | |
220 | # endif | |
221 | # ifdef USE_LOCALE_COLLATE | |
222 | "LC_COLLATE", | |
223 | # endif | |
224 | # ifdef USE_LOCALE_TIME | |
225 | "LC_TIME", | |
226 | # endif | |
227 | # ifdef USE_LOCALE_MESSAGES | |
228 | "LC_MESSAGES", | |
229 | # endif | |
230 | # ifdef USE_LOCALE_MONETARY | |
231 | "LC_MONETARY", | |
232 | # endif | |
9821811f KW |
233 | # ifdef USE_LOCALE_ADDRESS |
234 | "LC_ADDRESS", | |
235 | # endif | |
236 | # ifdef USE_LOCALE_IDENTIFICATION | |
237 | "LC_IDENTIFICATION", | |
238 | # endif | |
239 | # ifdef USE_LOCALE_MEASUREMENT | |
240 | "LC_MEASUREMENT", | |
241 | # endif | |
242 | # ifdef USE_LOCALE_PAPER | |
243 | "LC_PAPER", | |
244 | # endif | |
245 | # ifdef USE_LOCALE_TELEPHONE | |
246 | "LC_TELEPHONE", | |
247 | # endif | |
e5f10d49 KW |
248 | # ifdef LC_ALL |
249 | "LC_ALL", | |
250 | # endif | |
251 | NULL /* Placeholder */ | |
252 | }; | |
253 | ||
254 | # ifdef LC_ALL | |
255 | ||
256 | /* On systems with LC_ALL, it is kept in the highest index position. (-2 | |
257 | * to account for the final unused placeholder element.) */ | |
258 | # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2) | |
259 | ||
260 | # else | |
261 | ||
262 | /* On systems without LC_ALL, we pretend it is there, one beyond the real | |
263 | * top element, hence in the unused placeholder element. */ | |
264 | # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1) | |
265 | ||
266 | # endif | |
267 | ||
268 | /* Pretending there is an LC_ALL element just above allows us to avoid most | |
269 | * special cases. Most loops through these arrays in the code below are | |
270 | * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work | |
271 | * on either type of system. But the code must be written to not access the | |
948523db KW |
272 | * element at 'LC_ALL_INDEX' except on platforms that have it. This can be |
273 | * checked for at compile time by using the #define LC_ALL_INDEX which is only | |
274 | * defined if we do have LC_ALL. */ | |
e5f10d49 | 275 | |
b09aaf40 KW |
276 | STATIC const char * |
277 | S_category_name(const int category) | |
278 | { | |
279 | unsigned int i; | |
280 | ||
281 | #ifdef LC_ALL | |
282 | ||
283 | if (category == LC_ALL) { | |
284 | return "LC_ALL"; | |
285 | } | |
286 | ||
287 | #endif | |
288 | ||
289 | for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) { | |
290 | if (category == categories[i]) { | |
291 | return category_names[i]; | |
292 | } | |
293 | } | |
294 | ||
295 | { | |
296 | const char suffix[] = " (unknown)"; | |
297 | int temp = category; | |
298 | Size_t length = sizeof(suffix) + 1; | |
299 | char * unknown; | |
300 | dTHX; | |
301 | ||
302 | if (temp < 0) { | |
303 | length++; | |
304 | temp = - temp; | |
305 | } | |
306 | ||
307 | /* Calculate the number of digits */ | |
308 | while (temp >= 10) { | |
309 | temp /= 10; | |
310 | length++; | |
311 | } | |
312 | ||
313 | Newx(unknown, length, char); | |
314 | my_snprintf(unknown, length, "%d%s", category, suffix); | |
315 | SAVEFREEPV(unknown); | |
316 | return unknown; | |
317 | } | |
318 | } | |
319 | ||
948523db KW |
320 | /* Now create LC_foo_INDEX #defines for just those categories on this system */ |
321 | # ifdef USE_LOCALE_NUMERIC | |
322 | # define LC_NUMERIC_INDEX 0 | |
323 | # define _DUMMY_NUMERIC LC_NUMERIC_INDEX | |
324 | # else | |
325 | # define _DUMMY_NUMERIC -1 | |
326 | # endif | |
327 | # ifdef USE_LOCALE_CTYPE | |
328 | # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1 | |
329 | # define _DUMMY_CTYPE LC_CTYPE_INDEX | |
330 | # else | |
331 | # define _DUMMY_CTYPE _DUMMY_NUMERIC | |
332 | # endif | |
333 | # ifdef USE_LOCALE_COLLATE | |
334 | # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1 | |
335 | # define _DUMMY_COLLATE LC_COLLATE_INDEX | |
336 | # else | |
8203b3c3 | 337 | # define _DUMMY_COLLATE _DUMMY_CTYPE |
948523db KW |
338 | # endif |
339 | # ifdef USE_LOCALE_TIME | |
340 | # define LC_TIME_INDEX _DUMMY_COLLATE + 1 | |
341 | # define _DUMMY_TIME LC_TIME_INDEX | |
342 | # else | |
343 | # define _DUMMY_TIME _DUMMY_COLLATE | |
344 | # endif | |
345 | # ifdef USE_LOCALE_MESSAGES | |
346 | # define LC_MESSAGES_INDEX _DUMMY_TIME + 1 | |
347 | # define _DUMMY_MESSAGES LC_MESSAGES_INDEX | |
348 | # else | |
349 | # define _DUMMY_MESSAGES _DUMMY_TIME | |
350 | # endif | |
351 | # ifdef USE_LOCALE_MONETARY | |
352 | # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1 | |
353 | # define _DUMMY_MONETARY LC_MONETARY_INDEX | |
354 | # else | |
355 | # define _DUMMY_MONETARY _DUMMY_MESSAGES | |
356 | # endif | |
9821811f KW |
357 | # ifdef USE_LOCALE_ADDRESS |
358 | # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1 | |
359 | # define _DUMMY_ADDRESS LC_ADDRESS_INDEX | |
360 | # else | |
361 | # define _DUMMY_ADDRESS _DUMMY_MONETARY | |
362 | # endif | |
363 | # ifdef USE_LOCALE_IDENTIFICATION | |
364 | # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1 | |
365 | # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX | |
366 | # else | |
367 | # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS | |
368 | # endif | |
369 | # ifdef USE_LOCALE_MEASUREMENT | |
370 | # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1 | |
371 | # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX | |
372 | # else | |
373 | # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION | |
374 | # endif | |
375 | # ifdef USE_LOCALE_PAPER | |
376 | # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1 | |
377 | # define _DUMMY_PAPER LC_PAPER_INDEX | |
378 | # else | |
379 | # define _DUMMY_PAPER _DUMMY_MEASUREMENT | |
380 | # endif | |
381 | # ifdef USE_LOCALE_TELEPHONE | |
382 | # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1 | |
383 | # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX | |
384 | # else | |
385 | # define _DUMMY_TELEPHONE _DUMMY_PAPER | |
386 | # endif | |
948523db | 387 | # ifdef LC_ALL |
9821811f | 388 | # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1 |
948523db KW |
389 | # endif |
390 | #endif /* ifdef USE_LOCALE */ | |
8ef6e574 | 391 | |
d2b24094 | 392 | /* Windows requres a customized base-level setlocale() */ |
e9bc6d6b KW |
393 | #ifdef WIN32 |
394 | # define my_setlocale(cat, locale) win32_setlocale(cat, locale) | |
395 | #else | |
396 | # define my_setlocale(cat, locale) setlocale(cat, locale) | |
397 | #endif | |
398 | ||
399 | #ifndef USE_POSIX_2008_LOCALE | |
400 | ||
401 | /* "do_setlocale_c" is intended to be called when the category is a constant | |
402 | * known at compile time; "do_setlocale_r", not known until run time */ | |
403 | # define do_setlocale_c(cat, locale) my_setlocale(cat, locale) | |
404 | # define do_setlocale_r(cat, locale) my_setlocale(cat, locale) | |
405 | ||
406 | #else /* Below uses POSIX 2008 */ | |
407 | ||
408 | /* We emulate setlocale with our own function. LC_foo is not valid for the | |
409 | * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array | |
410 | * lookup to convert to. At compile time we have defined LC_foo_INDEX as the | |
411 | * proper offset into the array 'category_masks[]'. At runtime, we have to | |
412 | * search through the array (as the actual numbers may not be small contiguous | |
413 | * positive integers which would lend themselves to array lookup). */ | |
414 | # define do_setlocale_c(cat, locale) \ | |
415 | emulate_setlocale(cat, locale, cat ## _INDEX, TRUE) | |
416 | # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE) | |
417 | ||
418 | /* A third array, parallel to the ones above to map from category to its | |
419 | * equivalent mask */ | |
420 | const int category_masks[] = { | |
421 | # ifdef USE_LOCALE_NUMERIC | |
422 | LC_NUMERIC_MASK, | |
423 | # endif | |
424 | # ifdef USE_LOCALE_CTYPE | |
425 | LC_CTYPE_MASK, | |
426 | # endif | |
427 | # ifdef USE_LOCALE_COLLATE | |
428 | LC_COLLATE_MASK, | |
429 | # endif | |
430 | # ifdef USE_LOCALE_TIME | |
431 | LC_TIME_MASK, | |
432 | # endif | |
433 | # ifdef USE_LOCALE_MESSAGES | |
434 | LC_MESSAGES_MASK, | |
435 | # endif | |
436 | # ifdef USE_LOCALE_MONETARY | |
437 | LC_MONETARY_MASK, | |
438 | # endif | |
439 | # ifdef USE_LOCALE_ADDRESS | |
440 | LC_ADDRESS_MASK, | |
441 | # endif | |
442 | # ifdef USE_LOCALE_IDENTIFICATION | |
443 | LC_IDENTIFICATION_MASK, | |
444 | # endif | |
445 | # ifdef USE_LOCALE_MEASUREMENT | |
446 | LC_MEASUREMENT_MASK, | |
447 | # endif | |
448 | # ifdef USE_LOCALE_PAPER | |
449 | LC_PAPER_MASK, | |
450 | # endif | |
451 | # ifdef USE_LOCALE_TELEPHONE | |
452 | LC_TELEPHONE_MASK, | |
453 | # endif | |
454 | /* LC_ALL can't be turned off by a Configure | |
455 | * option, and in Posix 2008, should always be | |
456 | * here, so compile it in unconditionally. | |
457 | * This could catch some glitches at compile | |
458 | * time */ | |
459 | LC_ALL_MASK | |
460 | }; | |
461 | ||
462 | STATIC const char * | |
463 | S_emulate_setlocale(const int category, | |
464 | const char * locale, | |
465 | unsigned int index, | |
466 | const bool is_index_valid | |
467 | ) | |
468 | { | |
469 | /* This function effectively performs a setlocale() on just the current | |
470 | * thread; thus it is thread-safe. It does this by using the POSIX 2008 | |
471 | * locale functions to emulate the behavior of setlocale(). Similar to | |
472 | * regular setlocale(), the return from this function points to memory that | |
473 | * can be overwritten by other system calls, so needs to be copied | |
474 | * immediately if you need to retain it. The difference here is that | |
475 | * system calls besides another setlocale() can overwrite it. | |
476 | * | |
477 | * By doing this, most locale-sensitive functions become thread-safe. The | |
478 | * exceptions are mostly those that return a pointer to static memory. | |
479 | * | |
480 | * This function takes the same parameters, 'category' and 'locale', that | |
481 | * the regular setlocale() function does, but it also takes two additional | |
482 | * ones. This is because the 2008 functions don't use a category; instead | |
483 | * they use a corresponding mask. Because this function operates in both | |
484 | * worlds, it may need one or the other or both. This function can | |
485 | * calculate the mask from the input category, but to avoid this | |
486 | * calculation, if the caller knows at compile time what the mask is, it | |
487 | * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask | |
488 | * parameter is ignored. | |
489 | * | |
490 | * POSIX 2008, for some sick reason, chose not to provide a method to find | |
491 | * the category name of a locale. Some vendors have created a | |
492 | * querylocale() function to do just that. This function is a lot simpler | |
493 | * to implement on systems that have this. Otherwise, we have to keep | |
494 | * track of what the locale has been set to, so that we can return its | |
495 | * name to emulate setlocale(). It's also possible for C code in some | |
496 | * library to change the locale without us knowing it, though as of | |
497 | * September 2017, there are no occurrences in CPAN of uselocale(). Some | |
498 | * libraries do use setlocale(), but that changes the global locale, and | |
499 | * threads using per-thread locales will just ignore those changes. | |
500 | * Another problem is that without querylocale(), we have to guess at what | |
501 | * was meant by setting a locale of "". We handle this by not actually | |
502 | * ever setting to "" (unless querylocale exists), but to emulate what we | |
503 | * think should happen for "". | |
504 | */ | |
505 | ||
506 | int mask; | |
507 | locale_t old_obj; | |
508 | locale_t new_obj; | |
509 | dTHX; | |
510 | ||
511 | # ifdef DEBUGGING | |
512 | ||
513 | if (DEBUG_Lv_TEST || debug_initialization) { | |
514 | 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); | |
515 | } | |
516 | ||
517 | # endif | |
518 | ||
519 | /* If the input mask might be incorrect, calculate the correct one */ | |
520 | if (! is_index_valid) { | |
521 | unsigned int i; | |
522 | ||
523 | # ifdef DEBUGGING | |
524 | ||
525 | if (DEBUG_Lv_TEST || debug_initialization) { | |
526 | PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category)); | |
527 | } | |
528 | ||
529 | # endif | |
530 | ||
531 | for (i = 0; i <= LC_ALL_INDEX; i++) { | |
532 | if (category == categories[i]) { | |
533 | index = i; | |
534 | goto found_index; | |
535 | } | |
536 | } | |
537 | ||
538 | /* Here, we don't know about this category, so can't handle it. | |
539 | * Fallback to the early POSIX usages */ | |
540 | Perl_warner(aTHX_ packWARN(WARN_LOCALE), | |
541 | "Unknown locale category %d; can't set it to %s\n", | |
542 | category, locale); | |
543 | return NULL; | |
544 | ||
545 | found_index: ; | |
546 | ||
547 | # ifdef DEBUGGING | |
548 | ||
549 | if (DEBUG_Lv_TEST || debug_initialization) { | |
550 | PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category)); | |
551 | } | |
552 | ||
553 | # endif | |
554 | ||
555 | } | |
556 | ||
557 | mask = category_masks[index]; | |
558 | ||
559 | # ifdef DEBUGGING | |
560 | ||
561 | if (DEBUG_Lv_TEST || debug_initialization) { | |
562 | PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask); | |
563 | } | |
564 | ||
565 | # endif | |
566 | ||
567 | /* If just querying what the existing locale is ... */ | |
568 | if (locale == NULL) { | |
569 | locale_t cur_obj = uselocale((locale_t) 0); | |
570 | ||
571 | # ifdef DEBUGGING | |
572 | ||
573 | if (DEBUG_Lv_TEST || debug_initialization) { | |
574 | PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj); | |
575 | } | |
576 | ||
577 | # endif | |
578 | ||
579 | if (cur_obj == LC_GLOBAL_LOCALE) { | |
580 | return my_setlocale(category, NULL); | |
581 | } | |
582 | ||
583 | # ifdef HAS_QUERYLOCALE | |
584 | ||
585 | return (char *) querylocale(mask, cur_obj); | |
586 | ||
d2b24094 | 587 | # else |
ecdda939 KW |
588 | |
589 | /* If this assert fails, adjust the size of curlocales in intrpvar.h */ | |
590 | STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX); | |
591 | ||
8e13243a KW |
592 | # if defined(_NL_LOCALE_NAME) \ |
593 | && defined(DEBUGGING) \ | |
594 | && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME) | |
595 | /* On systems that accept any locale name, the real underlying locale | |
596 | * is often returned by this internal function, so we can't use it */ | |
e9bc6d6b KW |
597 | { |
598 | /* Internal glibc for querylocale(), but doesn't handle | |
599 | * empty-string ("") locale properly; who knows what other | |
3a732259 | 600 | * glitches. Check for it now, under debug. */ |
e9bc6d6b KW |
601 | |
602 | char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category), | |
603 | uselocale((locale_t) 0)); | |
604 | /* | |
605 | PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL"); | |
606 | PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index); | |
607 | PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]); | |
608 | */ | |
609 | if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) { | |
610 | if ( strNE(PL_curlocales[index], temp_name) | |
611 | && ! ( isNAME_C_OR_POSIX(temp_name) | |
612 | && isNAME_C_OR_POSIX(PL_curlocales[index]))) { | |
613 | ||
614 | # ifdef USE_C_BACKTRACE | |
615 | ||
616 | dump_c_backtrace(Perl_debug_log, 20, 1); | |
617 | ||
618 | # endif | |
619 | ||
620 | Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is" | |
621 | " (%s) and what internal glibc thinks" | |
622 | " (%s)\n", category_names[index], | |
623 | PL_curlocales[index], temp_name); | |
624 | } | |
625 | ||
626 | return temp_name; | |
627 | } | |
628 | } | |
629 | ||
630 | # endif | |
631 | ||
632 | /* Without querylocale(), we have to use our record-keeping we've | |
633 | * done. */ | |
634 | ||
635 | if (category != LC_ALL) { | |
636 | ||
637 | # ifdef DEBUGGING | |
638 | ||
639 | if (DEBUG_Lv_TEST || debug_initialization) { | |
640 | PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]); | |
641 | } | |
642 | ||
643 | # endif | |
644 | ||
645 | return PL_curlocales[index]; | |
646 | } | |
647 | else { /* For LC_ALL */ | |
648 | unsigned int i; | |
649 | Size_t names_len = 0; | |
650 | char * all_string; | |
7c93a471 | 651 | bool are_all_categories_the_same_locale = TRUE; |
e9bc6d6b KW |
652 | |
653 | /* If we have a valid LC_ALL value, just return it */ | |
654 | if (PL_curlocales[LC_ALL_INDEX]) { | |
655 | ||
656 | # ifdef DEBUGGING | |
657 | ||
658 | if (DEBUG_Lv_TEST || debug_initialization) { | |
659 | PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]); | |
660 | } | |
661 | ||
662 | # endif | |
663 | ||
664 | return PL_curlocales[LC_ALL_INDEX]; | |
665 | } | |
666 | ||
667 | /* Otherwise, we need to construct a string of name=value pairs. | |
668 | * We use the glibc syntax, like | |
669 | * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;... | |
7c93a471 KW |
670 | * First calculate the needed size. Along the way, check if all |
671 | * the locale names are the same */ | |
e9bc6d6b KW |
672 | for (i = 0; i < LC_ALL_INDEX; i++) { |
673 | ||
674 | # ifdef DEBUGGING | |
675 | ||
676 | if (DEBUG_Lv_TEST || debug_initialization) { | |
677 | 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]); | |
678 | } | |
679 | ||
680 | # endif | |
681 | ||
682 | names_len += strlen(category_names[i]) | |
683 | + 1 /* '=' */ | |
684 | + strlen(PL_curlocales[i]) | |
685 | + 1; /* ';' */ | |
7c93a471 KW |
686 | |
687 | if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) { | |
688 | are_all_categories_the_same_locale = FALSE; | |
689 | } | |
690 | } | |
691 | ||
692 | /* If they are the same, we don't actually have to construct the | |
693 | * string; we just make the entry in LC_ALL_INDEX valid, and be | |
694 | * that single name */ | |
695 | if (are_all_categories_the_same_locale) { | |
696 | PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]); | |
697 | return PL_curlocales[LC_ALL_INDEX]; | |
e9bc6d6b | 698 | } |
7c93a471 | 699 | |
e9bc6d6b KW |
700 | names_len++; /* Trailing '\0' */ |
701 | SAVEFREEPV(Newx(all_string, names_len, char)); | |
702 | *all_string = '\0'; | |
703 | ||
704 | /* Then fill in the string */ | |
705 | for (i = 0; i < LC_ALL_INDEX; i++) { | |
706 | ||
707 | # ifdef DEBUGGING | |
708 | ||
709 | if (DEBUG_Lv_TEST || debug_initialization) { | |
710 | 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]); | |
711 | } | |
712 | ||
713 | # endif | |
714 | ||
715 | my_strlcat(all_string, category_names[i], names_len); | |
716 | my_strlcat(all_string, "=", names_len); | |
717 | my_strlcat(all_string, PL_curlocales[i], names_len); | |
718 | my_strlcat(all_string, ";", names_len); | |
719 | } | |
720 | ||
721 | # ifdef DEBUGGING | |
722 | ||
723 | if (DEBUG_L_TEST || debug_initialization) { | |
724 | PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string); | |
725 | } | |
726 | ||
727 | #endif | |
728 | ||
729 | return all_string; | |
730 | } | |
731 | ||
732 | # ifdef EINVAL | |
733 | ||
734 | SETERRNO(EINVAL, LIB_INVARG); | |
735 | ||
736 | # endif | |
737 | ||
738 | return NULL; | |
739 | ||
d2b24094 KW |
740 | # endif |
741 | ||
6aba5c5e | 742 | } /* End of this being setlocale(LC_foo, NULL) */ |
e9bc6d6b | 743 | |
f8115d60 | 744 | /* Here, we are switching locales. */ |
e9bc6d6b KW |
745 | |
746 | # ifndef HAS_QUERYLOCALE | |
747 | ||
748 | if (strEQ(locale, "")) { | |
749 | ||
750 | /* For non-querylocale() systems, we do the setting of "" ourselves to | |
751 | * be sure that we really know what's going on. We follow the Linux | |
752 | * documented behavior (but if that differs from the actual behavior, | |
753 | * this won't work exactly as the OS implements). We go out and | |
754 | * examine the environment based on our understanding of how the system | |
755 | * works, and use that to figure things out */ | |
756 | ||
757 | const char * const lc_all = PerlEnv_getenv("LC_ALL"); | |
758 | ||
759 | /* Use any "LC_ALL" environment variable, as it overrides everything | |
760 | * else. */ | |
761 | if (lc_all && strNE(lc_all, "")) { | |
762 | locale = lc_all; | |
763 | } | |
764 | else { | |
765 | ||
766 | /* Otherwise, we need to dig deeper. Unless overridden, the | |
767 | * default is the LANG environment variable; if it doesn't exist, | |
768 | * then "C" */ | |
769 | ||
770 | const char * default_name; | |
771 | ||
207cc8df | 772 | default_name = PerlEnv_getenv("LANG"); |
e9bc6d6b KW |
773 | |
774 | if (! default_name || strEQ(default_name, "")) { | |
929a7f8c | 775 | default_name = "C"; |
e9bc6d6b KW |
776 | } |
777 | else if (PL_scopestack_ix != 0) { | |
207cc8df DM |
778 | /* To minimize other threads messing with the environment, |
779 | * we copy the variable, making it a temporary. But this | |
780 | * doesn't work upon program initialization before any | |
781 | * scopes are created, and at this time, there's nothing | |
782 | * else going on that would interfere. So skip the copy | |
783 | * in that case */ | |
784 | default_name = savepv(default_name); | |
e9bc6d6b KW |
785 | SAVEFREEPV(default_name); |
786 | } | |
787 | ||
788 | if (category != LC_ALL) { | |
789 | const char * const name = PerlEnv_getenv(category_names[index]); | |
790 | ||
791 | /* Here we are setting a single category. Assume will have the | |
792 | * default name */ | |
793 | locale = default_name; | |
794 | ||
795 | /* But then look for an overriding environment variable */ | |
796 | if (name && strNE(name, "")) { | |
797 | locale = name; | |
798 | } | |
799 | } | |
800 | else { | |
801 | bool did_override = FALSE; | |
802 | unsigned int i; | |
803 | ||
804 | /* Here, we are getting LC_ALL. Any categories that don't have | |
805 | * a corresponding environment variable set should be set to | |
806 | * LANG, or to "C" if there is no LANG. If no individual | |
807 | * categories differ from this, we can just set LC_ALL. This | |
808 | * is buggy on systems that have extra categories that we don't | |
809 | * know about. If there is an environment variable that sets | |
810 | * that category, we won't know to look for it, and so our use | |
811 | * of LANG or "C" improperly overrides it. On the other hand, | |
812 | * if we don't do what is done here, and there is no | |
813 | * environment variable, the category's locale should be set to | |
814 | * LANG or "C". So there is no good solution. khw thinks the | |
815 | * best is to look at systems to see what categories they have, | |
816 | * and include them, and then to assume that we know the | |
817 | * complete set */ | |
818 | ||
819 | for (i = 0; i < LC_ALL_INDEX; i++) { | |
820 | const char * const env_override | |
821 | = savepv(PerlEnv_getenv(category_names[i])); | |
822 | const char * this_locale = ( env_override | |
823 | && strNE(env_override, "")) | |
824 | ? env_override | |
825 | : default_name; | |
6435f98d KW |
826 | if (! emulate_setlocale(categories[i], this_locale, i, TRUE)) |
827 | { | |
828 | Safefree(env_override); | |
829 | return NULL; | |
830 | } | |
e9bc6d6b KW |
831 | |
832 | if (strNE(this_locale, default_name)) { | |
833 | did_override = TRUE; | |
834 | } | |
835 | ||
836 | Safefree(env_override); | |
837 | } | |
838 | ||
839 | /* If all the categories are the same, we can set LC_ALL to | |
840 | * that */ | |
841 | if (! did_override) { | |
842 | locale = default_name; | |
843 | } | |
844 | else { | |
845 | ||
846 | /* Here, LC_ALL is no longer valid, as some individual | |
847 | * categories don't match it. We call ourselves | |
848 | * recursively, as that will execute the code that | |
849 | * generates the proper locale string for this situation. | |
850 | * We don't do the remainder of this function, as that is | |
851 | * to update our records, and we've just done that for the | |
852 | * individual categories in the loop above, and doing so | |
853 | * would cause LC_ALL to be done as well */ | |
854 | return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE); | |
855 | } | |
856 | } | |
857 | } | |
6aba5c5e | 858 | } /* End of this being setlocale(LC_foo, "") */ |
e9bc6d6b KW |
859 | else if (strchr(locale, ';')) { |
860 | ||
861 | /* LC_ALL may actually incude a conglomeration of various categories. | |
862 | * Without querylocale, this code uses the glibc (as of this writing) | |
863 | * syntax for representing that, but that is not a stable API, and | |
864 | * other platforms do it differently, so we have to handle all cases | |
865 | * ourselves */ | |
866 | ||
f0023d47 | 867 | unsigned int i; |
e9bc6d6b KW |
868 | const char * s = locale; |
869 | const char * e = locale + strlen(locale); | |
870 | const char * p = s; | |
871 | const char * category_end; | |
872 | const char * name_start; | |
873 | const char * name_end; | |
874 | ||
f0023d47 KW |
875 | /* If the string that gives what to set doesn't include all categories, |
876 | * the omitted ones get set to "C". To get this behavior, first set | |
877 | * all the individual categories to "C", and override the furnished | |
878 | * ones below */ | |
879 | for (i = 0; i < LC_ALL_INDEX; i++) { | |
880 | if (! emulate_setlocale(categories[i], "C", i, TRUE)) { | |
881 | return NULL; | |
882 | } | |
883 | } | |
884 | ||
e9bc6d6b | 885 | while (s < e) { |
e9bc6d6b KW |
886 | |
887 | /* Parse through the category */ | |
888 | while (isWORDCHAR(*p)) { | |
889 | p++; | |
890 | } | |
891 | category_end = p; | |
892 | ||
893 | if (*p++ != '=') { | |
894 | Perl_croak(aTHX_ | |
895 | "panic: %s: %d: Unexpected character in locale name '%02X", | |
896 | __FILE__, __LINE__, *(p-1)); | |
897 | } | |
898 | ||
899 | /* Parse through the locale name */ | |
900 | name_start = p; | |
bee74f4b KW |
901 | while (p < e && *p != ';') { |
902 | if (! isGRAPH(*p)) { | |
903 | Perl_croak(aTHX_ | |
904 | "panic: %s: %d: Unexpected character in locale name '%02X", | |
905 | __FILE__, __LINE__, *(p-1)); | |
906 | } | |
e9bc6d6b KW |
907 | p++; |
908 | } | |
909 | name_end = p; | |
910 | ||
bee74f4b KW |
911 | /* Space past the semi-colon */ |
912 | if (p < e) { | |
913 | p++; | |
e9bc6d6b KW |
914 | } |
915 | ||
916 | /* Find the index of the category name in our lists */ | |
917 | for (i = 0; i < LC_ALL_INDEX; i++) { | |
3c76a412 | 918 | char * individ_locale; |
e9bc6d6b KW |
919 | |
920 | /* Keep going if this isn't the index. The strnNE() avoids a | |
921 | * Perl_form(), but would fail if ever a category name could be | |
922 | * a substring of another one, like if there were a | |
923 | * "LC_TIME_DATE" */ | |
924 | if strnNE(s, category_names[i], category_end - s) { | |
925 | continue; | |
926 | } | |
927 | ||
928 | /* If this index is for the single category we're changing, we | |
929 | * have found the locale to set it to. */ | |
930 | if (category == categories[i]) { | |
931 | locale = Perl_form(aTHX_ "%.*s", | |
932 | (int) (name_end - name_start), | |
933 | name_start); | |
934 | goto ready_to_set; | |
935 | } | |
936 | ||
3c76a412 | 937 | assert(category == LC_ALL); |
bee74f4b KW |
938 | individ_locale = Perl_form(aTHX_ "%.*s", |
939 | (int) (name_end - name_start), name_start); | |
6435f98d KW |
940 | if (! emulate_setlocale(categories[i], individ_locale, i, TRUE)) |
941 | { | |
942 | return NULL; | |
943 | } | |
e9bc6d6b KW |
944 | } |
945 | ||
946 | s = p; | |
947 | } | |
948 | ||
949 | /* Here we have set all the individual categories by recursive calls. | |
950 | * These collectively should have fixed up LC_ALL, so can just query | |
951 | * what that now is */ | |
952 | assert(category == LC_ALL); | |
953 | ||
954 | return do_setlocale_c(LC_ALL, NULL); | |
6aba5c5e KW |
955 | } /* End of this being setlocale(LC_ALL, |
956 | "LC_CTYPE=foo;LC_NUMERIC=bar;...") */ | |
e9bc6d6b KW |
957 | |
958 | ready_to_set: ; | |
959 | ||
f8115d60 KW |
960 | /* Here at the end of having to deal with the absence of querylocale(). |
961 | * Some cases have already been fully handled by recursive calls to this | |
962 | * function. But at this point, we haven't dealt with those, but are now | |
963 | * prepared to, knowing what the locale name to set this category to is. | |
964 | * This would have come for free if this system had had querylocale() */ | |
965 | ||
e9bc6d6b KW |
966 | # endif /* end of ! querylocale */ |
967 | ||
f8115d60 KW |
968 | assert(PL_C_locale_obj); |
969 | ||
970 | /* Switching locales generally entails freeing the current one's space (at | |
971 | * the C library's discretion). We need to stop using that locale before | |
972 | * the switch. So switch to a known locale object that we don't otherwise | |
973 | * mess with. This returns the locale object in effect at the time of the | |
974 | * switch. */ | |
975 | old_obj = uselocale(PL_C_locale_obj); | |
976 | ||
977 | # ifdef DEBUGGING | |
978 | ||
979 | if (DEBUG_Lv_TEST || debug_initialization) { | |
980 | PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj); | |
981 | } | |
982 | ||
983 | # endif | |
984 | ||
985 | if (! old_obj) { | |
986 | ||
987 | # ifdef DEBUGGING | |
988 | ||
989 | if (DEBUG_L_TEST || debug_initialization) { | |
990 | dSAVE_ERRNO; | |
991 | PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO); | |
992 | RESTORE_ERRNO; | |
993 | } | |
994 | ||
995 | # endif | |
996 | ||
997 | return NULL; | |
998 | } | |
999 | ||
1000 | # ifdef DEBUGGING | |
1001 | ||
1002 | if (DEBUG_Lv_TEST || debug_initialization) { | |
b22e9937 KW |
1003 | PerlIO_printf(Perl_debug_log, |
1004 | "%s:%d: emulate_setlocale now using %p\n", | |
1005 | __FILE__, __LINE__, PL_C_locale_obj); | |
f8115d60 KW |
1006 | } |
1007 | ||
1008 | # endif | |
1009 | ||
6aba5c5e KW |
1010 | /* If this call is to switch to the LC_ALL C locale, it already exists, and |
1011 | * in fact, we already have switched to it (in preparation for what | |
1012 | * normally is to come). But since we're already there, continue to use | |
70bd6bc8 KW |
1013 | * it instead of trying to create a new locale */ |
1014 | if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) { | |
1015 | ||
1016 | # ifdef DEBUGGING | |
1017 | ||
1018 | if (DEBUG_Lv_TEST || debug_initialization) { | |
1019 | PerlIO_printf(Perl_debug_log, | |
1020 | "%s:%d: will stay in C object\n", __FILE__, __LINE__); | |
1021 | } | |
1022 | ||
1023 | # endif | |
1024 | ||
1025 | new_obj = PL_C_locale_obj; | |
1026 | ||
1027 | /* We already had switched to the C locale in preparation for freeing | |
b22e9937 | 1028 | * 'old_obj' */ |
70bd6bc8 KW |
1029 | if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) { |
1030 | freelocale(old_obj); | |
1031 | } | |
1032 | } | |
1033 | else { | |
b22e9937 KW |
1034 | /* If we weren't in a thread safe locale, set so that newlocale() below |
1035 | * which uses 'old_obj', uses an empty one. Same for our reserved C | |
1036 | * object. The latter is defensive coding, so that, even if there is | |
1037 | * some bug, we will never end up trying to modify either of these, as | |
1038 | * if passed to newlocale(), they can be. */ | |
1039 | if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) { | |
1040 | old_obj = (locale_t) 0; | |
1041 | } | |
f8115d60 | 1042 | |
b22e9937 KW |
1043 | /* Ready to create a new locale by modification of the exising one */ |
1044 | new_obj = newlocale(mask, locale, old_obj); | |
e9bc6d6b | 1045 | |
b22e9937 KW |
1046 | if (! new_obj) { |
1047 | dSAVE_ERRNO; | |
e9bc6d6b KW |
1048 | |
1049 | # ifdef DEBUGGING | |
1050 | ||
b22e9937 KW |
1051 | if (DEBUG_L_TEST || debug_initialization) { |
1052 | PerlIO_printf(Perl_debug_log, | |
1053 | "%s:%d: emulate_setlocale creating new object" | |
1054 | " failed: %d\n", __FILE__, __LINE__, GET_ERRNO); | |
1055 | } | |
e9bc6d6b KW |
1056 | |
1057 | # endif | |
1058 | ||
b22e9937 | 1059 | if (! uselocale(old_obj)) { |
e9bc6d6b KW |
1060 | |
1061 | # ifdef DEBUGGING | |
1062 | ||
b22e9937 KW |
1063 | if (DEBUG_L_TEST || debug_initialization) { |
1064 | PerlIO_printf(Perl_debug_log, | |
1065 | "%s:%d: switching back failed: %d\n", | |
1066 | __FILE__, __LINE__, GET_ERRNO); | |
1067 | } | |
e9bc6d6b KW |
1068 | |
1069 | # endif | |
1070 | ||
b22e9937 KW |
1071 | } |
1072 | RESTORE_ERRNO; | |
1073 | return NULL; | |
e9bc6d6b | 1074 | } |
e9bc6d6b KW |
1075 | |
1076 | # ifdef DEBUGGING | |
1077 | ||
b22e9937 KW |
1078 | if (DEBUG_Lv_TEST || debug_initialization) { |
1079 | PerlIO_printf(Perl_debug_log, | |
1080 | "%s:%d: emulate_setlocale created %p", | |
1081 | __FILE__, __LINE__, new_obj); | |
1082 | if (old_obj) { | |
1083 | PerlIO_printf(Perl_debug_log, | |
1084 | "; should have freed %p", old_obj); | |
1085 | } | |
1086 | PerlIO_printf(Perl_debug_log, "\n"); | |
19ee3daf | 1087 | } |
e9bc6d6b KW |
1088 | |
1089 | # endif | |
1090 | ||
b22e9937 KW |
1091 | /* And switch into it */ |
1092 | if (! uselocale(new_obj)) { | |
1093 | dSAVE_ERRNO; | |
e9bc6d6b KW |
1094 | |
1095 | # ifdef DEBUGGING | |
1096 | ||
b22e9937 KW |
1097 | if (DEBUG_L_TEST || debug_initialization) { |
1098 | PerlIO_printf(Perl_debug_log, | |
1099 | "%s:%d: emulate_setlocale switching to new object" | |
1100 | " failed\n", __FILE__, __LINE__); | |
1101 | } | |
e9bc6d6b KW |
1102 | |
1103 | # endif | |
1104 | ||
b22e9937 | 1105 | if (! uselocale(old_obj)) { |
e9bc6d6b KW |
1106 | |
1107 | # ifdef DEBUGGING | |
1108 | ||
b22e9937 KW |
1109 | if (DEBUG_L_TEST || debug_initialization) { |
1110 | PerlIO_printf(Perl_debug_log, | |
1111 | "%s:%d: switching back failed: %d\n", | |
1112 | __FILE__, __LINE__, GET_ERRNO); | |
1113 | } | |
e9bc6d6b KW |
1114 | |
1115 | # endif | |
1116 | ||
b22e9937 KW |
1117 | } |
1118 | freelocale(new_obj); | |
1119 | RESTORE_ERRNO; | |
1120 | return NULL; | |
e9bc6d6b | 1121 | } |
70bd6bc8 | 1122 | } |
e9bc6d6b KW |
1123 | |
1124 | # ifdef DEBUGGING | |
1125 | ||
1126 | if (DEBUG_Lv_TEST || debug_initialization) { | |
b22e9937 KW |
1127 | PerlIO_printf(Perl_debug_log, |
1128 | "%s:%d: emulate_setlocale now using %p\n", | |
1129 | __FILE__, __LINE__, new_obj); | |
e9bc6d6b KW |
1130 | } |
1131 | ||
1132 | # endif | |
1133 | ||
1134 | /* We are done, except for updating our records (if the system doesn't keep | |
1135 | * them) and in the case of locale "", we don't actually know what the | |
1136 | * locale that got switched to is, as it came from the environment. So | |
1137 | * have to find it */ | |
1138 | ||
1139 | # ifdef HAS_QUERYLOCALE | |
1140 | ||
1141 | if (strEQ(locale, "")) { | |
1142 | locale = querylocale(mask, new_obj); | |
1143 | } | |
1144 | ||
1145 | # else | |
1146 | ||
1147 | /* Here, 'locale' is the return value */ | |
1148 | ||
1149 | /* Without querylocale(), we have to update our records */ | |
1150 | ||
1151 | if (category == LC_ALL) { | |
1152 | unsigned int i; | |
1153 | ||
1154 | /* For LC_ALL, we change all individual categories to correspond */ | |
1155 | /* PL_curlocales is a parallel array, so has same | |
1156 | * length as 'categories' */ | |
1157 | for (i = 0; i <= LC_ALL_INDEX; i++) { | |
1158 | Safefree(PL_curlocales[i]); | |
1159 | PL_curlocales[i] = savepv(locale); | |
1160 | } | |
1161 | } | |
1162 | else { | |
1163 | ||
1164 | /* For a single category, if it's not the same as the one in LC_ALL, we | |
1165 | * nullify LC_ALL */ | |
1166 | ||
1167 | if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) { | |
1168 | Safefree(PL_curlocales[LC_ALL_INDEX]); | |
1169 | PL_curlocales[LC_ALL_INDEX] = NULL; | |
1170 | } | |
1171 | ||
1172 | /* Then update the category's record */ | |
1173 | Safefree(PL_curlocales[index]); | |
1174 | PL_curlocales[index] = savepv(locale); | |
1175 | } | |
1176 | ||
1177 | # endif | |
1178 | ||
1179 | return locale; | |
1180 | } | |
1181 | ||
1182 | #endif /* USE_POSIX_2008_LOCALE */ | |
1183 | ||
1184 | #if 0 /* Code that was to emulate thread-safe locales on platforms that | |
1185 | didn't natively support them */ | |
1186 | ||
1187 | /* The way this would work is that we would keep a per-thread list of the | |
1188 | * correct locale for that thread. Any operation that was locale-sensitive | |
1189 | * would have to be changed so that it would look like this: | |
1190 | * | |
1191 | * LOCALE_LOCK; | |
1192 | * setlocale to the correct locale for this operation | |
1193 | * do operation | |
1194 | * LOCALE_UNLOCK | |
1195 | * | |
1196 | * This leaves the global locale in the most recently used operation's, but it | |
1197 | * was locked long enough to get the result. If that result is static, it | |
1198 | * needs to be copied before the unlock. | |
1199 | * | |
1200 | * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did | |
1201 | * the setup, but are no-ops when not needed, and similarly, | |
1202 | * END_LOCALE_DEPENDENT_OP for the tear-down | |
1203 | * | |
1204 | * But every call to a locale-sensitive function would have to be changed, and | |
1205 | * if a module didn't cooperate by using the mutex, things would break. | |
1206 | * | |
1207 | * This code was abandoned before being completed or tested, and is left as-is | |
1208 | */ | |
1209 | ||
1210 | # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE) | |
1211 | # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE) | |
1212 | ||
1213 | STATIC char * | |
1214 | S_locking_setlocale(pTHX_ | |
1215 | const int category, | |
1216 | const char * locale, | |
1217 | int index, | |
1218 | const bool is_index_valid | |
1219 | ) | |
1220 | { | |
1221 | /* This function kind of performs a setlocale() on just the current thread; | |
1222 | * thus it is kind of thread-safe. It does this by keeping a thread-level | |
1223 | * array of the current locales for each category. Every time a locale is | |
1224 | * switched to, it does the switch globally, but updates the thread's | |
1225 | * array. A query as to what the current locale is just returns the | |
1226 | * appropriate element from the array, and doesn't actually call the system | |
1227 | * setlocale(). The saving into the array is done in an uninterruptible | |
1228 | * section of code, so is unaffected by whatever any other threads might be | |
1229 | * doing. | |
1230 | * | |
1231 | * All locale-sensitive operations must work by first starting a critical | |
1232 | * section, then switching to the thread's locale as kept by this function, | |
1233 | * and then doing the operation, then ending the critical section. Thus, | |
1234 | * each gets done in the appropriate locale. simulating thread-safety. | |
1235 | * | |
1236 | * This function takes the same parameters, 'category' and 'locale', that | |
1237 | * the regular setlocale() function does, but it also takes two additional | |
1238 | * ones. This is because as described earlier. If we know on input the | |
1239 | * index corresponding to the category into the array where we store the | |
1240 | * current locales, we don't have to calculate it. If the caller knows at | |
06cc5386 | 1241 | * compile time what the index is, it can pass it, setting |
e9bc6d6b KW |
1242 | * 'is_index_valid' to TRUE; otherwise the index parameter is ignored. |
1243 | * | |
1244 | */ | |
1245 | ||
1246 | /* If the input index might be incorrect, calculate the correct one */ | |
1247 | if (! is_index_valid) { | |
1248 | unsigned int i; | |
1249 | ||
1250 | if (DEBUG_Lv_TEST || debug_initialization) { | |
1251 | PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category); | |
1252 | } | |
1253 | ||
1254 | for (i = 0; i <= LC_ALL_INDEX; i++) { | |
1255 | if (category == categories[i]) { | |
1256 | index = i; | |
1257 | goto found_index; | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | /* Here, we don't know about this category, so can't handle it. | |
1262 | * XXX best we can do is to unsafely set this | |
1263 | * XXX warning */ | |
1264 | ||
1265 | return my_setlocale(category, locale); | |
1266 | ||
1267 | found_index: ; | |
1268 | ||
1269 | if (DEBUG_Lv_TEST || debug_initialization) { | |
1270 | PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index); | |
1271 | } | |
1272 | } | |
1273 | ||
1274 | /* For a query, just return what's in our records */ | |
1275 | if (new_locale == NULL) { | |
1276 | return curlocales[index]; | |
1277 | } | |
1278 | ||
1279 | ||
1280 | /* Otherwise, we need to do the switch, and save the result, all in a | |
1281 | * critical section */ | |
1282 | ||
1283 | Safefree(curlocales[[index]]); | |
1284 | ||
1285 | /* It might be that this is called from an already-locked section of code. | |
1286 | * We would have to detect and skip the LOCK/UNLOCK if so */ | |
1287 | LOCALE_LOCK; | |
1288 | ||
1289 | curlocales[index] = savepv(my_setlocale(category, new_locale)); | |
1290 | ||
1291 | if (strEQ(new_locale, "")) { | |
1292 | ||
1293 | #ifdef LC_ALL | |
1294 | ||
1295 | /* The locale values come from the environment, and may not all be the | |
1296 | * same, so for LC_ALL, we have to update all the others, while the | |
1297 | * mutex is still locked */ | |
1298 | ||
1299 | if (category == LC_ALL) { | |
1300 | unsigned int i; | |
1301 | for (i = 0; i < LC_ALL_INDEX) { | |
1302 | curlocales[i] = my_setlocale(categories[i], NULL); | |
1303 | } | |
1304 | } | |
1305 | } | |
1306 | ||
1307 | #endif | |
1308 | ||
1309 | LOCALE_UNLOCK; | |
1310 | ||
1311 | return curlocales[index]; | |
1312 | } | |
1313 | ||
1314 | #endif | |
d36adde0 | 1315 | #ifdef USE_LOCALE |
837ce802 | 1316 | |
a4f00dcc | 1317 | STATIC void |
86799d2d | 1318 | S_set_numeric_radix(pTHX_ const bool use_locale) |
98994639 | 1319 | { |
86799d2d KW |
1320 | /* If 'use_locale' is FALSE, set to use a dot for the radix character. If |
1321 | * TRUE, use the radix character derived from the current locale */ | |
7d4bcc4a | 1322 | |
86799d2d KW |
1323 | #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \ |
1324 | || defined(HAS_NL_LANGINFO)) | |
98994639 | 1325 | |
87f8e8e7 | 1326 | const char * radix = (use_locale) |
4e6826bf | 1327 | ? my_nl_langinfo(RADIXCHAR, FALSE) |
87f8e8e7 KW |
1328 | /* FALSE => already in dest locale */ |
1329 | : "."; | |
2213a3be | 1330 | |
87f8e8e7 | 1331 | sv_setpv(PL_numeric_radix_sv, radix); |
86799d2d | 1332 | |
87f8e8e7 KW |
1333 | /* If this is valid UTF-8 that isn't totally ASCII, and we are in |
1334 | * a UTF-8 locale, then mark the radix as being in UTF-8 */ | |
1335 | if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv), | |
7a393424 | 1336 | SvCUR(PL_numeric_radix_sv)) |
87f8e8e7 KW |
1337 | && _is_cur_LC_category_utf8(LC_NUMERIC)) |
1338 | { | |
1339 | SvUTF8_on(PL_numeric_radix_sv); | |
1340 | } | |
86799d2d KW |
1341 | |
1342 | # ifdef DEBUGGING | |
7d4bcc4a | 1343 | |
2fcc0ca9 KW |
1344 | if (DEBUG_L_TEST || debug_initialization) { |
1345 | PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n", | |
3ca88433 KW |
1346 | SvPVX(PL_numeric_radix_sv), |
1347 | cBOOL(SvUTF8(PL_numeric_radix_sv))); | |
2fcc0ca9 | 1348 | } |
69014004 | 1349 | |
86799d2d | 1350 | # endif |
d36adde0 KW |
1351 | #else |
1352 | ||
1353 | PERL_UNUSED_ARG(use_locale); | |
1354 | ||
86799d2d | 1355 | #endif /* USE_LOCALE_NUMERIC and can find the radix char */ |
7d4bcc4a | 1356 | |
98994639 HS |
1357 | } |
1358 | ||
398eeea9 KW |
1359 | STATIC void |
1360 | S_new_numeric(pTHX_ const char *newnum) | |
98994639 | 1361 | { |
7d4bcc4a KW |
1362 | |
1363 | #ifndef USE_LOCALE_NUMERIC | |
1364 | ||
1365 | PERL_UNUSED_ARG(newnum); | |
1366 | ||
1367 | #else | |
0d071d52 | 1368 | |
291a84fb | 1369 | /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell |
0d071d52 KW |
1370 | * core Perl this and that 'newnum' is the name of the new locale. |
1371 | * It installs this locale as the current underlying default. | |
1372 | * | |
1373 | * The default locale and the C locale can be toggled between by use of the | |
5792c642 KW |
1374 | * set_numeric_underlying() and set_numeric_standard() functions, which |
1375 | * should probably not be called directly, but only via macros like | |
0d071d52 KW |
1376 | * SET_NUMERIC_STANDARD() in perl.h. |
1377 | * | |
1378 | * The toggling is necessary mainly so that a non-dot radix decimal point | |
1379 | * character can be output, while allowing internal calculations to use a | |
1380 | * dot. | |
1381 | * | |
1382 | * This sets several interpreter-level variables: | |
bb304765 | 1383 | * PL_numeric_name The underlying locale's name: a copy of 'newnum' |
892e6465 | 1384 | * PL_numeric_underlying A boolean indicating if the toggled state is such |
7738054c KW |
1385 | * that the current locale is the program's underlying |
1386 | * locale | |
1387 | * PL_numeric_standard An int indicating if the toggled state is such | |
4c68b815 KW |
1388 | * that the current locale is the C locale or |
1389 | * indistinguishable from the C locale. If non-zero, it | |
1390 | * is in C; if > 1, it means it may not be toggled away | |
7738054c | 1391 | * from C. |
4c68b815 KW |
1392 | * PL_numeric_underlying_is_standard A bool kept by this function |
1393 | * indicating that the underlying locale and the standard | |
1394 | * C locale are indistinguishable for the purposes of | |
1395 | * LC_NUMERIC. This happens when both of the above two | |
1396 | * variables are true at the same time. (Toggling is a | |
1397 | * no-op under these circumstances.) This variable is | |
1398 | * used to avoid having to recalculate. | |
398eeea9 | 1399 | */ |
0d071d52 | 1400 | |
b03f34cf | 1401 | char *save_newnum; |
98994639 HS |
1402 | |
1403 | if (! newnum) { | |
43c5f42d NC |
1404 | Safefree(PL_numeric_name); |
1405 | PL_numeric_name = NULL; | |
98994639 | 1406 | PL_numeric_standard = TRUE; |
892e6465 | 1407 | PL_numeric_underlying = TRUE; |
4c68b815 | 1408 | PL_numeric_underlying_is_standard = TRUE; |
98994639 HS |
1409 | return; |
1410 | } | |
1411 | ||
b03f34cf | 1412 | save_newnum = stdize_locale(savepv(newnum)); |
892e6465 | 1413 | PL_numeric_underlying = TRUE; |
4c68b815 KW |
1414 | PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum); |
1415 | ||
69c5e0db KW |
1416 | #ifndef TS_W32_BROKEN_LOCALECONV |
1417 | ||
4c68b815 | 1418 | /* If its name isn't C nor POSIX, it could still be indistinguishable from |
69c5e0db KW |
1419 | * them. But on broken Windows systems calling my_nl_langinfo() for |
1420 | * THOUSEP can currently (but rarely) cause a race, so avoid doing that, | |
1421 | * and just always change the locale if not C nor POSIX on those systems */ | |
4c68b815 | 1422 | if (! PL_numeric_standard) { |
4e6826bf | 1423 | PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR, |
4c68b815 | 1424 | FALSE /* Don't toggle locale */ )) |
4e6826bf | 1425 | && strEQ("", my_nl_langinfo(THOUSEP, FALSE))); |
4c68b815 | 1426 | } |
abe1abcf | 1427 | |
69c5e0db KW |
1428 | #endif |
1429 | ||
4c68b815 | 1430 | /* Save the new name if it isn't the same as the previous one, if any */ |
b03f34cf | 1431 | if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) { |
98994639 | 1432 | Safefree(PL_numeric_name); |
b03f34cf | 1433 | PL_numeric_name = save_newnum; |
b03f34cf | 1434 | } |
abe1abcf KW |
1435 | else { |
1436 | Safefree(save_newnum); | |
1437 | } | |
4c28b29c | 1438 | |
4c68b815 KW |
1439 | PL_numeric_underlying_is_standard = PL_numeric_standard; |
1440 | ||
7e5377f7 | 1441 | # ifdef HAS_POSIX_2008_LOCALE |
e1aa2579 KW |
1442 | |
1443 | PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK, | |
1444 | PL_numeric_name, | |
1445 | PL_underlying_numeric_obj); | |
1446 | ||
1447 | #endif | |
1448 | ||
4c68b815 KW |
1449 | if (DEBUG_L_TEST || debug_initialization) { |
1450 | PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name); | |
1451 | } | |
1452 | ||
4c28b29c KW |
1453 | /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't |
1454 | * have to worry about the radix being a non-dot. (Core operations that | |
1455 | * need the underlying locale change to it temporarily). */ | |
84f1d29f KW |
1456 | if (PL_numeric_standard) { |
1457 | set_numeric_radix(0); | |
1458 | } | |
1459 | else { | |
1460 | set_numeric_standard(); | |
1461 | } | |
4c28b29c | 1462 | |
98994639 | 1463 | #endif /* USE_LOCALE_NUMERIC */ |
7d4bcc4a | 1464 | |
98994639 HS |
1465 | } |
1466 | ||
1467 | void | |
1468 | Perl_set_numeric_standard(pTHX) | |
1469 | { | |
7d4bcc4a | 1470 | |
98994639 | 1471 | #ifdef USE_LOCALE_NUMERIC |
7d4bcc4a | 1472 | |
28c1bf33 KW |
1473 | /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like |
1474 | * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The | |
1475 | * macro avoids calling this routine if toggling isn't necessary according | |
1476 | * to our records (which could be wrong if some XS code has changed the | |
1477 | * locale behind our back) */ | |
0d071d52 | 1478 | |
7d4bcc4a KW |
1479 | # ifdef DEBUGGING |
1480 | ||
2fcc0ca9 KW |
1481 | if (DEBUG_L_TEST || debug_initialization) { |
1482 | PerlIO_printf(Perl_debug_log, | |
7bb8f702 | 1483 | "Setting LC_NUMERIC locale to standard C\n"); |
2fcc0ca9 | 1484 | } |
98994639 | 1485 | |
7d4bcc4a | 1486 | # endif |
7bb8f702 KW |
1487 | |
1488 | do_setlocale_c(LC_NUMERIC, "C"); | |
1489 | PL_numeric_standard = TRUE; | |
1490 | PL_numeric_underlying = PL_numeric_underlying_is_standard; | |
1491 | set_numeric_radix(0); | |
1492 | ||
98994639 | 1493 | #endif /* USE_LOCALE_NUMERIC */ |
7d4bcc4a | 1494 | |
98994639 HS |
1495 | } |
1496 | ||
1497 | void | |
5792c642 | 1498 | Perl_set_numeric_underlying(pTHX) |
98994639 | 1499 | { |
7d4bcc4a | 1500 | |
98994639 | 1501 | #ifdef USE_LOCALE_NUMERIC |
7d4bcc4a | 1502 | |
28c1bf33 | 1503 | /* Toggle the LC_NUMERIC locale to the current underlying default. Most |
7d4bcc4a KW |
1504 | * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h |
1505 | * instead of calling this directly. The macro avoids calling this routine | |
1506 | * if toggling isn't necessary according to our records (which could be | |
1507 | * wrong if some XS code has changed the locale behind our back) */ | |
a9b8c0d8 | 1508 | |
7d4bcc4a KW |
1509 | # ifdef DEBUGGING |
1510 | ||
2fcc0ca9 KW |
1511 | if (DEBUG_L_TEST || debug_initialization) { |
1512 | PerlIO_printf(Perl_debug_log, | |
7bb8f702 | 1513 | "Setting LC_NUMERIC locale to %s\n", |
2fcc0ca9 KW |
1514 | PL_numeric_name); |
1515 | } | |
98994639 | 1516 | |
7d4bcc4a | 1517 | # endif |
7bb8f702 KW |
1518 | |
1519 | do_setlocale_c(LC_NUMERIC, PL_numeric_name); | |
1520 | PL_numeric_standard = PL_numeric_underlying_is_standard; | |
1521 | PL_numeric_underlying = TRUE; | |
1522 | set_numeric_radix(! PL_numeric_standard); | |
1523 | ||
98994639 | 1524 | #endif /* USE_LOCALE_NUMERIC */ |
7d4bcc4a | 1525 | |
98994639 HS |
1526 | } |
1527 | ||
1528 | /* | |
1529 | * Set up for a new ctype locale. | |
1530 | */ | |
a4f00dcc KW |
1531 | STATIC void |
1532 | S_new_ctype(pTHX_ const char *newctype) | |
98994639 | 1533 | { |
7d4bcc4a KW |
1534 | |
1535 | #ifndef USE_LOCALE_CTYPE | |
1536 | ||
7d4bcc4a KW |
1537 | PERL_UNUSED_ARG(newctype); |
1538 | PERL_UNUSED_CONTEXT; | |
1539 | ||
1540 | #else | |
0d071d52 | 1541 | |
291a84fb | 1542 | /* Called after each libc setlocale() call affecting LC_CTYPE, to tell |
0d071d52 KW |
1543 | * core Perl this and that 'newctype' is the name of the new locale. |
1544 | * | |
1545 | * This function sets up the folding arrays for all 256 bytes, assuming | |
1546 | * that tofold() is tolc() since fold case is not a concept in POSIX, | |
1547 | * | |
1548 | * Any code changing the locale (outside this file) should use | |
9aac5db8 KW |
1549 | * Perl_setlocale or POSIX::setlocale, which call this function. Therefore |
1550 | * this function should be called directly only from this file and from | |
0d071d52 KW |
1551 | * POSIX::setlocale() */ |
1552 | ||
27da23d5 | 1553 | dVAR; |
013f4e03 | 1554 | unsigned int i; |
98994639 | 1555 | |
8b7358b9 KW |
1556 | /* Don't check for problems if we are suppressing the warnings */ |
1557 | bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST); | |
30d8090d | 1558 | bool maybe_utf8_turkic = FALSE; |
8b7358b9 | 1559 | |
7918f24d NC |
1560 | PERL_ARGS_ASSERT_NEW_CTYPE; |
1561 | ||
215c5139 KW |
1562 | /* We will replace any bad locale warning with 1) nothing if the new one is |
1563 | * ok; or 2) a new warning for the bad new locale */ | |
1564 | if (PL_warn_locale) { | |
1565 | SvREFCNT_dec_NN(PL_warn_locale); | |
1566 | PL_warn_locale = NULL; | |
1567 | } | |
1568 | ||
c1284011 | 1569 | PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE); |
31f05a37 KW |
1570 | |
1571 | /* A UTF-8 locale gets standard rules. But note that code still has to | |
1572 | * handle this specially because of the three problematic code points */ | |
1573 | if (PL_in_utf8_CTYPE_locale) { | |
1574 | Copy(PL_fold_latin1, PL_fold_locale, 256, U8); | |
30d8090d KW |
1575 | |
1576 | /* UTF-8 locales can have special handling for 'I' and 'i' if they are | |
1577 | * Turkic. Make sure these two are the only anomalies. (We don't use | |
1578 | * towupper and towlower because they aren't in C89.) */ | |
5b64f24c KW |
1579 | |
1580 | #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER) | |
1581 | ||
1582 | if (towupper('i') == 0x130 && towlower('I') == 0x131) { | |
1583 | ||
1584 | #else | |
1585 | ||
30d8090d | 1586 | if (toupper('i') == 'i' && tolower('I') == 'I') { |
5b64f24c KW |
1587 | |
1588 | #endif | |
30d8090d KW |
1589 | check_for_problems = TRUE; |
1590 | maybe_utf8_turkic = TRUE; | |
1591 | } | |
31f05a37 | 1592 | } |
8b7358b9 KW |
1593 | |
1594 | /* We don't populate the other lists if a UTF-8 locale, but do check that | |
1595 | * everything works as expected, unless checking turned off */ | |
1596 | if (check_for_problems || ! PL_in_utf8_CTYPE_locale) { | |
8c6180a9 KW |
1597 | /* Assume enough space for every character being bad. 4 spaces each |
1598 | * for the 94 printable characters that are output like "'x' "; and 5 | |
1599 | * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating | |
1600 | * NUL */ | |
8b7358b9 | 1601 | char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' }; |
8c6180a9 KW |
1602 | bool multi_byte_locale = FALSE; /* Assume is a single-byte locale |
1603 | to start */ | |
1604 | unsigned int bad_count = 0; /* Count of bad characters */ | |
1605 | ||
baa60164 | 1606 | for (i = 0; i < 256; i++) { |
8b7358b9 | 1607 | if (! PL_in_utf8_CTYPE_locale) { |
bd6d0898 KW |
1608 | if (isupper(i)) |
1609 | PL_fold_locale[i] = (U8) tolower(i); | |
1610 | else if (islower(i)) | |
1611 | PL_fold_locale[i] = (U8) toupper(i); | |
1612 | else | |
1613 | PL_fold_locale[i] = (U8) i; | |
8b7358b9 | 1614 | } |
8c6180a9 KW |
1615 | |
1616 | /* If checking for locale problems, see if the native ASCII-range | |
1617 | * printables plus \n and \t are in their expected categories in | |
1618 | * the new locale. If not, this could mean big trouble, upending | |
1619 | * Perl's and most programs' assumptions, like having a | |
1620 | * metacharacter with special meaning become a \w. Fortunately, | |
1621 | * it's very rare to find locales that aren't supersets of ASCII | |
1622 | * nowadays. It isn't a problem for most controls to be changed | |
1623 | * into something else; we check only \n and \t, though perhaps \r | |
1624 | * could be an issue as well. */ | |
7d4bcc4a | 1625 | if ( check_for_problems |
8c6180a9 KW |
1626 | && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n')) |
1627 | { | |
8b7358b9 | 1628 | bool is_bad = FALSE; |
6726b4f4 | 1629 | char name[4] = { '\0' }; |
8b7358b9 KW |
1630 | |
1631 | /* Convert the name into a string */ | |
6726b4f4 | 1632 | if (isGRAPH_A(i)) { |
8b7358b9 KW |
1633 | name[0] = i; |
1634 | name[1] = '\0'; | |
1635 | } | |
1636 | else if (i == '\n') { | |
6726b4f4 KW |
1637 | my_strlcpy(name, "\\n", sizeof(name)); |
1638 | } | |
1639 | else if (i == '\t') { | |
1640 | my_strlcpy(name, "\\t", sizeof(name)); | |
8b7358b9 KW |
1641 | } |
1642 | else { | |
6726b4f4 KW |
1643 | assert(i == ' '); |
1644 | my_strlcpy(name, "' '", sizeof(name)); | |
8b7358b9 KW |
1645 | } |
1646 | ||
1647 | /* Check each possibe class */ | |
1648 | if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) { | |
1649 | is_bad = TRUE; | |
1650 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1651 | "isalnum('%s') unexpectedly is %d\n", | |
1652 | name, cBOOL(isalnum(i)))); | |
1653 | } | |
1654 | if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) { | |
1655 | is_bad = TRUE; | |
1656 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1657 | "isalpha('%s') unexpectedly is %d\n", | |
1658 | name, cBOOL(isalpha(i)))); | |
1659 | } | |
1660 | if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) { | |
1661 | is_bad = TRUE; | |
1662 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1663 | "isdigit('%s') unexpectedly is %d\n", | |
1664 | name, cBOOL(isdigit(i)))); | |
1665 | } | |
1666 | if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) { | |
1667 | is_bad = TRUE; | |
1668 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1669 | "isgraph('%s') unexpectedly is %d\n", | |
1670 | name, cBOOL(isgraph(i)))); | |
1671 | } | |
1672 | if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) { | |
1673 | is_bad = TRUE; | |
1674 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1675 | "islower('%s') unexpectedly is %d\n", | |
1676 | name, cBOOL(islower(i)))); | |
1677 | } | |
1678 | if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) { | |
1679 | is_bad = TRUE; | |
1680 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1681 | "isprint('%s') unexpectedly is %d\n", | |
1682 | name, cBOOL(isprint(i)))); | |
1683 | } | |
1684 | if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) { | |
1685 | is_bad = TRUE; | |
1686 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1687 | "ispunct('%s') unexpectedly is %d\n", | |
1688 | name, cBOOL(ispunct(i)))); | |
1689 | } | |
1690 | if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) { | |
1691 | is_bad = TRUE; | |
1692 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1693 | "isspace('%s') unexpectedly is %d\n", | |
1694 | name, cBOOL(isspace(i)))); | |
1695 | } | |
1696 | if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) { | |
1697 | is_bad = TRUE; | |
1698 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1699 | "isupper('%s') unexpectedly is %d\n", | |
1700 | name, cBOOL(isupper(i)))); | |
1701 | } | |
1702 | if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) { | |
1703 | is_bad = TRUE; | |
1704 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1705 | "isxdigit('%s') unexpectedly is %d\n", | |
1706 | name, cBOOL(isxdigit(i)))); | |
1707 | } | |
1708 | if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) { | |
1709 | is_bad = TRUE; | |
1710 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1711 | "tolower('%s')=0x%x instead of the expected 0x%x\n", | |
1712 | name, tolower(i), (int) toLOWER_A(i))); | |
1713 | } | |
1714 | if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) { | |
1715 | is_bad = TRUE; | |
1716 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1717 | "toupper('%s')=0x%x instead of the expected 0x%x\n", | |
1718 | name, toupper(i), (int) toUPPER_A(i))); | |
1719 | } | |
1720 | if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) { | |
1721 | is_bad = TRUE; | |
1722 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
1723 | "'\\n' (=%02X) is not a control\n", (int) i)); | |
1724 | } | |
1725 | ||
1726 | /* Add to the list; Separate multiple entries with a blank */ | |
1727 | if (is_bad) { | |
1728 | if (bad_count) { | |
1729 | my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list)); | |
8c6180a9 | 1730 | } |
8b7358b9 KW |
1731 | my_strlcat(bad_chars_list, name, sizeof(bad_chars_list)); |
1732 | bad_count++; | |
8c6180a9 KW |
1733 | } |
1734 | } | |
1735 | } | |
1736 | ||
30d8090d KW |
1737 | if (bad_count == 2 && maybe_utf8_turkic) { |
1738 | bad_count = 0; | |
1739 | *bad_chars_list = '\0'; | |
1740 | PL_fold_locale['I'] = 'I'; | |
1741 | PL_fold_locale['i'] = 'i'; | |
1742 | PL_in_utf8_turkic_locale = TRUE; | |
1743 | DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n", | |
1744 | __FILE__, __LINE__, newctype)); | |
1745 | } | |
1746 | else { | |
b8df1494 | 1747 | PL_in_utf8_turkic_locale = FALSE; |
30d8090d | 1748 | } |
b8df1494 | 1749 | |
7d4bcc4a KW |
1750 | # ifdef MB_CUR_MAX |
1751 | ||
8c6180a9 | 1752 | /* We only handle single-byte locales (outside of UTF-8 ones; so if |
d35fca5f | 1753 | * this locale requires more than one byte, there are going to be |
8c6180a9 | 1754 | * problems. */ |
9c8a6dc2 KW |
1755 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
1756 | "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n", | |
1757 | __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX)); | |
1758 | ||
8b7358b9 KW |
1759 | if ( check_for_problems && MB_CUR_MAX > 1 |
1760 | && ! PL_in_utf8_CTYPE_locale | |
ba1a4362 KW |
1761 | |
1762 | /* Some platforms return MB_CUR_MAX > 1 for even the "C" | |
1763 | * locale. Just assume that the implementation for them (plus | |
1764 | * for POSIX) is correct and the > 1 value is spurious. (Since | |
1765 | * these are specially handled to never be considered UTF-8 | |
1766 | * locales, as long as this is the only problem, everything | |
1767 | * should work fine */ | |
1768 | && strNE(newctype, "C") && strNE(newctype, "POSIX")) | |
1769 | { | |
8c6180a9 KW |
1770 | multi_byte_locale = TRUE; |
1771 | } | |
7d4bcc4a KW |
1772 | |
1773 | # endif | |
8c6180a9 | 1774 | |
30d8090d KW |
1775 | /* If we found problems and we want them output, do so */ |
1776 | if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) | |
1777 | && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST))) | |
1778 | { | |
8b7358b9 KW |
1779 | if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) { |
1780 | PL_warn_locale = Perl_newSVpvf(aTHX_ | |
1781 | "Locale '%s' contains (at least) the following characters" | |
578a6a87 KW |
1782 | " which have\nunexpected meanings: %s\nThe Perl program" |
1783 | " will use the expected meanings", | |
8b7358b9 | 1784 | newctype, bad_chars_list); |
8b7358b9 KW |
1785 | } |
1786 | else { | |
1787 | PL_warn_locale = Perl_newSVpvf(aTHX_ | |
8c6180a9 | 1788 | "Locale '%s' may not work well.%s%s%s\n", |
780fcc9f | 1789 | newctype, |
8c6180a9 KW |
1790 | (multi_byte_locale) |
1791 | ? " Some characters in it are not recognized by" | |
1792 | " Perl." | |
1793 | : "", | |
1794 | (bad_count) | |
1795 | ? "\nThe following characters (and maybe others)" | |
1796 | " may not have the same meaning as the Perl" | |
1797 | " program expects:\n" | |
1798 | : "", | |
1799 | (bad_count) | |
1800 | ? bad_chars_list | |
1801 | : "" | |
1802 | ); | |
8b7358b9 KW |
1803 | } |
1804 | ||
b119c2be KW |
1805 | # ifdef HAS_NL_LANGINFO |
1806 | ||
1807 | Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s", | |
1808 | /* parameter FALSE is a don't care here */ | |
4e6826bf | 1809 | my_nl_langinfo(CODESET, FALSE)); |
b119c2be KW |
1810 | |
1811 | # endif | |
1812 | ||
8b7358b9 KW |
1813 | Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n"); |
1814 | ||
cc9eaeb0 | 1815 | /* If we are actually in the scope of the locale or are debugging, |
bddebb56 KW |
1816 | * output the message now. If not in that scope, we save the |
1817 | * message to be output at the first operation using this locale, | |
1818 | * if that actually happens. Most programs don't use locales, so | |
1819 | * they are immune to bad ones. */ | |
cc9eaeb0 | 1820 | if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) { |
780fcc9f | 1821 | |
780fcc9f KW |
1822 | /* The '0' below suppresses a bogus gcc compiler warning */ |
1823 | Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0); | |
bddebb56 | 1824 | |
bddebb56 KW |
1825 | if (IN_LC(LC_CTYPE)) { |
1826 | SvREFCNT_dec_NN(PL_warn_locale); | |
1827 | PL_warn_locale = NULL; | |
1828 | } | |
780fcc9f | 1829 | } |
baa60164 | 1830 | } |
31f05a37 | 1831 | } |
98994639 HS |
1832 | |
1833 | #endif /* USE_LOCALE_CTYPE */ | |
7d4bcc4a | 1834 | |
98994639 HS |
1835 | } |
1836 | ||
98994639 | 1837 | void |
2726666d KW |
1838 | Perl__warn_problematic_locale() |
1839 | { | |
2726666d KW |
1840 | |
1841 | #ifdef USE_LOCALE_CTYPE | |
1842 | ||
5f04a188 KW |
1843 | dTHX; |
1844 | ||
1845 | /* Internal-to-core function that outputs the message in PL_warn_locale, | |
1846 | * and then NULLS it. Should be called only through the macro | |
1847 | * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */ | |
1848 | ||
2726666d | 1849 | if (PL_warn_locale) { |
2726666d KW |
1850 | Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), |
1851 | SvPVX(PL_warn_locale), | |
1852 | 0 /* dummy to avoid compiler warning */ ); | |
2726666d KW |
1853 | SvREFCNT_dec_NN(PL_warn_locale); |
1854 | PL_warn_locale = NULL; | |
1855 | } | |
1856 | ||
1857 | #endif | |
1858 | ||
1859 | } | |
1860 | ||
a4f00dcc KW |
1861 | STATIC void |
1862 | S_new_collate(pTHX_ const char *newcoll) | |
98994639 | 1863 | { |
7d4bcc4a KW |
1864 | |
1865 | #ifndef USE_LOCALE_COLLATE | |
1866 | ||
1867 | PERL_UNUSED_ARG(newcoll); | |
1868 | PERL_UNUSED_CONTEXT; | |
1869 | ||
1870 | #else | |
0d071d52 | 1871 | |
291a84fb | 1872 | /* Called after each libc setlocale() call affecting LC_COLLATE, to tell |
0d071d52 KW |
1873 | * core Perl this and that 'newcoll' is the name of the new locale. |
1874 | * | |
d35fca5f KW |
1875 | * The design of locale collation is that every locale change is given an |
1876 | * index 'PL_collation_ix'. The first time a string particpates in an | |
1877 | * operation that requires collation while locale collation is active, it | |
1878 | * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That | |
1879 | * magic includes the collation index, and the transformation of the string | |
1880 | * by strxfrm(), q.v. That transformation is used when doing comparisons, | |
1881 | * instead of the string itself. If a string changes, the magic is | |
1882 | * cleared. The next time the locale changes, the index is incremented, | |
1883 | * and so we know during a comparison that the transformation is not | |
1884 | * necessarily still valid, and so is recomputed. Note that if the locale | |
1885 | * changes enough times, the index could wrap (a U32), and it is possible | |
1886 | * that a transformation would improperly be considered valid, leading to | |
1887 | * an unlikely bug */ | |
0d071d52 | 1888 | |
98994639 HS |
1889 | if (! newcoll) { |
1890 | if (PL_collation_name) { | |
1891 | ++PL_collation_ix; | |
1892 | Safefree(PL_collation_name); | |
1893 | PL_collation_name = NULL; | |
1894 | } | |
1895 | PL_collation_standard = TRUE; | |
00bf60ca | 1896 | is_standard_collation: |
98994639 HS |
1897 | PL_collxfrm_base = 0; |
1898 | PL_collxfrm_mult = 2; | |
165a1c52 | 1899 | PL_in_utf8_COLLATE_locale = FALSE; |
f28f4d2a | 1900 | PL_strxfrm_NUL_replacement = '\0'; |
a4a439fb | 1901 | PL_strxfrm_max_cp = 0; |
98994639 HS |
1902 | return; |
1903 | } | |
1904 | ||
d35fca5f | 1905 | /* If this is not the same locale as currently, set the new one up */ |
98994639 HS |
1906 | if (! PL_collation_name || strNE(PL_collation_name, newcoll)) { |
1907 | ++PL_collation_ix; | |
1908 | Safefree(PL_collation_name); | |
1909 | PL_collation_name = stdize_locale(savepv(newcoll)); | |
a39edc4c | 1910 | PL_collation_standard = isNAME_C_OR_POSIX(newcoll); |
00bf60ca KW |
1911 | if (PL_collation_standard) { |
1912 | goto is_standard_collation; | |
1913 | } | |
98994639 | 1914 | |
165a1c52 | 1915 | PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE); |
f28f4d2a | 1916 | PL_strxfrm_NUL_replacement = '\0'; |
a4a439fb | 1917 | PL_strxfrm_max_cp = 0; |
165a1c52 | 1918 | |
59c018b9 KW |
1919 | /* A locale collation definition includes primary, secondary, tertiary, |
1920 | * etc. weights for each character. To sort, the primary weights are | |
1921 | * used, and only if they compare equal, then the secondary weights are | |
1922 | * used, and only if they compare equal, then the tertiary, etc. | |
1923 | * | |
1924 | * strxfrm() works by taking the input string, say ABC, and creating an | |
1925 | * output transformed string consisting of first the primary weights, | |
1926 | * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the | |
1927 | * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters | |
1928 | * may not have weights at every level. In our example, let's say B | |
1929 | * doesn't have a tertiary weight, and A doesn't have a secondary | |
1930 | * weight. The constructed string is then going to be | |
1931 | * A¹B¹C¹ B²C² A³C³ .... | |
1932 | * This has the desired effect that strcmp() will look at the secondary | |
1933 | * or tertiary weights only if the strings compare equal at all higher | |
1934 | * priority weights. The spaces shown here, like in | |
c342d20e | 1935 | * "A¹B¹C¹ A²B²C² " |
59c018b9 KW |
1936 | * are not just for readability. In the general case, these must |
1937 | * actually be bytes, which we will call here 'separator weights'; and | |
1938 | * they must be smaller than any other weight value, but since these | |
1939 | * are C strings, only the terminating one can be a NUL (some | |
1940 | * implementations may include a non-NUL separator weight just before | |
1941 | * the NUL). Implementations tend to reserve 01 for the separator | |
1942 | * weights. They are needed so that a shorter string's secondary | |
1943 | * weights won't be misconstrued as primary weights of a longer string, | |
1944 | * etc. By making them smaller than any other weight, the shorter | |
1945 | * string will sort first. (Actually, if all secondary weights are | |
1946 | * smaller than all primary ones, there is no need for a separator | |
1947 | * weight between those two levels, etc.) | |
1948 | * | |
1949 | * The length of the transformed string is roughly a linear function of | |
1950 | * the input string. It's not exactly linear because some characters | |
1951 | * don't have weights at all levels. When we call strxfrm() we have to | |
1952 | * allocate some memory to hold the transformed string. The | |
1953 | * calculations below try to find coefficients 'm' and 'b' for this | |
1954 | * locale so that m*x + b equals how much space we need, given the size | |
1955 | * of the input string in 'x'. If we calculate too small, we increase | |
1956 | * the size as needed, and call strxfrm() again, but it is better to | |
1957 | * get it right the first time to avoid wasted expensive string | |
1958 | * transformations. */ | |
1959 | ||
98994639 | 1960 | { |
79f120c8 KW |
1961 | /* We use the string below to find how long the tranformation of it |
1962 | * is. Almost all locales are supersets of ASCII, or at least the | |
1963 | * ASCII letters. We use all of them, half upper half lower, | |
1964 | * because if we used fewer, we might hit just the ones that are | |
1965 | * outliers in a particular locale. Most of the strings being | |
1966 | * collated will contain a preponderance of letters, and even if | |
1967 | * they are above-ASCII, they are likely to have the same number of | |
1968 | * weight levels as the ASCII ones. It turns out that digits tend | |
1969 | * to have fewer levels, and some punctuation has more, but those | |
1970 | * are relatively sparse in text, and khw believes this gives a | |
1971 | * reasonable result, but it could be changed if experience so | |
1972 | * dictates. */ | |
1973 | const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz"; | |
1974 | char * x_longer; /* Transformed 'longer' */ | |
1975 | Size_t x_len_longer; /* Length of 'x_longer' */ | |
1976 | ||
1977 | char * x_shorter; /* We also transform a substring of 'longer' */ | |
1978 | Size_t x_len_shorter; | |
1979 | ||
a4a439fb | 1980 | /* _mem_collxfrm() is used get the transformation (though here we |
79f120c8 KW |
1981 | * are interested only in its length). It is used because it has |
1982 | * the intelligence to handle all cases, but to work, it needs some | |
1983 | * values of 'm' and 'b' to get it started. For the purposes of | |
1984 | * this calculation we use a very conservative estimate of 'm' and | |
1985 | * 'b'. This assumes a weight can be multiple bytes, enough to | |
1986 | * hold any UV on the platform, and there are 5 levels, 4 weight | |
1987 | * bytes, and a trailing NUL. */ | |
1988 | PL_collxfrm_base = 5; | |
1989 | PL_collxfrm_mult = 5 * sizeof(UV); | |
1990 | ||
1991 | /* Find out how long the transformation really is */ | |
a4a439fb KW |
1992 | x_longer = _mem_collxfrm(longer, |
1993 | sizeof(longer) - 1, | |
1994 | &x_len_longer, | |
1995 | ||
1996 | /* We avoid converting to UTF-8 in the | |
1997 | * called function by telling it the | |
1998 | * string is in UTF-8 if the locale is a | |
1999 | * UTF-8 one. Since the string passed | |
2000 | * here is invariant under UTF-8, we can | |
2001 | * claim it's UTF-8 even though it isn't. | |
2002 | * */ | |
2003 | PL_in_utf8_COLLATE_locale); | |
79f120c8 KW |
2004 | Safefree(x_longer); |
2005 | ||
2006 | /* Find out how long the transformation of a substring of 'longer' | |
2007 | * is. Together the lengths of these transformations are | |
2008 | * sufficient to calculate 'm' and 'b'. The substring is all of | |
2009 | * 'longer' except the first character. This minimizes the chances | |
2010 | * of being swayed by outliers */ | |
a4a439fb | 2011 | x_shorter = _mem_collxfrm(longer + 1, |
79f120c8 | 2012 | sizeof(longer) - 2, |
a4a439fb KW |
2013 | &x_len_shorter, |
2014 | PL_in_utf8_COLLATE_locale); | |
79f120c8 KW |
2015 | Safefree(x_shorter); |
2016 | ||
2017 | /* If the results are nonsensical for this simple test, the whole | |
2018 | * locale definition is suspect. Mark it so that locale collation | |
2019 | * is not active at all for it. XXX Should we warn? */ | |
2020 | if ( x_len_shorter == 0 | |
2021 | || x_len_longer == 0 | |
2022 | || x_len_shorter >= x_len_longer) | |
2023 | { | |
2024 | PL_collxfrm_mult = 0; | |
2025 | PL_collxfrm_base = 0; | |
2026 | } | |
2027 | else { | |
2028 | SSize_t base; /* Temporary */ | |
2029 | ||
2030 | /* We have both: m * strlen(longer) + b = x_len_longer | |
2031 | * m * strlen(shorter) + b = x_len_shorter; | |
2032 | * subtracting yields: | |
2033 | * m * (strlen(longer) - strlen(shorter)) | |
2034 | * = x_len_longer - x_len_shorter | |
2035 | * But we have set things up so that 'shorter' is 1 byte smaller | |
2036 | * than 'longer'. Hence: | |
2037 | * m = x_len_longer - x_len_shorter | |
2038 | * | |
2039 | * But if something went wrong, make sure the multiplier is at | |
2040 | * least 1. | |
2041 | */ | |
2042 | if (x_len_longer > x_len_shorter) { | |
2043 | PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter; | |
2044 | } | |
2045 | else { | |
2046 | PL_collxfrm_mult = 1; | |
2047 | } | |
2048 | ||
2049 | /* mx + b = len | |
2050 | * so: b = len - mx | |
2051 | * but in case something has gone wrong, make sure it is | |
2052 | * non-negative */ | |
2053 | base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1); | |
2054 | if (base < 0) { | |
2055 | base = 0; | |
2056 | } | |
2057 | ||
2058 | /* Add 1 for the trailing NUL */ | |
2059 | PL_collxfrm_base = base + 1; | |
2060 | } | |
58eebef2 | 2061 | |
7d4bcc4a KW |
2062 | # ifdef DEBUGGING |
2063 | ||
58eebef2 KW |
2064 | if (DEBUG_L_TEST || debug_initialization) { |
2065 | PerlIO_printf(Perl_debug_log, | |
b07929e4 KW |
2066 | "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, " |
2067 | "x_len_longer=%zu," | |
2068 | " collate multipler=%zu, collate base=%zu\n", | |
58eebef2 KW |
2069 | __FILE__, __LINE__, |
2070 | PL_in_utf8_COLLATE_locale, | |
2071 | x_len_shorter, x_len_longer, | |
2072 | PL_collxfrm_mult, PL_collxfrm_base); | |
2073 | } | |
7d4bcc4a KW |
2074 | # endif |
2075 | ||
98994639 HS |
2076 | } |
2077 | } | |
2078 | ||
2079 | #endif /* USE_LOCALE_COLLATE */ | |
7d4bcc4a | 2080 | |
98994639 HS |
2081 | } |
2082 | ||
d36adde0 KW |
2083 | #endif |
2084 | ||
d2b24094 | 2085 | #ifdef WIN32 |
b8cc575c | 2086 | |
6c332036 TC |
2087 | #define USE_WSETLOCALE |
2088 | ||
2089 | #ifdef USE_WSETLOCALE | |
2090 | ||
2091 | STATIC char * | |
2092 | S_wrap_wsetlocale(pTHX_ int category, const char *locale) { | |
2093 | wchar_t *wlocale; | |
2094 | wchar_t *wresult; | |
2095 | char *result; | |
2096 | ||
2097 | if (locale) { | |
2098 | int req_size = | |
2099 | MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0); | |
2100 | ||
2101 | if (!req_size) { | |
2102 | errno = EINVAL; | |
2103 | return NULL; | |
2104 | } | |
2105 | ||
2106 | Newx(wlocale, req_size, wchar_t); | |
2107 | if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) { | |
2108 | Safefree(wlocale); | |
2109 | errno = EINVAL; | |
2110 | return NULL; | |
2111 | } | |
2112 | } | |
2113 | else { | |
2114 | wlocale = NULL; | |
2115 | } | |
2116 | wresult = _wsetlocale(category, wlocale); | |
2117 | Safefree(wlocale); | |
2118 | if (wresult) { | |
2119 | int req_size = | |
2120 | WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL); | |
2121 | Newx(result, req_size, char); | |
2122 | SAVEFREEPV(result); /* is there something better we can do here? */ | |
2123 | if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1, | |
2124 | result, req_size, NULL, NULL)) { | |
2125 | errno = EINVAL; | |
2126 | return NULL; | |
2127 | } | |
2128 | } | |
2129 | else { | |
2130 | result = NULL; | |
2131 | } | |
2132 | ||
2133 | return result; | |
2134 | } | |
2135 | ||
2136 | #endif | |
2137 | ||
a4f00dcc | 2138 | STATIC char * |
b8cc575c | 2139 | S_win32_setlocale(pTHX_ int category, const char* locale) |
b385bb4d KW |
2140 | { |
2141 | /* This, for Windows, emulates POSIX setlocale() behavior. There is no | |
7d4bcc4a KW |
2142 | * difference between the two unless the input locale is "", which normally |
2143 | * means on Windows to get the machine default, which is set via the | |
2144 | * computer's "Regional and Language Options" (or its current equivalent). | |
2145 | * In POSIX, it instead means to find the locale from the user's | |
2146 | * environment. This routine changes the Windows behavior to first look in | |
2147 | * the environment, and, if anything is found, use that instead of going to | |
2148 | * the machine default. If there is no environment override, the machine | |
2149 | * default is used, by calling the real setlocale() with "". | |
2150 | * | |
2151 | * The POSIX behavior is to use the LC_ALL variable if set; otherwise to | |
2152 | * use the particular category's variable if set; otherwise to use the LANG | |
2153 | * variable. */ | |
b385bb4d | 2154 | |
175c4cf9 | 2155 | bool override_LC_ALL = FALSE; |
89f7b9aa | 2156 | char * result; |
e5f10d49 | 2157 | unsigned int i; |
89f7b9aa | 2158 | |
b385bb4d | 2159 | if (locale && strEQ(locale, "")) { |
7d4bcc4a KW |
2160 | |
2161 | # ifdef LC_ALL | |
2162 | ||
b385bb4d KW |
2163 | locale = PerlEnv_getenv("LC_ALL"); |
2164 | if (! locale) { | |
e5f10d49 KW |
2165 | if (category == LC_ALL) { |
2166 | override_LC_ALL = TRUE; | |
2167 | } | |
2168 | else { | |
7d4bcc4a KW |
2169 | |
2170 | # endif | |
7d4bcc4a | 2171 | |
e5f10d49 KW |
2172 | for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) { |
2173 | if (category == categories[i]) { | |
2174 | locale = PerlEnv_getenv(category_names[i]); | |
2175 | goto found_locale; | |
2176 | } | |
2177 | } | |
7d4bcc4a | 2178 | |
b385bb4d | 2179 | locale = PerlEnv_getenv("LANG"); |
481465ea | 2180 | if (! locale) { |
b385bb4d KW |
2181 | locale = ""; |
2182 | } | |
e5f10d49 KW |
2183 | |
2184 | found_locale: ; | |
7d4bcc4a KW |
2185 | |
2186 | # ifdef LC_ALL | |
2187 | ||
e5f10d49 | 2188 | } |
b385bb4d | 2189 | } |
7d4bcc4a KW |
2190 | |
2191 | # endif | |
2192 | ||
b385bb4d KW |
2193 | } |
2194 | ||
6c332036 TC |
2195 | #ifdef USE_WSETLOCALE |
2196 | result = S_wrap_wsetlocale(aTHX_ category, locale); | |
2197 | #else | |
89f7b9aa | 2198 | result = setlocale(category, locale); |
6c332036 | 2199 | #endif |
48015184 KW |
2200 | DEBUG_L(STMT_START { |
2201 | dSAVE_ERRNO; | |
2202 | PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__, | |
2203 | setlocale_debug_string(category, locale, result)); | |
2204 | RESTORE_ERRNO; | |
2205 | } STMT_END); | |
89f7b9aa | 2206 | |
481465ea | 2207 | if (! override_LC_ALL) { |
89f7b9aa KW |
2208 | return result; |
2209 | } | |
2210 | ||
dfd77d7a | 2211 | /* Here the input category was LC_ALL, and we have set it to what is in the |
481465ea KW |
2212 | * LANG variable or the system default if there is no LANG. But these have |
2213 | * lower priority than the other LC_foo variables, so override it for each | |
2214 | * one that is set. (If they are set to "", it means to use the same thing | |
2215 | * we just set LC_ALL to, so can skip) */ | |
7d4bcc4a | 2216 | |
948523db | 2217 | for (i = 0; i < LC_ALL_INDEX; i++) { |
e5f10d49 KW |
2218 | result = PerlEnv_getenv(category_names[i]); |
2219 | if (result && strNE(result, "")) { | |
6c332036 | 2220 | #ifdef USE_WSETLOCALE |
17a1e9ac | 2221 | S_wrap_wsetlocale(aTHX_ categories[i], result); |
6c332036 | 2222 | #else |
17a1e9ac | 2223 | setlocale(categories[i], result); |
6c332036 | 2224 | #endif |
e5f10d49 KW |
2225 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", |
2226 | __FILE__, __LINE__, | |
2227 | setlocale_debug_string(categories[i], result, "not captured"))); | |
2228 | } | |
89f7b9aa | 2229 | } |
7d4bcc4a | 2230 | |
bbc98134 | 2231 | result = setlocale(LC_ALL, NULL); |
48015184 KW |
2232 | DEBUG_L(STMT_START { |
2233 | dSAVE_ERRNO; | |
2234 | PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", | |
bbc98134 | 2235 | __FILE__, __LINE__, |
48015184 KW |
2236 | setlocale_debug_string(LC_ALL, NULL, result)); |
2237 | RESTORE_ERRNO; | |
2238 | } STMT_END); | |
89f7b9aa | 2239 | |
bbc98134 | 2240 | return result; |
b385bb4d KW |
2241 | } |
2242 | ||
2243 | #endif | |
2244 | ||
9aac5db8 KW |
2245 | /* |
2246 | ||
2247 | =head1 Locale-related functions and macros | |
2248 | ||
2249 | =for apidoc Perl_setlocale | |
2250 | ||
2251 | This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>, | |
2252 | taking the same parameters, and returning the same information, except that it | |
9487427b KW |
2253 | returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will |
2254 | instead return C<C> if the underlying locale has a non-dot decimal point | |
2255 | character, or a non-empty thousands separator for displaying floating point | |
2256 | numbers. This is because perl keeps that locale category such that it has a | |
2257 | dot and empty separator, changing the locale briefly during the operations | |
2258 | where the underlying one is required. C<Perl_setlocale> knows about this, and | |
2259 | compensates; regular C<setlocale> doesn't. | |
9aac5db8 | 2260 | |
e9bc6d6b | 2261 | Another reason it isn't completely a drop-in replacement is that it is |
9aac5db8 | 2262 | declared to return S<C<const char *>>, whereas the system setlocale omits the |
163deff3 KW |
2263 | C<const> (presumably because its API was specified long ago, and can't be |
2264 | updated; it is illegal to change the information C<setlocale> returns; doing | |
2265 | so leads to segfaults.) | |
9aac5db8 | 2266 | |
e9bc6d6b KW |
2267 | Finally, C<Perl_setlocale> works under all circumstances, whereas plain |
2268 | C<setlocale> can be completely ineffective on some platforms under some | |
2269 | configurations. | |
2270 | ||
9aac5db8 | 2271 | C<Perl_setlocale> should not be used to change the locale except on systems |
e9bc6d6b KW |
2272 | where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems, |
2273 | the system C<setlocale()> is ineffective, returning the wrong information, and | |
2274 | failing to actually change the locale. C<Perl_setlocale>, however works | |
2275 | properly in all circumstances. | |
9aac5db8 KW |
2276 | |
2277 | The return points to a per-thread static buffer, which is overwritten the next | |
2278 | time C<Perl_setlocale> is called from the same thread. | |
2279 | ||
2280 | =cut | |
2281 | ||
2282 | */ | |
2283 | ||
2284 | const char * | |
2285 | Perl_setlocale(const int category, const char * locale) | |
a4f00dcc KW |
2286 | { |
2287 | /* This wraps POSIX::setlocale() */ | |
2288 | ||
52129632 | 2289 | #ifndef USE_LOCALE |
d36adde0 KW |
2290 | |
2291 | PERL_UNUSED_ARG(category); | |
2292 | PERL_UNUSED_ARG(locale); | |
2293 | ||
2294 | return "C"; | |
2295 | ||
2296 | #else | |
2297 | ||
9aac5db8 KW |
2298 | const char * retval; |
2299 | const char * newlocale; | |
2300 | dSAVEDERRNO; | |
a4f00dcc | 2301 | dTHX; |
d36adde0 | 2302 | DECLARATION_FOR_LC_NUMERIC_MANIPULATION; |
a4f00dcc | 2303 | |
a4f00dcc KW |
2304 | #ifdef USE_LOCALE_NUMERIC |
2305 | ||
291a84fb KW |
2306 | /* A NULL locale means only query what the current one is. We have the |
2307 | * LC_NUMERIC name saved, because we are normally switched into the C | |
9487427b KW |
2308 | * (or equivalent) locale for it. For an LC_ALL query, switch back to get |
2309 | * the correct results. All other categories don't require special | |
2310 | * handling */ | |
a4f00dcc KW |
2311 | if (locale == NULL) { |
2312 | if (category == LC_NUMERIC) { | |
9aac5db8 KW |
2313 | |
2314 | /* We don't have to copy this return value, as it is a per-thread | |
2315 | * variable, and won't change until a future setlocale */ | |
2316 | return PL_numeric_name; | |
a4f00dcc KW |
2317 | } |
2318 | ||
7d4bcc4a | 2319 | # ifdef LC_ALL |
a4f00dcc | 2320 | |
9aac5db8 KW |
2321 | else if (category == LC_ALL) { |
2322 | STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); | |
a4f00dcc KW |
2323 | } |
2324 | ||
7d4bcc4a | 2325 | # endif |
a4f00dcc KW |
2326 | |
2327 | } | |
2328 | ||
2329 | #endif | |
2330 | ||
33e5a354 KW |
2331 | retval = save_to_buffer(do_setlocale_r(category, locale), |
2332 | &PL_setlocale_buf, &PL_setlocale_bufsize, 0); | |
9aac5db8 KW |
2333 | SAVE_ERRNO; |
2334 | ||
2335 | #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL) | |
2336 | ||
2337 | if (locale == NULL && category == LC_ALL) { | |
2338 | RESTORE_LC_NUMERIC(); | |
2339 | } | |
2340 | ||
2341 | #endif | |
a4f00dcc KW |
2342 | |
2343 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
2344 | "%s:%d: %s\n", __FILE__, __LINE__, | |
2345 | setlocale_debug_string(category, locale, retval))); | |
7d4bcc4a | 2346 | |
9aac5db8 KW |
2347 | RESTORE_ERRNO; |
2348 | ||
2349 | if (! retval) { | |
a4f00dcc KW |
2350 | return NULL; |
2351 | } | |
2352 | ||
9aac5db8 | 2353 | /* If locale == NULL, we are just querying the state */ |
a4f00dcc | 2354 | if (locale == NULL) { |
a4f00dcc KW |
2355 | return retval; |
2356 | } | |
a4f00dcc | 2357 | |
1159483a KW |
2358 | /* Now that have switched locales, we have to update our records to |
2359 | * correspond. */ | |
a4f00dcc | 2360 | |
1159483a | 2361 | switch (category) { |
a4f00dcc | 2362 | |
1159483a | 2363 | #ifdef USE_LOCALE_CTYPE |
a4f00dcc | 2364 | |
1159483a KW |
2365 | case LC_CTYPE: |
2366 | new_ctype(retval); | |
2367 | break; | |
a4f00dcc | 2368 | |
1159483a | 2369 | #endif |
a4f00dcc KW |
2370 | #ifdef USE_LOCALE_COLLATE |
2371 | ||
1159483a KW |
2372 | case LC_COLLATE: |
2373 | new_collate(retval); | |
2374 | break; | |
a4f00dcc | 2375 | |
1159483a | 2376 | #endif |
a4f00dcc KW |
2377 | #ifdef USE_LOCALE_NUMERIC |
2378 | ||
1159483a KW |
2379 | case LC_NUMERIC: |
2380 | new_numeric(retval); | |
2381 | break; | |
a4f00dcc | 2382 | |
1159483a KW |
2383 | #endif |
2384 | #ifdef LC_ALL | |
a4f00dcc | 2385 | |
1159483a | 2386 | case LC_ALL: |
a4f00dcc | 2387 | |
1159483a KW |
2388 | /* LC_ALL updates all the things we care about. The values may not |
2389 | * be the same as 'retval', as the locale "" may have set things | |
2390 | * individually */ | |
a4f00dcc | 2391 | |
1159483a | 2392 | # ifdef USE_LOCALE_CTYPE |
a4f00dcc | 2393 | |
b0bde642 | 2394 | newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL)); |
1159483a | 2395 | new_ctype(newlocale); |
b0bde642 | 2396 | Safefree(newlocale); |
a4f00dcc | 2397 | |
1159483a KW |
2398 | # endif /* USE_LOCALE_CTYPE */ |
2399 | # ifdef USE_LOCALE_COLLATE | |
2400 | ||
b0bde642 | 2401 | newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL)); |
1159483a | 2402 | new_collate(newlocale); |
b0bde642 | 2403 | Safefree(newlocale); |
a4f00dcc | 2404 | |
7d4bcc4a | 2405 | # endif |
1159483a | 2406 | # ifdef USE_LOCALE_NUMERIC |
a4f00dcc | 2407 | |
b0bde642 | 2408 | newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL)); |
1159483a | 2409 | new_numeric(newlocale); |
b0bde642 | 2410 | Safefree(newlocale); |
a4f00dcc | 2411 | |
1159483a KW |
2412 | # endif /* USE_LOCALE_NUMERIC */ |
2413 | #endif /* LC_ALL */ | |
a4f00dcc | 2414 | |
1159483a KW |
2415 | default: |
2416 | break; | |
a4f00dcc KW |
2417 | } |
2418 | ||
2419 | return retval; | |
2420 | ||
d36adde0 KW |
2421 | #endif |
2422 | ||
f7416781 KW |
2423 | } |
2424 | ||
2425 | PERL_STATIC_INLINE const char * | |
2426 | S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset) | |
2427 | { | |
2428 | /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size', | |
2429 | * growing it if necessary */ | |
2430 | ||
0b6697c5 | 2431 | Size_t string_size; |
f7416781 KW |
2432 | |
2433 | PERL_ARGS_ASSERT_SAVE_TO_BUFFER; | |
2434 | ||
0b6697c5 KW |
2435 | if (! string) { |
2436 | return NULL; | |
2437 | } | |
2438 | ||
2439 | string_size = strlen(string) + offset + 1; | |
2440 | ||
f7416781 KW |
2441 | if (*buf_size == 0) { |
2442 | Newx(*buf, string_size, char); | |
2443 | *buf_size = string_size; | |
2444 | } | |
2445 | else if (string_size > *buf_size) { | |
2446 | Renew(*buf, string_size, char); | |
2447 | *buf_size = string_size; | |
2448 | } | |
2449 | ||
2450 | Copy(string, *buf + offset, string_size - offset, char); | |
2451 | return *buf; | |
2452 | } | |
2453 | ||
2454 | /* | |
2455 | ||
f7416781 KW |
2456 | =for apidoc Perl_langinfo |
2457 | ||
ce181bf6 | 2458 | This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>, |
f7416781 KW |
2459 | taking the same C<item> parameter values, and returning the same information. |
2460 | But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks | |
2461 | of Perl's locale handling from your code, and can be used on systems that lack | |
2462 | a native C<nl_langinfo>. | |
2463 | ||
2464 | Expanding on these: | |
2465 | ||
2466 | =over | |
2467 | ||
2468 | =item * | |
2469 | ||
ce181bf6 KW |
2470 | The reason it isn't quite a drop-in replacement is actually an advantage. The |
2471 | only difference is that it returns S<C<const char *>>, whereas plain | |
2472 | C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation) | |
2473 | forbidden to write into the buffer. By declaring this C<const>, the compiler | |
2474 | enforces this restriction, so if it is violated, you know at compilation time, | |
2475 | rather than getting segfaults at runtime. | |
2476 | ||
2477 | =item * | |
2478 | ||
69e2275e | 2479 | It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items, |
f7416781 KW |
2480 | without you having to write extra code. The reason for the extra code would be |
2481 | because these are from the C<LC_NUMERIC> locale category, which is normally | |
9487427b KW |
2482 | kept set by Perl so that the radix is a dot, and the separator is the empty |
2483 | string, no matter what the underlying locale is supposed to be, and so to get | |
2484 | the expected results, you have to temporarily toggle into the underlying | |
2485 | locale, and later toggle back. (You could use plain C<nl_langinfo> and | |
2486 | C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get | |
2487 | the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C | |
2488 | (or equivalent) locale would break a lot of CPAN, which is expecting the radix | |
2489 | (decimal point) character to be a dot.) | |
ce181bf6 KW |
2490 | |
2491 | =item * | |
2492 | ||
2493 | The system function it replaces can have its static return buffer trashed, | |
2494 | not only by a subesequent call to that function, but by a C<freelocale>, | |
2495 | C<setlocale>, or other locale change. The returned buffer of this function is | |
2496 | not changed until the next call to it, so the buffer is never in a trashed | |
2497 | state. | |
f7416781 KW |
2498 | |
2499 | =item * | |
2500 | ||
ce181bf6 KW |
2501 | Its return buffer is per-thread, so it also is never overwritten by a call to |
2502 | this function from another thread; unlike the function it replaces. | |
2503 | ||
2504 | =item * | |
2505 | ||
2506 | But most importantly, it works on systems that don't have C<nl_langinfo>, such | |
2507 | as Windows, hence makes your code more portable. Of the fifty-some possible | |
2508 | items specified by the POSIX 2008 standard, | |
f7416781 | 2509 | L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>, |
8d72e74e KW |
2510 | only one is completely unimplemented, though on non-Windows platforms, another |
2511 | significant one is also not implemented). It uses various techniques to | |
2512 | recover the other items, including calling C<L<localeconv(3)>>, and | |
2513 | C<L<strftime(3)>>, both of which are specified in C89, so should be always be | |
2514 | available. Later C<strftime()> versions have additional capabilities; C<""> is | |
2515 | returned for those not available on your system. | |
ce181bf6 KW |
2516 | |
2517 | It is important to note that when called with an item that is recovered by | |
2518 | using C<localeconv>, the buffer from any previous explicit call to | |
2519 | C<localeconv> will be overwritten. This means you must save that buffer's | |
c3997e39 KW |
2520 | contents if you need to access them after a call to this function. (But note |
2521 | that you might not want to be using C<localeconv()> directly anyway, because of | |
2522 | issues like the ones listed in the second item of this list (above) for | |
2523 | C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to | |
2524 | call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to | |
2525 | unpack). | |
9f0bc911 | 2526 | |
6201c220 KW |
2527 | The details for those items which may deviate from what this emulation returns |
2528 | and what a native C<nl_langinfo()> would return are specified in | |
2529 | L<I18N::Langinfo>. | |
f7416781 | 2530 | |
ce181bf6 KW |
2531 | =back |
2532 | ||
f7416781 KW |
2533 | When using C<Perl_langinfo> on systems that don't have a native |
2534 | C<nl_langinfo()>, you must | |
2535 | ||
2536 | #include "perl_langinfo.h" | |
2537 | ||
2538 | before the C<perl.h> C<#include>. You can replace your C<langinfo.h> | |
2539 | C<#include> with this one. (Doing it this way keeps out the symbols that plain | |
c3997e39 KW |
2540 | C<langinfo.h> would try to import into the namespace for code that doesn't need |
2541 | it.) | |
f7416781 | 2542 | |
f7416781 KW |
2543 | The original impetus for C<Perl_langinfo()> was so that code that needs to |
2544 | find out the current currency symbol, floating point radix character, or digit | |
2545 | grouping separator can use, on all systems, the simpler and more | |
2546 | thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a | |
2547 | pain to make thread-friendly. For other fields returned by C<localeconv>, it | |
2548 | is better to use the methods given in L<perlcall> to call | |
2549 | L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly. | |
2550 | ||
2551 | =cut | |
2552 | ||
2553 | */ | |
2554 | ||
2555 | const char * | |
2556 | #ifdef HAS_NL_LANGINFO | |
2557 | Perl_langinfo(const nl_item item) | |
2558 | #else | |
2559 | Perl_langinfo(const int item) | |
2560 | #endif | |
2561 | { | |
f61748ac KW |
2562 | return my_nl_langinfo(item, TRUE); |
2563 | } | |
2564 | ||
59a15af0 | 2565 | STATIC const char * |
f61748ac KW |
2566 | #ifdef HAS_NL_LANGINFO |
2567 | S_my_nl_langinfo(const nl_item item, bool toggle) | |
2568 | #else | |
2569 | S_my_nl_langinfo(const int item, bool toggle) | |
2570 | #endif | |
2571 | { | |
ae74815b | 2572 | dTHX; |
0b6697c5 | 2573 | const char * retval; |
f7416781 | 2574 | |
d36adde0 KW |
2575 | #ifdef USE_LOCALE_NUMERIC |
2576 | ||
5a854ab3 KW |
2577 | /* We only need to toggle into the underlying LC_NUMERIC locale for these |
2578 | * two items, and only if not already there */ | |
4e6826bf | 2579 | if (toggle && (( item != RADIXCHAR && item != THOUSEP) |
5a854ab3 | 2580 | || PL_numeric_underlying)) |
d36adde0 KW |
2581 | |
2582 | #endif /* No toggling needed if not using LC_NUMERIC */ | |
2583 | ||
5a854ab3 | 2584 | toggle = FALSE; |
5a854ab3 | 2585 | |
ab340fff | 2586 | #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */ |
ee90eb2d | 2587 | # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \ |
80306bb6 KW |
2588 | || ! defined(HAS_POSIX_2008_LOCALE) \ |
2589 | || ! defined(DUPLOCALE) | |
f7416781 | 2590 | |
ab340fff | 2591 | /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC |
ae74815b KW |
2592 | * for those items dependent on it. This must be copied to a buffer before |
2593 | * switching back, as some systems destroy the buffer when setlocale() is | |
2594 | * called */ | |
f7416781 | 2595 | |
038d3702 KW |
2596 | { |
2597 | DECLARATION_FOR_LC_NUMERIC_MANIPULATION; | |
2598 | ||
b2fee59c | 2599 | if (toggle) { |
038d3702 | 2600 | STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); |
b2fee59c | 2601 | } |
f7416781 | 2602 | |
5acc4454 KW |
2603 | LOCALE_LOCK; /* Prevent interference from another thread executing |
2604 | this code section (the only call to nl_langinfo in | |
2605 | the core) */ | |
2606 | ||
ee90eb2d | 2607 | |
628ff75a KW |
2608 | /* Copy to a per-thread buffer, which is also one that won't be |
2609 | * destroyed by a subsequent setlocale(), such as the | |
2610 | * RESTORE_LC_NUMERIC may do just below. */ | |
0b6697c5 KW |
2611 | retval = save_to_buffer(nl_langinfo(item), |
2612 | &PL_langinfo_buf, &PL_langinfo_bufsize, 0); | |
f7416781 | 2613 | |
5acc4454 KW |
2614 | LOCALE_UNLOCK; |
2615 | ||
b2fee59c | 2616 | if (toggle) { |
038d3702 | 2617 | RESTORE_LC_NUMERIC(); |
b2fee59c | 2618 | } |
f7416781 KW |
2619 | } |
2620 | ||
ab340fff KW |
2621 | # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */ |
2622 | ||
5a854ab3 | 2623 | { |
b2fee59c KW |
2624 | bool do_free = FALSE; |
2625 | locale_t cur = uselocale((locale_t) 0); | |
ab340fff | 2626 | |
b2fee59c KW |
2627 | if (cur == LC_GLOBAL_LOCALE) { |
2628 | cur = duplocale(LC_GLOBAL_LOCALE); | |
2629 | do_free = TRUE; | |
2630 | } | |
ab340fff | 2631 | |
d36adde0 KW |
2632 | # ifdef USE_LOCALE_NUMERIC |
2633 | ||
b2fee59c | 2634 | if (toggle) { |
e1aa2579 KW |
2635 | if (PL_underlying_numeric_obj) { |
2636 | cur = PL_underlying_numeric_obj; | |
2637 | } | |
2638 | else { | |
2639 | cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur); | |
2640 | do_free = TRUE; | |
2641 | } | |
b2fee59c | 2642 | } |
ab340fff | 2643 | |
d36adde0 KW |
2644 | # endif |
2645 | ||
628ff75a KW |
2646 | /* We have to save it to a buffer, because the freelocale() just below |
2647 | * can invalidate the internal one */ | |
0b6697c5 KW |
2648 | retval = save_to_buffer(nl_langinfo_l(item, cur), |
2649 | &PL_langinfo_buf, &PL_langinfo_bufsize, 0); | |
ee90eb2d | 2650 | |
b2fee59c KW |
2651 | if (do_free) { |
2652 | freelocale(cur); | |
2653 | } | |
5a854ab3 | 2654 | } |
ab340fff | 2655 | |
c1566110 KW |
2656 | # endif |
2657 | ||
0b6697c5 | 2658 | if (strEQ(retval, "")) { |
4e6826bf | 2659 | if (item == YESSTR) { |
c1566110 KW |
2660 | return "yes"; |
2661 | } | |
4e6826bf | 2662 | if (item == NOSTR) { |
c1566110 KW |
2663 | return "no"; |
2664 | } | |
2665 | } | |
2666 | ||
0b6697c5 | 2667 | return retval; |
ab340fff | 2668 | |
f7416781 | 2669 | #else /* Below, emulate nl_langinfo as best we can */ |
43dd6b15 KW |
2670 | |
2671 | { | |
2672 | ||
f7416781 KW |
2673 | # ifdef HAS_LOCALECONV |
2674 | ||
43dd6b15 | 2675 | const struct lconv* lc; |
628ff75a | 2676 | const char * temp; |
038d3702 | 2677 | DECLARATION_FOR_LC_NUMERIC_MANIPULATION; |
f7416781 | 2678 | |
69c5e0db KW |
2679 | # ifdef TS_W32_BROKEN_LOCALECONV |
2680 | ||
2681 | const char * save_global; | |
2682 | const char * save_thread; | |
2683 | int needed_size; | |
2684 | char * ptr; | |
2685 | char * e; | |
2686 | char * item_start; | |
2687 | ||
2688 | # endif | |
f7416781 KW |
2689 | # endif |
2690 | # ifdef HAS_STRFTIME | |
2691 | ||
43dd6b15 KW |
2692 | struct tm tm; |
2693 | bool return_format = FALSE; /* Return the %format, not the value */ | |
2694 | const char * format; | |
f7416781 KW |
2695 | |
2696 | # endif | |
2697 | ||
43dd6b15 KW |
2698 | /* We copy the results to a per-thread buffer, even if not |
2699 | * multi-threaded. This is in part to simplify this code, and partly | |
2700 | * because we need a buffer anyway for strftime(), and partly because a | |
2701 | * call of localeconv() could otherwise wipe out the buffer, and the | |
2702 | * programmer would not be expecting this, as this is a nl_langinfo() | |
2703 | * substitute after all, so s/he might be thinking their localeconv() | |
2704 | * is safe until another localeconv() call. */ | |
f7416781 | 2705 | |
43dd6b15 KW |
2706 | switch (item) { |
2707 | Size_t len; | |
f7416781 | 2708 | |
8d72e74e | 2709 | /* This is unimplemented */ |
4e6826bf | 2710 | case ERA: /* For use with strftime() %E modifier */ |
f7416781 | 2711 | |
43dd6b15 KW |
2712 | default: |
2713 | return ""; | |
f7416781 | 2714 | |
43dd6b15 | 2715 | /* We use only an English set, since we don't know any more */ |
4e6826bf KW |
2716 | case YESEXPR: return "^[+1yY]"; |
2717 | case YESSTR: return "yes"; | |
2718 | case NOEXPR: return "^[-0nN]"; | |
2719 | case NOSTR: return "no"; | |
2720 | ||
8d72e74e KW |
2721 | case CODESET: |
2722 | ||
2723 | # ifndef WIN32 | |
2724 | ||
2725 | /* On non-windows, this is unimplemented, in part because of | |
2726 | * inconsistencies between vendors. The Darwin native | |
2727 | * nl_langinfo() implementation simply looks at everything past | |
2728 | * any dot in the name, but that doesn't work for other | |
2729 | * vendors. Many Linux locales that don't have UTF-8 in their | |
2730 | * names really are UTF-8, for example; z/OS locales that do | |
2731 | * have UTF-8 in their names, aren't really UTF-8 */ | |
2732 | return ""; | |
2733 | ||
2734 | # else | |
2735 | ||
2736 | { /* But on Windows, the name does seem to be consistent, so | |
2737 | use that. */ | |
2738 | const char * p; | |
2739 | const char * first; | |
2740 | Size_t offset = 0; | |
2741 | const char * name = my_setlocale(LC_CTYPE, NULL); | |
2742 | ||
2743 | if (isNAME_C_OR_POSIX(name)) { | |
2744 | return "ANSI_X3.4-1968"; | |
2745 | } | |
2746 | ||
2747 | /* Find the dot in the locale name */ | |
2748 | first = (const char *) strchr(name, '.'); | |
2749 | if (! first) { | |
2750 | first = name; | |
2751 | goto has_nondigit; | |
2752 | } | |
2753 | ||
2754 | /* Look at everything past the dot */ | |
2755 | first++; | |
2756 | p = first; | |
2757 | ||
2758 | while (*p) { | |
2759 | if (! isDIGIT(*p)) { | |
2760 | goto has_nondigit; | |
2761 | } | |
2762 | ||
2763 | p++; | |
2764 | } | |
2765 | ||
2766 | /* Here everything past the dot is a digit. Treat it as a | |
2767 | * code page */ | |
18503cbd | 2768 | retval = save_to_buffer("CP", &PL_langinfo_buf, |
32a62865 | 2769 | &PL_langinfo_bufsize, 0); |
8d72e74e | 2770 | offset = STRLENs("CP"); |
f7416781 | 2771 | |
8d72e74e KW |
2772 | has_nondigit: |
2773 | ||
2774 | retval = save_to_buffer(first, &PL_langinfo_buf, | |
2775 | &PL_langinfo_bufsize, offset); | |
2776 | } | |
2777 | ||
2778 | break; | |
2779 | ||
2780 | # endif | |
f7416781 KW |
2781 | # ifdef HAS_LOCALECONV |
2782 | ||
4e6826bf | 2783 | case CRNCYSTR: |
f7416781 | 2784 | |
43dd6b15 KW |
2785 | /* We don't bother with localeconv_l() because any system that |
2786 | * has it is likely to also have nl_langinfo() */ | |
291a84fb | 2787 | |
69c5e0db KW |
2788 | LOCALE_LOCK_V; /* Prevent interference with other threads |
2789 | using localeconv() */ | |
2790 | ||
2791 | # ifdef TS_W32_BROKEN_LOCALECONV | |
2792 | ||
2793 | /* This is a workaround for a Windows bug prior to VS 15. | |
2794 | * What we do here is, while locked, switch to the global | |
2795 | * locale so localeconv() works; then switch back just before | |
2796 | * the unlock. This can screw things up if some thread is | |
2797 | * already using the global locale while assuming no other is. | |
2798 | * A different workaround would be to call GetCurrencyFormat on | |
2799 | * a known value, and parse it; patches welcome | |
2800 | * | |
2801 | * We have to use LC_ALL instead of LC_MONETARY because of | |
2802 | * another bug in Windows */ | |
2803 | ||
2804 | save_thread = savepv(my_setlocale(LC_ALL, NULL)); | |
2805 | _configthreadlocale(_DISABLE_PER_THREAD_LOCALE); | |
2806 | save_global= savepv(my_setlocale(LC_ALL, NULL)); | |
2807 | my_setlocale(LC_ALL, save_thread); | |
2808 | ||
2809 | # endif | |
5acc4454 | 2810 | |
43dd6b15 KW |
2811 | lc = localeconv(); |
2812 | if ( ! lc | |
2813 | || ! lc->currency_symbol | |
2814 | || strEQ("", lc->currency_symbol)) | |
2815 | { | |
69c5e0db | 2816 | LOCALE_UNLOCK_V; |
43dd6b15 KW |
2817 | return ""; |
2818 | } | |
f7416781 | 2819 | |
43dd6b15 | 2820 | /* Leave the first spot empty to be filled in below */ |
0b6697c5 KW |
2821 | retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf, |
2822 | &PL_langinfo_bufsize, 1); | |
43dd6b15 KW |
2823 | if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, "")) |
2824 | { /* khw couldn't figure out how the localedef specifications | |
2825 | would show that the $ should replace the radix; this is | |
2826 | just a guess as to how it might work.*/ | |
a34edee3 | 2827 | PL_langinfo_buf[0] = '.'; |
43dd6b15 KW |
2828 | } |
2829 | else if (lc->p_cs_precedes) { | |
a34edee3 | 2830 | PL_langinfo_buf[0] = '-'; |
43dd6b15 KW |
2831 | } |
2832 | else { | |
a34edee3 | 2833 | PL_langinfo_buf[0] = '+'; |
43dd6b15 | 2834 | } |
f7416781 | 2835 | |
69c5e0db KW |
2836 | # ifdef TS_W32_BROKEN_LOCALECONV |
2837 | ||
2838 | my_setlocale(LC_ALL, save_global); | |
2839 | _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); | |
2840 | my_setlocale(LC_ALL, save_thread); | |
2841 | Safefree(save_global); | |
2842 | Safefree(save_thread); | |
2843 | ||
2844 | # endif | |
2845 | ||
2846 | LOCALE_UNLOCK_V; | |
43dd6b15 | 2847 | break; |
f7416781 | 2848 | |
69c5e0db KW |
2849 | # ifdef TS_W32_BROKEN_LOCALECONV |
2850 | ||
4e6826bf | 2851 | case RADIXCHAR: |
69c5e0db KW |
2852 | |
2853 | /* For this, we output a known simple floating point number to | |
2854 | * a buffer, and parse it, looking for the radix */ | |
2855 | ||
2856 | if (toggle) { | |
2857 | STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); | |
2858 | } | |
2859 | ||
2860 | if (PL_langinfo_bufsize < 10) { | |
2861 | PL_langinfo_bufsize = 10; | |
2862 | Renew(PL_langinfo_buf, PL_langinfo_bufsize, char); | |
2863 | } | |
2864 | ||
2865 | needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize, | |
2866 | "%.1f", 1.5); | |
2867 | if (needed_size >= (int) PL_langinfo_bufsize) { | |
2868 | PL_langinfo_bufsize = needed_size + 1; | |
2869 | Renew(PL_langinfo_buf, PL_langinfo_bufsize, char); | |
2870 | needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize, | |
2871 | "%.1f", 1.5); | |
2872 | assert(needed_size < (int) PL_langinfo_bufsize); | |
2873 | } | |
2874 | ||
2875 | ptr = PL_langinfo_buf; | |
2876 | e = PL_langinfo_buf + PL_langinfo_bufsize; | |
2877 | ||
2878 | /* Find the '1' */ | |
2879 | while (ptr < e && *ptr != '1') { | |
2880 | ptr++; | |
2881 | } | |
2882 | ptr++; | |
2883 | ||
2884 | /* Find the '5' */ | |
2885 | item_start = ptr; | |
2886 | while (ptr < e && *ptr != '5') { | |
2887 | ptr++; | |
2888 | } | |
2889 | ||
2890 | /* Everything in between is the radix string */ | |
2891 | if (ptr >= e) { | |
2892 | PL_langinfo_buf[0] = '?'; | |
2893 | PL_langinfo_buf[1] = '\0'; | |
2894 | } | |
2895 | else { | |
2896 | *ptr = '\0'; | |
2897 | Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char); | |
2898 | } | |
2899 | ||
2900 | if (toggle) { | |
2901 | RESTORE_LC_NUMERIC(); | |
2902 | } | |
2903 | ||
2904 | retval = PL_langinfo_buf; | |
2905 | break; | |
2906 | ||
2907 | # else | |
2908 | ||
2909 | case RADIXCHAR: /* No special handling needed */ | |
2910 | ||
2911 | # endif | |
2912 | ||
4e6826bf | 2913 | case THOUSEP: |
f7416781 | 2914 | |
43dd6b15 | 2915 | if (toggle) { |
038d3702 | 2916 | STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); |
c0d737a8 | 2917 | } |
f7416781 | 2918 | |
69c5e0db KW |
2919 | LOCALE_LOCK_V; /* Prevent interference with other threads |
2920 | using localeconv() */ | |
2921 | ||
2922 | # ifdef TS_W32_BROKEN_LOCALECONV | |
2923 | ||
2924 | /* This should only be for the thousands separator. A | |
2925 | * different work around would be to use GetNumberFormat on a | |
2926 | * known value and parse the result to find the separator */ | |
2927 | save_thread = savepv(my_setlocale(LC_ALL, NULL)); | |
2928 | _configthreadlocale(_DISABLE_PER_THREAD_LOCALE); | |
2929 | save_global = savepv(my_setlocale(LC_ALL, NULL)); | |
2930 | my_setlocale(LC_ALL, save_thread); | |
2931 | # if 0 | |
2932 | /* This is the start of code that for broken Windows replaces | |
2933 | * the above and below code, and instead calls | |
2934 | * GetNumberFormat() and then would parse that to find the | |
2935 | * thousands separator. It needs to handle UTF-16 vs -8 | |
2936 | * issues. */ | |
2937 | ||
2938 | needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize); | |
2939 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
2940 | "%s: %d: return from GetNumber, count=%d, val=%s\n", | |
2941 | __FILE__, __LINE__, needed_size, PL_langinfo_buf)); | |
2942 | ||
2943 | # endif | |
2944 | # endif | |
5acc4454 | 2945 | |
43dd6b15 KW |
2946 | lc = localeconv(); |
2947 | if (! lc) { | |
628ff75a | 2948 | temp = ""; |
33394adc | 2949 | } |
43dd6b15 | 2950 | else { |
4e6826bf | 2951 | temp = (item == RADIXCHAR) |
43dd6b15 KW |
2952 | ? lc->decimal_point |
2953 | : lc->thousands_sep; | |
628ff75a KW |
2954 | if (! temp) { |
2955 | temp = ""; | |
43dd6b15 KW |
2956 | } |
2957 | } | |
f7416781 | 2958 | |
0b6697c5 KW |
2959 | retval = save_to_buffer(temp, &PL_langinfo_buf, |
2960 | &PL_langinfo_bufsize, 0); | |
f7416781 | 2961 | |
69c5e0db KW |
2962 | # ifdef TS_W32_BROKEN_LOCALECONV |
2963 | ||
2964 | my_setlocale(LC_ALL, save_global); | |
2965 | _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); | |
2966 | my_setlocale(LC_ALL, save_thread); | |
2967 | Safefree(save_global); | |
2968 | Safefree(save_thread); | |
2969 | ||
2970 | # endif | |
2971 | ||
2972 | LOCALE_UNLOCK_V; | |
5acc4454 | 2973 | |
43dd6b15 | 2974 | if (toggle) { |
038d3702 | 2975 | RESTORE_LC_NUMERIC(); |
43dd6b15 | 2976 | } |
f7416781 | 2977 | |
43dd6b15 | 2978 | break; |
f7416781 KW |
2979 | |
2980 | # endif | |
2981 | # ifdef HAS_STRFTIME | |
2982 | ||
43dd6b15 KW |
2983 | /* These are defined by C89, so we assume that strftime supports |
2984 | * them, and so are returned unconditionally; they may not be what | |
2985 | * the locale actually says, but should give good enough results | |
2986 | * for someone using them as formats (as opposed to trying to parse | |
2987 | * them to figure out what the locale says). The other format | |
2988 | * items are actually tested to verify they work on the platform */ | |
4e6826bf KW |
2989 | case D_FMT: return "%x"; |
2990 | case T_FMT: return "%X"; | |
2991 | case D_T_FMT: return "%c"; | |
43dd6b15 KW |
2992 | |
2993 | /* These formats are only available in later strfmtime's */ | |
4e6826bf | 2994 | case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM: |
43dd6b15 KW |
2995 | |
2996 | /* The rest can be gotten from most versions of strftime(). */ | |
4e6826bf KW |
2997 | case ABDAY_1: case ABDAY_2: case ABDAY_3: |
2998 | case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7: | |
2999 | case ALT_DIGITS: | |
3000 | case AM_STR: case PM_STR: | |
3001 | case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4: | |
3002 | case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8: | |
3003 | case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12: | |
3004 | case DAY_1: case DAY_2: case DAY_3: case DAY_4: | |
3005 | case DAY_5: case DAY_6: case DAY_7: | |
3006 | case MON_1: case MON_2: case MON_3: case MON_4: | |
3007 | case MON_5: case MON_6: case MON_7: case MON_8: | |
3008 | case MON_9: case MON_10: case MON_11: case MON_12: | |
43dd6b15 KW |
3009 | |
3010 | LOCALE_LOCK; | |
3011 | ||
3012 | init_tm(&tm); /* Precaution against core dumps */ | |
3013 | tm.tm_sec = 30; | |
3014 | tm.tm_min = 30; | |
3015 | tm.tm_hour = 6; | |
3016 | tm.tm_year = 2017 - 1900; | |
3017 | tm.tm_wday = 0; | |
3018 | tm.tm_mon = 0; | |
3019 | switch (item) { | |
3020 | default: | |
3021 | LOCALE_UNLOCK; | |
3022 | Perl_croak(aTHX_ | |
3023 | "panic: %s: %d: switch case: %d problem", | |
3024 | __FILE__, __LINE__, item); | |
3025 | NOT_REACHED; /* NOTREACHED */ | |
3026 | ||
4e6826bf KW |
3027 | case PM_STR: tm.tm_hour = 18; |
3028 | case AM_STR: | |
43dd6b15 KW |
3029 | format = "%p"; |
3030 | break; | |
3031 | ||
4e6826bf KW |
3032 | case ABDAY_7: tm.tm_wday++; |
3033 | case ABDAY_6: tm.tm_wday++; | |
3034 | case ABDAY_5: tm.tm_wday++; | |
3035 | case ABDAY_4: tm.tm_wday++; | |
3036 | case ABDAY_3: tm.tm_wday++; | |
3037 | case ABDAY_2: tm.tm_wday++; | |
3038 | case ABDAY_1: | |
43dd6b15 KW |
3039 | format = "%a"; |
3040 | break; | |
3041 | ||
4e6826bf KW |
3042 | case DAY_7: tm.tm_wday++; |
3043 | case DAY_6: tm.tm_wday++; | |
3044 | case DAY_5: tm.tm_wday++; | |
3045 | case DAY_4: tm.tm_wday++; | |
3046 | case DAY_3: tm.tm_wday++; | |
3047 | case DAY_2: tm.tm_wday++; | |
3048 | case DAY_1: | |
43dd6b15 KW |
3049 | format = "%A"; |
3050 | break; | |
3051 | ||
4e6826bf KW |
3052 | case ABMON_12: tm.tm_mon++; |
3053 | case ABMON_11: tm.tm_mon++; | |
3054 | case ABMON_10: tm.tm_mon++; | |
3055 | case ABMON_9: tm.tm_mon++; | |
3056 | case ABMON_8: tm.tm_mon++; | |
3057 | case ABMON_7: tm.tm_mon++; | |
3058 | case ABMON_6: tm.tm_mon++; | |
3059 | case ABMON_5: tm.tm_mon++; | |
3060 | case ABMON_4: tm.tm_mon++; | |
3061 | case ABMON_3: tm.tm_mon++; | |
3062 | case ABMON_2: tm.tm_mon++; | |
3063 | case ABMON_1: | |
43dd6b15 KW |
3064 | format = "%b"; |
3065 | break; | |
3066 | ||
4e6826bf KW |
3067 | case MON_12: tm.tm_mon++; |
3068 | case MON_11: tm.tm_mon++; | |
3069 | case MON_10: tm.tm_mon++; | |
3070 | case MON_9: tm.tm_mon++; | |
3071 | case MON_8: tm.tm_mon++; | |
3072 | case MON_7: tm.tm_mon++; | |
3073 | case MON_6: tm.tm_mon++; | |
3074 | case MON_5: tm.tm_mon++; | |
3075 | case MON_4: tm.tm_mon++; | |
3076 | case MON_3: tm.tm_mon++; | |
3077 | case MON_2: tm.tm_mon++; | |
3078 | case MON_1: | |
43dd6b15 KW |
3079 | format = "%B"; |
3080 | break; | |
3081 | ||
4e6826bf | 3082 | case T_FMT_AMPM: |
43dd6b15 KW |
3083 | format = "%r"; |
3084 | return_format = TRUE; | |
3085 | break; | |
3086 | ||
4e6826bf | 3087 | case ERA_D_FMT: |
43dd6b15 KW |
3088 | format = "%Ex"; |
3089 | return_format = TRUE; | |
3090 | break; | |
3091 | ||
4e6826bf | 3092 | case ERA_T_FMT: |
43dd6b15 KW |
3093 | format = "%EX"; |
3094 | return_format = TRUE; | |
3095 | break; | |
3096 | ||
4e6826bf | 3097 | case ERA_D_T_FMT: |
43dd6b15 KW |
3098 | format = "%Ec"; |
3099 | return_format = TRUE; | |
3100 | break; | |
3101 | ||
4e6826bf | 3102 | case ALT_DIGITS: |
43dd6b15 KW |
3103 | tm.tm_wday = 0; |
3104 | format = "%Ow"; /* Find the alternate digit for 0 */ | |
3105 | break; | |
3106 | } | |
f7416781 | 3107 | |
43dd6b15 KW |
3108 | /* We can't use my_strftime() because it doesn't look at |
3109 | * tm_wday */ | |
3110 | while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize, | |
3111 | format, &tm)) | |
3112 | { | |
3113 | /* A zero return means one of: | |
3114 | * a) there wasn't enough space in PL_langinfo_buf | |
3115 | * b) the format, like a plain %p, returns empty | |
3116 | * c) it was an illegal format, though some | |
3117 | * implementations of strftime will just return the | |
3118 | * illegal format as a plain character sequence. | |
3119 | * | |
3120 | * To quickly test for case 'b)', try again but precede | |
3121 | * the format with a plain character. If that result is | |
3122 | * still empty, the problem is either 'a)' or 'c)' */ | |
3123 | ||
3124 | Size_t format_size = strlen(format) + 1; | |
3125 | Size_t mod_size = format_size + 1; | |
3126 | char * mod_format; | |
3127 | char * temp_result; | |
3128 | ||
3129 | Newx(mod_format, mod_size, char); | |
3130 | Newx(temp_result, PL_langinfo_bufsize, char); | |
6873aa47 | 3131 | *mod_format = ' '; |
43dd6b15 KW |
3132 | my_strlcpy(mod_format + 1, format, mod_size); |
3133 | len = strftime(temp_result, | |
3134 | PL_langinfo_bufsize, | |
3135 | mod_format, &tm); | |
3136 | Safefree(mod_format); | |
3137 | Safefree(temp_result); | |
3138 | ||
3139 | /* If 'len' is non-zero, it means that we had a case like | |
3140 | * %p which means the current locale doesn't use a.m. or | |
3141 | * p.m., and that is valid */ | |
3142 | if (len == 0) { | |
3143 | ||
3144 | /* Here, still didn't work. If we get well beyond a | |
3145 | * reasonable size, bail out to prevent an infinite | |
3146 | * loop. */ | |
3147 | ||
3148 | if (PL_langinfo_bufsize > 100 * format_size) { | |
3149 | *PL_langinfo_buf = '\0'; | |
3150 | } | |
3151 | else { | |
3152 | /* Double the buffer size to retry; Add 1 in case | |
3153 | * original was 0, so we aren't stuck at 0. */ | |
3154 | PL_langinfo_bufsize *= 2; | |
3155 | PL_langinfo_bufsize++; | |
3156 | Renew(PL_langinfo_buf, PL_langinfo_bufsize, char); | |
3157 | continue; | |
3158 | } | |
3159 | } | |
f7416781 | 3160 | |
f7416781 | 3161 | break; |
43dd6b15 | 3162 | } |
f7416781 | 3163 | |
43dd6b15 KW |
3164 | /* Here, we got a result. |
3165 | * | |
3166 | * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the | |
3167 | * alternate format for wday 0. If the value is the same as | |
3168 | * the normal 0, there isn't an alternate, so clear the buffer. | |
3169 | * */ | |
4e6826bf | 3170 | if ( item == ALT_DIGITS |
43dd6b15 KW |
3171 | && strEQ(PL_langinfo_buf, "0")) |
3172 | { | |
3173 | *PL_langinfo_buf = '\0'; | |
3174 | } | |
f7416781 | 3175 | |
43dd6b15 KW |
3176 | /* ALT_DIGITS is problematic. Experiments on it showed that |
3177 | * strftime() did not always work properly when going from | |
3178 | * alt-9 to alt-10. Only a few locales have this item defined, | |
3179 | * and in all of them on Linux that khw was able to find, | |
3180 | * nl_langinfo() merely returned the alt-0 character, possibly | |
3181 | * doubled. Most Unicode digits are in blocks of 10 | |
3182 | * consecutive code points, so that is sufficient information | |
3183 | * for those scripts, as we can infer alt-1, alt-2, .... But | |
3184 | * for a Japanese locale, a CJK ideographic 0 is returned, and | |
3185 | * the CJK digits are not in code point order, so you can't | |
3186 | * really infer anything. The localedef for this locale did | |
3187 | * specify the succeeding digits, so that strftime() works | |
3188 | * properly on them, without needing to infer anything. But | |
3189 | * the nl_langinfo() return did not give sufficient information | |
3190 | * for the caller to understand what's going on. So until | |
3191 | * there is evidence that it should work differently, this | |
3192 | * returns the alt-0 string for ALT_DIGITS. | |
3193 | * | |
3194 | * wday was chosen because its range is all a single digit. | |
3195 | * Things like tm_sec have two digits as the minimum: '00' */ | |
f7416781 | 3196 | |
43dd6b15 | 3197 | LOCALE_UNLOCK; |
f7416781 | 3198 | |
d1b150d9 KW |
3199 | retval = PL_langinfo_buf; |
3200 | ||
43dd6b15 KW |
3201 | /* If to return the format, not the value, overwrite the buffer |
3202 | * with it. But some strftime()s will keep the original format | |
3203 | * if illegal, so change those to "" */ | |
3204 | if (return_format) { | |
3205 | if (strEQ(PL_langinfo_buf, format)) { | |
f7416781 KW |
3206 | *PL_langinfo_buf = '\0'; |
3207 | } | |
43dd6b15 | 3208 | else { |
0b6697c5 KW |
3209 | retval = save_to_buffer(format, &PL_langinfo_buf, |
3210 | &PL_langinfo_bufsize, 0); | |
f7416781 KW |
3211 | } |
3212 | } | |
3213 | ||
3214 | break; | |
f7416781 KW |
3215 | |
3216 | # endif | |
3217 | ||
43dd6b15 | 3218 | } |
f7416781 KW |
3219 | } |
3220 | ||
0b6697c5 | 3221 | return retval; |
f7416781 KW |
3222 | |
3223 | #endif | |
3224 | ||
a4f00dcc | 3225 | } |
b385bb4d | 3226 | |
98994639 HS |
3227 | /* |
3228 | * Initialize locale awareness. | |
3229 | */ | |
3230 | int | |
3231 | Perl_init_i18nl10n(pTHX_ int printwarn) | |
3232 | { | |
0e92a118 KW |
3233 | /* printwarn is |
3234 | * | |
3235 | * 0 if not to output warning when setup locale is bad | |
3236 | * 1 if to output warning based on value of PERL_BADLANG | |
3237 | * >1 if to output regardless of PERL_BADLANG | |
3238 | * | |
3239 | * returns | |
98994639 | 3240 | * 1 = set ok or not applicable, |
0e92a118 KW |
3241 | * 0 = fallback to a locale of lower priority |
3242 | * -1 = fallback to all locales failed, not even to the C locale | |
6b058d42 KW |
3243 | * |
3244 | * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is | |
3245 | * set, debugging information is output. | |
3246 | * | |
3247 | * This looks more complicated than it is, mainly due to the #ifdefs. | |
3248 | * | |
3249 | * We try to set LC_ALL to the value determined by the environment. If | |
3250 | * there is no LC_ALL on this platform, we try the individual categories we | |
3251 | * know about. If this works, we are done. | |
3252 | * | |
3253 | * But if it doesn't work, we have to do something else. We search the | |
3254 | * environment variables ourselves instead of relying on the system to do | |
3255 | * it. We look at, in order, LC_ALL, LANG, a system default locale (if we | |
3256 | * think there is one), and the ultimate fallback "C". This is all done in | |
3257 | * the same loop as above to avoid duplicating code, but it makes things | |
7d4bcc4a KW |
3258 | * more complex. The 'trial_locales' array is initialized with just one |
3259 | * element; it causes the behavior described in the paragraph above this to | |
3260 | * happen. If that fails, we add elements to 'trial_locales', and do extra | |
3261 | * loop iterations to cause the behavior described in this paragraph. | |
6b058d42 KW |
3262 | * |
3263 | * On Ultrix, the locale MUST come from the environment, so there is | |
3264 | * preliminary code to set it. I (khw) am not sure that it is necessary, | |
3265 | * and that this couldn't be folded into the loop, but barring any real | |
3266 | * platforms to test on, it's staying as-is | |
3267 | * | |
3268 | * A slight complication is that in embedded Perls, the locale may already | |
3269 | * be set-up, and we don't want to get it from the normal environment | |
3270 | * variables. This is handled by having a special environment variable | |
3271 | * indicate we're in this situation. We simply set setlocale's 2nd | |
3272 | * parameter to be a NULL instead of "". That indicates to setlocale that | |
3273 | * it is not to change anything, but to return the current value, | |
3274 | * effectively initializing perl's db to what the locale already is. | |
3275 | * | |
3276 | * We play the same trick with NULL if a LC_ALL succeeds. We call | |
3277 | * setlocale() on the individual categores with NULL to get their existing | |
3278 | * values for our db, instead of trying to change them. | |
3279 | * */ | |
98994639 | 3280 | |
1565c085 DM |
3281 | dVAR; |
3282 | ||
0e92a118 KW |
3283 | int ok = 1; |
3284 | ||
7d4bcc4a KW |
3285 | #ifndef USE_LOCALE |
3286 | ||
3287 | PERL_UNUSED_ARG(printwarn); | |
3288 | ||
3289 | #else /* USE_LOCALE */ | |
7d4bcc4a KW |
3290 | # ifdef __GLIBC__ |
3291 | ||
175c4cf9 | 3292 | const char * const language = savepv(PerlEnv_getenv("LANGUAGE")); |
7d4bcc4a KW |
3293 | |
3294 | # endif | |
65ebb059 | 3295 | |
ccd65d51 KW |
3296 | /* NULL uses the existing already set up locale */ |
3297 | const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT")) | |
3298 | ? NULL | |
3299 | : ""; | |
c3fcd832 KW |
3300 | const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */ |
3301 | unsigned int trial_locales_count; | |
175c4cf9 KW |
3302 | const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL")); |
3303 | const char * const lang = savepv(PerlEnv_getenv("LANG")); | |
98994639 | 3304 | bool setlocale_failure = FALSE; |
65ebb059 | 3305 | unsigned int i; |
175c4cf9 KW |
3306 | |
3307 | /* A later getenv() could zap this, so only use here */ | |
3308 | const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG"); | |
3309 | ||
3310 | const bool locwarn = (printwarn > 1 | |
e5f10d49 KW |
3311 | || ( printwarn |
3312 | && ( ! bad_lang_use_once | |
22ff3130 | 3313 | || ( |
e5f10d49 KW |
3314 | /* disallow with "" or "0" */ |
3315 | *bad_lang_use_once | |
3316 | && strNE("0", bad_lang_use_once))))); | |
ea92aad8 | 3317 | |
291a84fb | 3318 | /* setlocale() return vals; not copied so must be looked at immediately */ |
8de4332b | 3319 | const char * sl_result[NOMINAL_LC_ALL_INDEX + 1]; |
291a84fb KW |
3320 | |
3321 | /* current locale for given category; should have been copied so aren't | |
3322 | * volatile */ | |
8de4332b | 3323 | const char * curlocales[NOMINAL_LC_ALL_INDEX + 1]; |
291a84fb | 3324 | |
7d4bcc4a KW |
3325 | # ifdef WIN32 |
3326 | ||
6bce99ee JH |
3327 | /* In some systems you can find out the system default locale |
3328 | * and use that as the fallback locale. */ | |
7d4bcc4a KW |
3329 | # define SYSTEM_DEFAULT_LOCALE |
3330 | # endif | |
3331 | # ifdef SYSTEM_DEFAULT_LOCALE | |
3332 | ||
65ebb059 | 3333 | const char *system_default_locale = NULL; |
98994639 | 3334 | |
7d4bcc4a | 3335 | # endif |
948523db KW |
3336 | |
3337 | # ifndef DEBUGGING | |
3338 | # define DEBUG_LOCALE_INIT(a,b,c) | |
3339 | # else | |
7d4bcc4a | 3340 | |
8298454c | 3341 | DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT"))); |
7d4bcc4a KW |
3342 | |
3343 | # define DEBUG_LOCALE_INIT(category, locale, result) \ | |
2fcc0ca9 KW |
3344 | STMT_START { \ |
3345 | if (debug_initialization) { \ | |
3346 | PerlIO_printf(Perl_debug_log, \ | |
3347 | "%s:%d: %s\n", \ | |
3348 | __FILE__, __LINE__, \ | |
a4f00dcc | 3349 | setlocale_debug_string(category, \ |
2fcc0ca9 KW |
3350 | locale, \ |
3351 | result)); \ | |
3352 | } \ | |
3353 | } STMT_END | |
2fcc0ca9 | 3354 | |
948523db KW |
3355 | /* Make sure the parallel arrays are properly set up */ |
3356 | # ifdef USE_LOCALE_NUMERIC | |
3357 | assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC); | |
3358 | assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC")); | |
e9bc6d6b KW |
3359 | # ifdef USE_POSIX_2008_LOCALE |
3360 | assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK); | |
3361 | # endif | |
948523db KW |
3362 | # endif |
3363 | # ifdef USE_LOCALE_CTYPE | |
3364 | assert(categories[LC_CTYPE_INDEX] == LC_CTYPE); | |
3365 | assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE")); | |
e9bc6d6b KW |
3366 | # ifdef USE_POSIX_2008_LOCALE |
3367 | assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK); | |
3368 | # endif | |
948523db KW |
3369 | # endif |
3370 | # ifdef USE_LOCALE_COLLATE | |
3371 | assert(categories[LC_COLLATE_INDEX] == LC_COLLATE); | |
3372 | assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE")); | |
e9bc6d6b KW |
3373 | # ifdef USE_POSIX_2008_LOCALE |
3374 | assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK); | |
3375 | # endif | |
948523db KW |
3376 | # endif |
3377 | # ifdef USE_LOCALE_TIME | |
3378 | assert(categories[LC_TIME_INDEX] == LC_TIME); | |
3379 | assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME")); | |
e9bc6d6b KW |
3380 | # ifdef USE_POSIX_2008_LOCALE |
3381 | assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK); | |
3382 | # endif | |
948523db KW |
3383 | # endif |
3384 | # ifdef USE_LOCALE_MESSAGES | |
3385 | assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES); | |
3386 | assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES")); | |
e9bc6d6b KW |
3387 | # ifdef USE_POSIX_2008_LOCALE |
3388 | assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK); | |
3389 | # endif | |
948523db KW |
3390 | # endif |
3391 | # ifdef USE_LOCALE_MONETARY | |
3392 | assert(categories[LC_MONETARY_INDEX] == LC_MONETARY); | |
3393 | assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY")); | |
e9bc6d6b KW |
3394 | # ifdef USE_POSIX_2008_LOCALE |
3395 | assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK); | |
3396 | # endif | |
948523db | 3397 | # endif |
9821811f KW |
3398 | # ifdef USE_LOCALE_ADDRESS |
3399 | assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS); | |
3400 | assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS")); | |
e9bc6d6b KW |
3401 | # ifdef USE_POSIX_2008_LOCALE |
3402 | assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK); | |
3403 | # endif | |
9821811f KW |
3404 | # endif |
3405 | # ifdef USE_LOCALE_IDENTIFICATION | |
3406 | assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION); | |
3407 | assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION")); | |
e9bc6d6b KW |
3408 | # ifdef USE_POSIX_2008_LOCALE |
3409 | assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK); | |
3410 | # endif | |
9821811f KW |
3411 | # endif |
3412 | # ifdef USE_LOCALE_MEASUREMENT | |
3413 | assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT); | |
3414 | assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT")); | |
e9bc6d6b KW |
3415 | # ifdef USE_POSIX_2008_LOCALE |
3416 | assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK); | |
3417 | # endif | |
9821811f KW |
3418 | # endif |
3419 | # ifdef USE_LOCALE_PAPER | |
3420 | assert(categories[LC_PAPER_INDEX] == LC_PAPER); | |
3421 | assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER")); | |
e9bc6d6b KW |
3422 | # ifdef USE_POSIX_2008_LOCALE |
3423 | assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK); | |
3424 | # endif | |
9821811f KW |
3425 | # endif |
3426 | # ifdef USE_LOCALE_TELEPHONE | |
3427 | assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE); | |
3428 | assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE")); | |
e9bc6d6b KW |
3429 | # ifdef USE_POSIX_2008_LOCALE |
3430 | assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK); | |
3431 | # endif | |
9821811f | 3432 | # endif |
948523db KW |
3433 | # ifdef LC_ALL |
3434 | assert(categories[LC_ALL_INDEX] == LC_ALL); | |
3435 | assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL")); | |
3436 | assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX); | |
e9bc6d6b KW |
3437 | # ifdef USE_POSIX_2008_LOCALE |
3438 | assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK); | |
3439 | # endif | |
948523db KW |
3440 | # endif |
3441 | # endif /* DEBUGGING */ | |
47280b20 KW |
3442 | |
3443 | /* Initialize the cache of the program's UTF-8ness for the always known | |
3444 | * locales C and POSIX */ | |
3445 | my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness, | |
3446 | sizeof(PL_locale_utf8ness)); | |
3447 | ||
e9bc6d6b KW |
3448 | # ifdef USE_THREAD_SAFE_LOCALE |
3449 | # ifdef WIN32 | |
3450 | ||
3451 | _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); | |
3452 | ||
3453 | # endif | |
3454 | # endif | |
39e69e77 | 3455 | # ifdef USE_POSIX_2008_LOCALE |
e9bc6d6b KW |
3456 | |
3457 | PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0); | |
3458 | if (! PL_C_locale_obj) { | |
3459 | Perl_croak_nocontext( | |
3460 | "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno); | |
3461 | } | |
3462 | if (DEBUG_Lv_TEST || debug_initialization) { | |
3463 | PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj); | |
3464 | } | |
3465 | ||
3466 | # endif | |
3467 | ||
78f7c09f KW |
3468 | # ifdef USE_LOCALE_NUMERIC |
3469 | ||
3ca88433 KW |
3470 | PL_numeric_radix_sv = newSVpvs("."); |
3471 | ||
78f7c09f KW |
3472 | # endif |
3473 | ||
e9bc6d6b KW |
3474 | # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE) |
3475 | ||
3476 | /* Initialize our records. If we have POSIX 2008, we have LC_ALL */ | |
3477 | do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL)); | |
3478 | ||
3479 | # endif | |
b56c4436 | 3480 | # ifdef LOCALE_ENVIRON_REQUIRED |
98994639 HS |
3481 | |
3482 | /* | |
3483 | * Ultrix setlocale(..., "") fails if there are no environment | |
3484 | * variables from which to get a locale name. | |
3485 | */ | |
3486 | ||
b56c4436 KW |
3487 | # ifndef LC_ALL |
3488 | # error Ultrix without LC_ALL not implemented | |
3489 | # else | |
7d4bcc4a | 3490 | |
b56c4436 KW |
3491 | { |
3492 | bool done = FALSE; | |
ea92aad8 KW |
3493 | if (lang) { |
3494 | sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init); | |
3495 | DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]); | |
3496 | if (sl_result[LC_ALL_INDEX]) | |
3497 | done = TRUE; | |
3498 | else | |
e5f10d49 | 3499 | setlocale_failure = TRUE; |
ea92aad8 KW |
3500 | } |
3501 | if (! setlocale_failure) { | |
3502 | const char * locale_param; | |
3503 | for (i = 0; i < LC_ALL_INDEX; i++) { | |
3504 | locale_param = (! done && (lang || PerlEnv_getenv(category_names[i]))) | |
3505 | ? setlocale_init | |
3506 | : NULL; | |
3507 | sl_result[i] = do_setlocale_r(categories[i], locale_param); | |
3508 | if (! sl_result[i]) { | |
3509 | setlocale_failure = TRUE; | |
3510 | } | |
3511 | DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]); | |
e5f10d49 | 3512 | } |
c835d6be | 3513 | } |
7d4bcc4a KW |
3514 | } |
3515 | ||
3516 | # endif /* LC_ALL */ | |
e5f10d49 | 3517 | # endif /* LOCALE_ENVIRON_REQUIRED */ |
98994639 | 3518 | |
65ebb059 | 3519 | /* We try each locale in the list until we get one that works, or exhaust |
20a240df KW |
3520 | * the list. Normally the loop is executed just once. But if setting the |
3521 | * locale fails, inside the loop we add fallback trials to the array and so | |
3522 | * will execute the loop multiple times */ | |
c3fcd832 KW |
3523 | trial_locales[0] = setlocale_init; |
3524 | trial_locales_count = 1; | |
7d4bcc4a | 3525 | |
65ebb059 KW |
3526 | for (i= 0; i < trial_locales_count; i++) { |
3527 | const char * trial_locale = trial_locales[i]; | |
3528 | ||
3529 | if (i > 0) { | |
3530 | ||
3531 | /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED | |
3532 | * when i==0, but I (khw) don't think that behavior makes much | |
3533 | * sense */ | |
3534 | setlocale_failure = FALSE; | |
3535 | ||
7d4bcc4a | 3536 | # ifdef SYSTEM_DEFAULT_LOCALE |
291a84fb | 3537 | # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */ |
7d4bcc4a | 3538 | |
65ebb059 KW |
3539 | /* On Windows machines, an entry of "" after the 0th means to use |
3540 | * the system default locale, which we now proceed to get. */ | |
3541 | if (strEQ(trial_locale, "")) { | |
3542 | unsigned int j; | |
3543 | ||
3544 | /* Note that this may change the locale, but we are going to do | |
3545 | * that anyway just below */ | |
837ce802 | 3546 | system_default_locale = do_setlocale_c(LC_ALL, ""); |
5d1187d1 | 3547 | DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale); |
65ebb059 | 3548 | |
7d4bcc4a | 3549 | /* Skip if invalid or if it's already on the list of locales to |
65ebb059 KW |
3550 | * try */ |
3551 | if (! system_default_locale) { | |
3552 | goto next_iteration; | |
3553 | } | |
3554 | for (j = 0; j < trial_locales_count; j++) { | |
3555 | if (strEQ(system_default_locale, trial_locales[j])) { | |
3556 | goto next_iteration; | |
3557 | } | |
3558 | } | |
3559 | ||
3560 | trial_locale = system_default_locale; | |
3561 | } | |
ec0202b5 KW |
3562 | # else |
3563 | # error SYSTEM_DEFAULT_LOCALE only implemented for Win32 | |
3564 | # endif | |
7d4bcc4a | 3565 | # endif /* SYSTEM_DEFAULT_LOCALE */ |
291a84fb KW |
3566 | |
3567 | } /* For i > 0 */ | |
65ebb059 | 3568 | |
7d4bcc4a KW |
3569 | # ifdef LC_ALL |
3570 | ||
948523db KW |
3571 | sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale); |
3572 | DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]); | |
3573 | if (! sl_result[LC_ALL_INDEX]) { | |
49c85077 | 3574 | setlocale_failure = TRUE; |
7cd8b568 KW |
3575 | } |
3576 | else { | |
3577 | /* Since LC_ALL succeeded, it should have changed all the other | |
3578 | * categories it can to its value; so we massage things so that the | |
3579 | * setlocales below just return their category's current values. | |
3580 | * This adequately handles the case in NetBSD where LC_COLLATE may | |
3581 | * not be defined for a locale, and setting it individually will | |
7d4bcc4a | 3582 | * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to |
7cd8b568 KW |
3583 | * the POSIX locale. */ |
3584 | trial_locale = NULL; | |
3585 | } | |
7d4bcc4a KW |
3586 | |
3587 | # endif /* LC_ALL */ | |
98994639 | 3588 | |
e5f10d49 KW |
3589 | if (! setlocale_failure) { |
3590 | unsigned int j; | |
3591 | for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) { | |
3592 | curlocales[j] | |
3593 | = savepv(do_setlocale_r(categories[j], trial_locale)); | |
3594 | if (! curlocales[j]) { | |
3595 | setlocale_failure = TRUE; | |
3596 | } | |
3597 | DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]); | |
3598 | } | |
c835d6be | 3599 | |
e5f10d49 KW |
3600 | if (! setlocale_failure) { /* All succeeded */ |
3601 | break; /* Exit trial_locales loop */ | |
49c85077 | 3602 | } |
65ebb059 | 3603 | } |
98994639 | 3604 | |
49c85077 KW |
3605 | /* Here, something failed; will need to try a fallback. */ |
3606 | ok = 0; | |
65ebb059 | 3607 | |
49c85077 KW |
3608 | if (i == 0) { |
3609 | unsigned int j; | |
98994639 | 3610 | |
65ebb059 | 3611 | if (locwarn) { /* Output failure info only on the first one */ |
7d4bcc4a KW |
3612 | |
3613 | # ifdef LC_ALL | |
98994639 | 3614 | |
49c85077 KW |
3615 | PerlIO_printf(Perl_error_log, |
3616 | "perl: warning: Setting locale failed.\n"); | |
98994639 | 3617 | |
7d4bcc4a | 3618 | # else /* !LC_ALL */ |
98994639 | 3619 | |
49c85077 KW |
3620 | PerlIO_printf(Perl_error_log, |
3621 | "perl: warning: Setting locale failed for the categories:\n\t"); | |
7d4bcc4a | 3622 | |
e5f10d49 KW |
3623 | for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) { |
3624 | if (! curlocales[j]) { | |
3625 | PerlIO_printf(Perl_error_log, category_names[j]); | |
3626 | } | |
3627 | else { | |
3628 | Safefree(curlocales[j]); | |
3629 | } | |
3630 | } | |
7d4bcc4a | 3631 | |
7d4bcc4a | 3632 | # endif /* LC_ALL */ |
98994639 | 3633 | |
49c85077 KW |
3634 | PerlIO_printf(Perl_error_log, |
3635 | "perl: warning: Please check that your locale settings:\n"); | |
98994639 | 3636 | |
7d4bcc4a KW |
3637 | # ifdef __GLIBC__ |
3638 | ||
49c85077 KW |
3639 | PerlIO_printf(Perl_error_log, |
3640 | "\tLANGUAGE = %c%s%c,\n", | |
3641 | language ? '"' : '(', | |
3642 | language ? language : "unset", | |
3643 | language ? '"' : ')'); | |
7d4bcc4a | 3644 | # endif |
98994639 | 3645 | |
49c85077 KW |
3646 | PerlIO_printf(Perl_error_log, |
3647 | "\tLC_ALL = %c%s%c,\n", | |
3648 | lc_all ? '"' : '(', | |
3649 | lc_all ? lc_all : "unset", | |
3650 | lc_all ? '"' : ')'); | |
98994639 | 3651 | |
7d4bcc4a KW |
3652 | # if defined(USE_ENVIRON_ARRAY) |
3653 | ||
49c85077 | 3654 | { |
cd999af9 | 3655 | char **e; |
d5e32b93 KW |
3656 | |
3657 | /* Look through the environment for any variables of the | |
3658 | * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was | |
3659 | * already handled above. These are assumed to be locale | |
3660 | * settings. Output them and their values. */ | |
cd999af9 | 3661 | for (e = environ; *e; e++) { |
d5e32b93 KW |
3662 | const STRLEN prefix_len = sizeof("LC_") - 1; |
3663 | STRLEN uppers_len; | |
3664 | ||
cd999af9 | 3665 | if ( strBEGINs(*e, "LC_") |
c8b388b0 | 3666 | && ! strBEGINs(*e, "LC_ALL=") |
d5e32b93 KW |
3667 | && (uppers_len = strspn(*e + prefix_len, |
3668 | "ABCDEFGHIJKLMNOPQRSTUVWXYZ")) | |
3669 | && ((*e)[prefix_len + uppers_len] == '=')) | |
cd999af9 KW |
3670 | { |
3671 | PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n", | |
d5e32b93 KW |
3672 | (int) (prefix_len + uppers_len), *e, |
3673 | *e + prefix_len + uppers_len + 1); | |
cd999af9 KW |
3674 | } |
3675 | } | |
49c85077 | 3676 | } |
7d4bcc4a KW |
3677 | |
3678 | # else | |
3679 | ||
49c85077 KW |
3680 | PerlIO_printf(Perl_error_log, |
3681 | "\t(possibly more locale environment variables)\n"); | |
7d4bcc4a KW |
3682 | |
3683 | # endif | |
98994639 | 3684 | |
49c85077 KW |
3685 | PerlIO_printf(Perl_error_log, |
3686 | "\tLANG = %c%s%c\n", | |
3687 | lang ? '"' : '(', | |
3688 | lang ? lang : "unset", | |
3689 | lang ? '"' : ')'); | |
98994639 | 3690 | |
49c85077 KW |
3691 | PerlIO_printf(Perl_error_log, |
3692 | " are supported and installed on your system.\n"); | |
3693 | } | |
98994639 | 3694 | |
65ebb059 | 3695 | /* Calculate what fallback locales to try. We have avoided this |
f6bab5f6 | 3696 | * until we have to, because failure is quite unlikely. This will |
65ebb059 KW |
3697 | * usually change the upper bound of the loop we are in. |
3698 | * | |
3699 | * Since the system's default way of setting the locale has not | |
3700 | * found one that works, We use Perl's defined ordering: LC_ALL, | |
3701 | * LANG, and the C locale. We don't try the same locale twice, so | |
3702 | * don't add to the list if already there. (On POSIX systems, the | |
3703 | * LC_ALL element will likely be a repeat of the 0th element "", | |
6b058d42 KW |
3704 | * but there's no harm done by doing it explicitly. |
3705 | * | |
3706 | * Note that this tries the LC_ALL environment variable even on | |
3707 | * systems which have no LC_ALL locale setting. This may or may | |
3708 | * not have been originally intentional, but there's no real need | |
3709 | * to change the behavior. */ | |
65ebb059 KW |
3710 | if (lc_all) { |
3711 | for (j = 0; j < trial_locales_count; j++) { | |
3712 | if (strEQ(lc_all, trial_locales[j])) { | |
3713 | goto done_lc_all; | |
3714 | } | |
3715 | } | |
3716 | trial_locales[trial_locales_count++] = lc_all; | |
3717 | } | |
3718 | done_lc_all: | |
98994639 | 3719 | |
65ebb059 KW |
3720 | if (lang) { |
3721 | for (j = 0; j < trial_locales_count; j++) { | |
3722 | if (strEQ(lang, trial_locales[j])) { | |
3723 | goto done_lang; | |
3724 | } | |
3725 | } | |
3726 | trial_locales[trial_locales_count++] = lang; | |
3727 | } | |
3728 | done_lang: | |
3729 | ||
7d4bcc4a KW |
3730 | # if defined(WIN32) && defined(LC_ALL) |
3731 | ||
65ebb059 KW |
3732 | /* For Windows, we also try the system default locale before "C". |
3733 | * (If there exists a Windows without LC_ALL we skip this because | |
3734 | * it gets too complicated. For those, the "C" is the next | |
3735 | * fallback possibility). The "" is the same as the 0th element of | |
3736 | * the array, but the code at the loop above knows to treat it | |
3737 | * differently when not the 0th */ | |
3738 | trial_locales[trial_locales_count++] = ""; | |
7d4bcc4a KW |
3739 | |
3740 | # endif | |
65ebb059 KW |
3741 | |
3742 | for (j = 0; j < trial_locales_count; j++) { | |
3743 | if (strEQ("C", trial_locales[j])) { | |
3744 | goto done_C; | |
3745 | } | |
3746 | } | |
3747 | trial_locales[trial_locales_count++] = "C"; | |
98994639 | 3748 | |
65ebb059 KW |
3749 | done_C: ; |
3750 | } /* end of first time through the loop */ | |
98994639 | 3751 | |
7d4bcc4a KW |
3752 | # ifdef WIN32 |
3753 | ||
65ebb059 | 3754 | next_iteration: ; |
7d4bcc4a KW |
3755 | |
3756 | # endif | |
65ebb059 KW |
3757 | |
3758 | } /* end of looping through the trial locales */ | |
3759 | ||
3760 | if (ok < 1) { /* If we tried to fallback */ | |
3761 | const char* msg; | |
3762 | if (! setlocale_failure) { /* fallback succeeded */ | |
3763 | msg = "Falling back to"; | |
3764 | } | |
3765 | else { /* fallback failed */ | |
e5f10d49 | 3766 | unsigned int j; |
98994639 | 3767 | |
65ebb059 KW |
3768 | /* We dropped off the end of the loop, so have to decrement i to |
3769 | * get back to the value the last time through */ | |
3770 | i--; | |
98994639 | 3771 | |
65ebb059 KW |
3772 | ok = -1; |
3773 | msg = "Failed to fall back to"; | |
3774 | ||
3775 | /* To continue, we should use whatever values we've got */ | |
7d4bcc4a | 3776 | |
e5f10d49 KW |
3777 | for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) { |
3778 | Safefree(curlocales[j]); | |
3779 | curlocales[j] = savepv(do_setlocale_r(categories[j], NULL)); | |
3780 | DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]); | |
3781 | } | |
65ebb059 KW |
3782 | } |
3783 | ||
3784 | if (locwarn) { | |
3785 | const char * description; | |
3786 | const char * name = ""; | |
3787 | if (strEQ(trial_locales[i], "C")) { | |
3788 | description = "the standard locale"; | |
3789 | name = "C"; | |
3790 | } | |
7d4bcc4a KW |
3791 | |
3792 | # ifdef SYSTEM_DEFAULT_LOCALE | |
3793 | ||
65ebb059 KW |
3794 | else if (strEQ(trial_locales[i], "")) { |
3795 | description = "the system default locale"; | |
3796 | if (system_default_locale) { | |
3797 | name = system_default_locale; | |
3798 | } | |
3799 | } | |
7d4bcc4a KW |
3800 | |
3801 | # endif /* SYSTEM_DEFAULT_LOCALE */ | |
3802 | ||
65ebb059 KW |
3803 | else { |
3804 | description = "a fallback locale"; | |
3805 | name = trial_locales[i]; | |
3806 | } | |
3807 | if (name && strNE(name, "")) { | |
3808 | PerlIO_printf(Perl_error_log, | |
3809 | "perl: warning: %s %s (\"%s\").\n", msg, description, name); | |
3810 | } | |
3811 | else { | |
3812 | PerlIO_printf(Perl_error_log, | |
3813 | "perl: warning: %s %s.\n", msg, description); | |
3814 | } | |
3815 | } | |
3816 | } /* End of tried to fallback */ | |
98994639 | 3817 | |
e5f10d49 KW |
3818 | /* Done with finding the locales; update our records */ |
3819 | ||
7d4bcc4a KW |
3820 | # ifdef USE_LOCALE_CTYPE |
3821 | ||
948523db | 3822 | new_ctype(curlocales[LC_CTYPE_INDEX]); |
98994639 | 3823 | |
e5f10d49 | 3824 | # endif |
7d4bcc4a KW |
3825 | # ifdef USE_LOCALE_COLLATE |
3826 | ||
948523db | 3827 | new_collate(curlocales[LC_COLLATE_INDEX]); |
98994639 | 3828 | |
e5f10d49 | 3829 | # endif |
7d4bcc4a KW |
3830 | # ifdef USE_LOCALE_NUMERIC |
3831 | ||
948523db | 3832 | new_numeric(curlocales[LC_NUMERIC_INDEX]); |
e5f10d49 KW |
3833 | |
3834 | # endif | |
3835 | ||
948523db | 3836 | for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) { |
47280b20 | 3837 | |
e9bc6d6b | 3838 | # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE) |
47280b20 KW |
3839 | |
3840 | /* This caches whether each category's locale is UTF-8 or not. This | |
3841 | * may involve changing the locale. It is ok to do this at | |
e9bc6d6b KW |
3842 | * initialization time before any threads have started, but not later |
3843 | * unless thread-safe operations are used. | |
47280b20 KW |
3844 | * Caching means that if the program heeds our dictate not to change |
3845 | * locales in threaded applications, this data will remain valid, and | |
9de6fe47 KW |
3846 | * it may get queried without having to change locales. If the |
3847 | * environment is such that all categories have the same locale, this | |
3848 | * isn't needed, as the code will not change the locale; but this | |
3849 | * handles the uncommon case where the environment has disparate | |
3850 | * locales for the categories */ | |
47280b20 KW |
3851 | (void) _is_cur_LC_category_utf8(categories[i]); |
3852 | ||
3853 | # endif | |
3854 | ||
e5f10d49 KW |
3855 | Safefree(curlocales[i]); |
3856 | } | |
b310b053 | 3857 | |
7d4bcc4a KW |
3858 | # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE) |
3859 | ||
49c85077 | 3860 | /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE |
50bf02bd KW |
3861 | * locale is UTF-8. The call to new_ctype() just above has already |
3862 | * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If | |
3863 | * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE}) | |
3864 | * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on | |
3865 | * STDIN, STDOUT, STDERR, _and_ the default open discipline. */ | |
3866 | PL_utf8locale = PL_in_utf8_CTYPE_locale; | |
49c85077 | 3867 | |
a05d7ebb | 3868 | /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO. |
fde18df1 JH |
3869 | This is an alternative to using the -C command line switch |
3870 | (the -C if present will override this). */ | |
3871 | { | |
dd374669 | 3872 | const char *p = PerlEnv_getenv("PERL_UNICODE"); |
a05d7ebb | 3873 | PL_unicode = p ? parse_unicode_opts(&p) : 0; |
5a22a2bb NC |
3874 | if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG) |
3875 | PL_utf8cache = -1; | |
b310b053 JH |
3876 | } |
3877 | ||
7d4bcc4a | 3878 | # endif |
7d4bcc4a KW |
3879 | # ifdef __GLIBC__ |
3880 | ||
175c4cf9 | 3881 | Safefree(language); |
7d4bcc4a KW |
3882 | |
3883 | # endif | |
175c4cf9 KW |
3884 | |
3885 | Safefree(lc_all); | |
3886 | Safefree(lang); | |
3887 | ||
e3305790 | 3888 | #endif /* USE_LOCALE */ |
2fcc0ca9 | 3889 | #ifdef DEBUGGING |
7d4bcc4a | 3890 | |
2fcc0ca9 | 3891 | /* So won't continue to output stuff */ |
27cdc72e | 3892 | DEBUG_INITIALIZATION_set(FALSE); |
7d4bcc4a | 3893 | |
2fcc0ca9 KW |
3894 | #endif |
3895 | ||
98994639 HS |
3896 | return ok; |
3897 | } | |
3898 | ||
98994639 HS |
3899 | #ifdef USE_LOCALE_COLLATE |
3900 | ||
a4a439fb | 3901 | char * |
a4a439fb KW |
3902 | Perl__mem_collxfrm(pTHX_ const char *input_string, |
3903 | STRLEN len, /* Length of 'input_string' */ | |
3904 | STRLEN *xlen, /* Set to length of returned string | |
3905 | (not including the collation index | |
3906 | prefix) */ | |
3907 | bool utf8 /* Is the input in UTF-8? */ | |
6696cfa7 | 3908 | ) |
98994639 | 3909 | { |
a4a439fb KW |
3910 | |
3911 | /* _mem_collxfrm() is a bit like strxfrm() but with two important | |
3912 | * differences. First, it handles embedded NULs. Second, it allocates a bit | |
3913 | * more memory than needed for the transformed data itself. The real | |
55e5378d | 3914 | * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to |
a4a439fb KW |
3915 | * the length of that, and doesn't include the collation index size. |
3916 | * Please see sv_collxfrm() to see how this is used. */ | |
3917 | ||
55e5378d KW |
3918 | #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix) |
3919 | ||
6696cfa7 KW |
3920 | char * s = (char *) input_string; |
3921 | STRLEN s_strlen = strlen(input_string); | |
79f120c8 | 3922 | char *xbuf = NULL; |
55e5378d | 3923 | STRLEN xAlloc; /* xalloc is a reserved word in VC */ |
17f41037 | 3924 | STRLEN length_in_chars; |
c664130f | 3925 | bool first_time = TRUE; /* Cleared after first loop iteration */ |
98994639 | 3926 | |
a4a439fb KW |
3927 | PERL_ARGS_ASSERT__MEM_COLLXFRM; |
3928 | ||
3929 | /* Must be NUL-terminated */ | |
3930 | assert(*(input_string + len) == '\0'); | |
7918f24d | 3931 | |
79f120c8 KW |
3932 | /* If this locale has defective collation, skip */ |
3933 | if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) { | |
c7202dee KW |
3934 | DEBUG_L(PerlIO_printf(Perl_debug_log, |
3935 | "_mem_collxfrm: locale's collation is defective\n")); | |
79f120c8 KW |
3936 | goto bad; |
3937 | } | |
3938 | ||
6696cfa7 KW |
3939 | /* Replace any embedded NULs with the control that sorts before any others. |
3940 | * This will give as good as possible results on strings that don't | |
3941 | * otherwise contain that character, but otherwise there may be | |
3942 | * less-than-perfect results with that character and NUL. This is | |
fdc080f3 | 3943 | * unavoidable unless we replace strxfrm with our own implementation. */ |
fd43f63c KW |
3944 | if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded |
3945 | NUL */ | |
6696cfa7 KW |
3946 | char * e = s + len; |
3947 | char * sans_nuls; | |
fdc080f3 | 3948 | STRLEN sans_nuls_len; |
94762aa0 | 3949 | int try_non_controls; |
afc4976f KW |
3950 | char this_replacement_char[] = "?\0"; /* Room for a two-byte string, |
3951 | making sure 2nd byte is NUL. | |
3952 | */ | |
3953 | STRLEN this_replacement_len; | |
3954 | ||
1e4c9676 KW |
3955 | /* If we don't know what non-NUL control character sorts lowest for |
3956 | * this locale, find it */ | |
f28f4d2a | 3957 | if (PL_strxfrm_NUL_replacement == '\0') { |
6696cfa7 | 3958 | int j; |
afc4976f | 3959 | char * cur_min_x = NULL; /* The min_char's xfrm, (except it also |
6696cfa7 KW |
3960 | includes the collation index |
3961 | prefixed. */ | |
3962 | ||
91c0e2e0 | 3963 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n")); |
94762aa0 KW |
3964 | |
3965 | /* Unlikely, but it may be that no control will work to replace | |
1e4c9676 KW |
3966 | * NUL, in which case we instead look for any character. Controls |
3967 | * are preferred because collation order is, in general, context | |
3968 | * sensitive, with adjoining characters affecting the order, and | |
3969 | * controls are less likely to have such interactions, allowing the | |
3970 | * NUL-replacement to stand on its own. (Another way to look at it | |
3971 | * is to imagine what would happen if the NUL were replaced by a | |
3972 | * combining character; it wouldn't work out all that well.) */ | |
94762aa0 KW |
3973 | for (try_non_controls = 0; |
3974 | try_non_controls < 2; | |
3975 | try_non_controls++) | |
3976 | { | |
d4ff9586 KW |
3977 | /* Look through all legal code points (NUL isn't) */ |
3978 | for (j = 1; j < 256; j++) { | |
3979 | char * x; /* j's xfrm plus collation index */ | |
3980 | STRLEN x_len; /* length of 'x' */ | |
3981 | STRLEN trial_len = 1; | |
736a4fed | 3982 | char cur_source[] = { '\0', '\0' }; |
d4ff9586 | 3983 | |
736a4fed KW |
3984 | /* Skip non-controls the first time through the loop. The |
3985 | * controls in a UTF-8 locale are the L1 ones */ | |
afc4976f KW |
3986 | if (! try_non_controls && (PL_in_utf8_COLLATE_locale) |
3987 | ? ! isCNTRL_L1(j) | |
3988 | : ! isCNTRL_LC(j)) | |
3989 | { | |
d4ff9586 | 3990 | continue; |
6696cfa7 | 3991 | } |
6696cfa7 | 3992 | |
736a4fed KW |
3993 | /* Create a 1-char string of the current code point */ |
3994 | cur_source[0] = (char) j; | |
3995 | ||
d4ff9586 KW |
3996 | /* Then transform it */ |
3997 | x = _mem_collxfrm(cur_source, trial_len, &x_len, | |
afc4976f | 3998 | 0 /* The string is not in UTF-8 */); |
6696cfa7 | 3999 | |
1e4c9676 | 4000 | /* Ignore any character that didn't successfully transform. |
d4ff9586 KW |
4001 | * */ |
4002 | if (! x) { | |
4003 | continue; | |
4004 | } | |
6696cfa7 | 4005 | |
d4ff9586 KW |
4006 | /* If this character's transformation is lower than |
4007 | * the current lowest, this one becomes the lowest */ | |
4008 | if ( cur_min_x == NULL | |
4009 | || strLT(x + COLLXFRM_HDR_LEN, | |
4010 | cur_min_x + COLLXFRM_HDR_LEN)) | |
4011 | { | |
f28f4d2a | 4012 | PL_strxfrm_NUL_replacement = j; |
62d66863 | 4013 | Safefree(cur_min_x); |
d4ff9586 | 4014 | cur_min_x = x; |
d4ff9586 KW |
4015 | } |
4016 | else { | |
4017 | Safefree(x); | |
4018 | } | |
1e4c9676 | 4019 | } /* end of loop through all 255 characters */ |
6696cfa7 | 4020 | |
1e4c9676 | 4021 | /* Stop looking if found */ |
94762aa0 KW |
4022 | if (cur_min_x) { |
4023 | break; | |
4024 | } | |
4025 | ||
4026 | /* Unlikely, but possible, if there aren't any controls that | |
4027 | * work in the locale, repeat the loop, looking for any | |
4028 | * character that works */ | |
4029 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
4030 | "_mem_collxfrm: No control worked. Trying non-controls\n")); | |
1e4c9676 | 4031 | } /* End of loop to try first the controls, then any char */ |
6696cfa7 | 4032 | |
94762aa0 KW |
4033 | if (! cur_min_x) { |
4034 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
4035 | "_mem_collxfrm: Couldn't find any character to replace" | |
4036 | " embedded NULs in locale %s with", PL_collation_name)); | |
4037 | goto bad; | |
58eebef2 KW |
4038 | } |
4039 | ||
94762aa0 KW |
4040 | DEBUG_L(PerlIO_printf(Perl_debug_log, |
4041 | "_mem_collxfrm: Replacing embedded NULs in locale %s with " | |
f28f4d2a | 4042 | "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement)); |
94762aa0 | 4043 | |
6696cfa7 | 4044 | Safefree(cur_min_x); |
1e4c9676 | 4045 | } /* End of determining the character that is to replace NULs */ |
afc4976f KW |
4046 | |
4047 | /* If the replacement is variant under UTF-8, it must match the | |
291a84fb | 4048 | * UTF8-ness of the original */ |
f28f4d2a KW |
4049 | if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) { |
4050 | this_replacement_char[0] = | |
4051 | UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement); | |
4052 | this_replacement_char[1] = | |
4053 | UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement); | |
afc4976f KW |
4054 | this_replacement_len = 2; |
4055 | } | |
4056 | else { | |
f28f4d2a | 4057 | this_replacement_char[0] = PL_strxfrm_NUL_replacement; |
afc4976f KW |
4058 | /* this_replacement_char[1] = '\0' was done at initialization */ |
4059 | this_replacement_len = 1; | |
6696cfa7 KW |
4060 | } |
4061 | ||
4062 | /* The worst case length for the replaced string would be if every | |
4063 | * character in it is NUL. Multiply that by the length of each | |
4064 | * replacement, and allow for a trailing NUL */ | |
afc4976f | 4065 | sans_nuls_len = (len * this_replacement_len) + 1; |
fdc080f3 | 4066 | Newx(sans_nuls, sans_nuls_len, char); |
6696cfa7 KW |
4067 | *sans_nuls = '\0'; |
4068 | ||
6696cfa7 KW |
4069 | /* Replace each NUL with the lowest collating control. Loop until have |
4070 | * exhausted all the NULs */ | |
4071 | while (s + s_strlen < e) { | |
6069d6c5 | 4072 | my_strlcat(sans_nuls, s, sans_nuls_len); |
6696cfa7 KW |
4073 | |
4074 | /* Do the actual replacement */ | |
6069d6c5 | 4075 | my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len); |
6696cfa7 KW |
4076 | |
4077 | /* Move past the input NUL */ | |
4078 | s += s_strlen + 1; | |
4079 | s_strlen = strlen(s); | |
4080 | } | |
4081 | ||
4082 | /* And add anything that trails the final NUL */ | |
6069d6c5 | 4083 | my_strlcat(sans_nuls, s, sans_nuls_len); |
6696cfa7 KW |
4084 | |
4085 | /* Switch so below we transform this modified string */ | |
4086 | s = sans_nuls; | |
4087 | len = strlen(s); | |
1e4c9676 | 4088 | } /* End of replacing NULs */ |
6696cfa7 | 4089 | |
a4a439fb KW |
4090 | /* Make sure the UTF8ness of the string and locale match */ |
4091 | if (utf8 != PL_in_utf8_COLLATE_locale) { | |
9de6fe47 | 4092 | /* XXX convert above Unicode to 10FFFF? */ |
a4a439fb KW |
4093 | const char * const t = s; /* Temporary so we can later find where the |
4094 | input was */ | |
4095 | ||
4096 | /* Here they don't match. Change the string's to be what the locale is | |
4097 | * expecting */ | |
4098 | ||
4099 | if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */ | |
4100 | s = (char *) bytes_to_utf8((const U8 *) s, &len); | |
4101 | utf8 = TRUE; | |
4102 | } | |
4103 | else { /* locale is not UTF-8; but input is; downgrade the input */ | |
4104 | ||
4105 | s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8); | |
4106 | ||
4107 | /* If the downgrade was successful we are done, but if the input | |
4108 | * contains things that require UTF-8 to represent, have to do | |
4109 | * damage control ... */ | |
4110 | if (UNLIKELY(utf8)) { | |
4111 | ||
4112 | /* What we do is construct a non-UTF-8 string with | |
4113 | * 1) the characters representable by a single byte converted | |
4114 | * to be so (if necessary); | |
4115 | * 2) and the rest converted to collate the same as the | |
4116 | * highest collating representable character. That makes | |
4117 | * them collate at the end. This is similar to how we | |
4118 | * handle embedded NULs, but we use the highest collating | |
4119 | * code point instead of the smallest. Like the NUL case, | |
4120 | * this isn't perfect, but is the best we can reasonably | |
4121 | * do. Every above-255 code point will sort the same as | |
4122 | * the highest-sorting 0-255 code point. If that code | |
4123 | * point can combine in a sequence with some other code | |
4124 | * points for weight calculations, us changing something to | |
4125 | * be it can adversely affect the results. But in most | |
4126 | * cases, it should work reasonably. And note that this is | |
4127 | * really an illegal situation: using code points above 255 | |
4128 | * on a locale where only 0-255 are valid. If two strings | |
4129 | * sort entirely equal, then the sort order for the | |
4130 | * above-255 code points will be in code point order. */ | |
4131 | ||
4132 | utf8 = FALSE; | |
4133 | ||
4134 | /* If we haven't calculated the code point with the maximum | |
4135 | * collating order for this locale, do so now */ | |
4136 | if (! PL_strxfrm_max_cp) { | |
4137 | int j; | |
4138 | ||
4139 | /* The current transformed string that collates the | |
4140 | * highest (except it also includes the prefixed collation | |
4141 | * index. */ | |
4142 | char * cur_max_x = NULL; | |
4143 | ||
4144 | /* Look through all legal code points (NUL isn't) */ | |
4145 | for (j = 1; j < 256; j++) { | |
4146 | char * x; | |
4147 | STRLEN x_len; | |
736a4fed | 4148 | char cur_source[] = { '\0', '\0' }; |
a4a439fb | 4149 | |
736a4fed KW |
4150 | /* Create a 1-char string of the current code point */ |
4151 | cur_source[0] = (char) j; | |
a4a439fb KW |
4152 | |
4153 | /* Then transform it */ | |
4154 | x = _mem_collxfrm(cur_source, 1, &x_len, FALSE); | |
4155 | ||
4156 | /* If something went wrong (which it shouldn't), just | |
4157 | * ignore this code point */ | |
94762aa0 | 4158 | if (! x) { |
a4a439fb KW |
4159 | continue; |
4160 | } | |
4161 | ||
4162 | /* If this character's transformation is higher than | |
4163 | * the current highest, this one becomes the highest */ | |
4164 | if ( cur_max_x == NULL | |
55e5378d KW |
4165 | || strGT(x + COLLXFRM_HDR_LEN, |
4166 | cur_max_x + COLLXFRM_HDR_LEN)) | |
a4a439fb KW |
4167 | { |
4168 | PL_strxfrm_max_cp = j; | |
62d66863 | 4169 | Safefree(cur_max_x); |
a4a439fb KW |
4170 | cur_max_x = x; |
4171 | } | |
4172 | else { | |
4173 | Safefree(x); | |
4174 | } | |
4175 | } | |
4176 | ||
94762aa0 KW |
4177 | if (! cur_max_x) { |
4178 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
4179 | "_mem_collxfrm: Couldn't find any character to" | |
4180 | " replace above-Latin1 chars in locale %s with", | |
4181 | PL_collation_name)); | |
4182 | goto bad; | |
4183 | } | |
4184 | ||
58eebef2 KW |
4185 | DEBUG_L(PerlIO_printf(Perl_debug_log, |
4186 | "_mem_collxfrm: highest 1-byte collating character" | |
4187 | " in locale %s is 0x%02X\n", | |
4188 | PL_collation_name, | |
4189 | PL_strxfrm_max_cp)); | |
58eebef2 | 4190 | |
a4a439fb KW |
4191 | Safefree(cur_max_x); |
4192 | } | |
4193 | ||
4194 | /* Here we know which legal code point collates the highest. | |
4195 | * We are ready to construct the non-UTF-8 string. The length | |
4196 | * will be at least 1 byte smaller than the input string | |
4197 | * (because we changed at least one 2-byte character into a | |
4198 | * single byte), but that is eaten up by the trailing NUL */ | |
4199 | Newx(s, len, char); | |
4200 | ||
4201 | { | |
4202 | STRLEN i; | |
4203 | STRLEN d= 0; | |
042d9e50 | 4204 | char * e = (char *) t + len; |
a4a439fb KW |
4205 | |
4206 | for (i = 0; i < len; i+= UTF8SKIP(t + i)) { | |
4207 | U8 cur_char = t[i]; | |
4208 | if (UTF8_IS_INVARIANT(cur_char)) { | |
4209 | s[d++] = cur_char; | |
4210 | } | |
042d9e50 | 4211 | else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) { |
a4a439fb KW |
4212 | s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]); |
4213 | } | |
4214 | else { /* Replace illegal cp with highest collating | |
4215 | one */ | |
4216 | s[d++] = PL_strxfrm_max_cp; | |
4217 | } | |
4218 | } | |
4219 | s[d++] = '\0'; | |
4220 | Renew(s, d, char); /* Free up unused space */ | |
4221 | } | |
4222 | } | |
4223 | } | |
4224 | ||
4225 | /* Here, we have constructed a modified version of the input. It could | |
4226 | * be that we already had a modified copy before we did this version. | |
4227 | * If so, that copy is no longer needed */ | |
4228 | if (t != input_string) { | |
4229 | Safefree(t); | |
4230 | } | |
4231 | } | |
4232 | ||
17f41037 KW |
4233 | length_in_chars = (utf8) |
4234 | ? utf8_length((U8 *) s, (U8 *) s + len) | |
4235 | : len; | |
4236 | ||
59c018b9 KW |
4237 | /* The first element in the output is the collation id, used by |
4238 | * sv_collxfrm(); then comes the space for the transformed string. The | |
4239 | * equation should give us a good estimate as to how much is needed */ | |
55e5378d | 4240 | xAlloc = COLLXFRM_HDR_LEN |
a4a439fb | 4241 | + PL_collxfrm_base |
17f41037 | 4242 | + (PL_collxfrm_mult * length_in_chars); |
a02a5408 | 4243 | Newx(xbuf, xAlloc, char); |
c7202dee KW |
4244 | if (UNLIKELY(! xbuf)) { |
4245 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
4246 | "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc)); | |
98994639 | 4247 | goto bad; |
c7202dee | 4248 | } |
98994639 | 4249 | |
d35fca5f | 4250 | /* Store the collation id */ |
98994639 | 4251 | *(U32*)xbuf = PL_collation_ix; |
d35fca5f KW |
4252 | |
4253 | /* Then the transformation of the input. We loop until successful, or we | |
4254 | * give up */ | |
4ebeff16 | 4255 | for (;;) { |
1adab0a7 | 4256 | |
55e5378d | 4257 | *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN); |
4ebeff16 KW |
4258 | |
4259 | /* If the transformed string occupies less space than we told strxfrm() | |
4260 | * was available, it means it successfully transformed the whole | |
4261 | * string. */ | |
55e5378d | 4262 | if (*xlen < xAlloc - COLLXFRM_HDR_LEN) { |
17f41037 | 4263 | |
1adab0a7 KW |
4264 | /* Some systems include a trailing NUL in the returned length. |
4265 | * Ignore it, using a loop in case multiple trailing NULs are | |
4266 | * returned. */ | |
4267 | while ( (*xlen) > 0 | |
4268 | && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0') | |
4269 | { | |
4270 | (*xlen)--; | |
4271 | } | |
4272 | ||
17f41037 KW |
4273 | /* If the first try didn't get it, it means our prediction was low. |
4274 | * Modify the coefficients so that we predict a larger value in any | |
4275 | * future transformations */ | |
4276 | if (! first_time) { | |
4277 | STRLEN needed = *xlen + 1; /* +1 For trailing NUL */ | |
4278 | STRLEN computed_guess = PL_collxfrm_base | |
4279 | + (PL_collxfrm_mult * length_in_chars); | |
e1c30f0c KW |
4280 | |
4281 | /* On zero-length input, just keep current slope instead of | |
4282 | * dividing by 0 */ | |
4283 | const STRLEN new_m = (length_in_chars != 0) | |
4284 | ? needed / length_in_chars | |
4285 | : PL_collxfrm_mult; | |
17f41037 KW |
4286 | |
4287 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, | |
b07929e4 KW |
4288 | "%s: %d: initial size of %zu bytes for a length " |
4289 | "%zu string was insufficient, %zu needed\n", | |
17f41037 | 4290 | __FILE__, __LINE__, |
b07929e4 | 4291 | computed_guess, length_in_chars, needed)); |
17f41037 KW |
4292 | |
4293 | /* If slope increased, use it, but discard this result for | |
4294 | * length 1 strings, as we can't be sure that it's a real slope | |
4295 | * change */ | |
4296 | if (length_in_chars > 1 && new_m > PL_collxfrm_mult) { | |
7d4bcc4a KW |
4297 | |
4298 | # ifdef DEBUGGING | |
4299 | ||
17f41037 KW |
4300 | STRLEN old_m = PL_collxfrm_mult; |
4301 | STRLEN old_b = PL_collxfrm_base; | |
7d4bcc4a KW |
4302 | |
4303 | # endif | |
4304 | ||
17f41037 KW |
4305 | PL_collxfrm_mult = new_m; |
4306 | PL_collxfrm_base = 1; /* +1 For trailing NUL */ | |
4307 | computed_guess = PL_collxfrm_base | |
4308 | + (PL_collxfrm_mult * length_in_chars); | |
4309 | if (computed_guess < needed) { | |
4310 | PL_collxfrm_base += needed - computed_guess; | |
4311 | } | |
4312 | ||
4313 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, | |
b07929e4 KW |
4314 | "%s: %d: slope is now %zu; was %zu, base " |
4315 | "is now %zu; was %zu\n", | |
17f41037 | 4316 | __FILE__, __LINE__, |
b07929e4 KW |
4317 | PL_collxfrm_mult, old_m, |
4318 | PL_collxfrm_base, old_b)); | |
17f41037 KW |
4319 | } |
4320 | else { /* Slope didn't change, but 'b' did */ | |
4321 | const STRLEN new_b = needed | |
4322 | - computed_guess | |
4323 | + PL_collxfrm_base; | |
4324 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, | |
b07929e4 | 4325 | "%s: %d: base is now %zu; was %zu\n", |
17f41037 | 4326 | __FILE__, __LINE__, |
b07929e4 | 4327 | new_b, PL_collxfrm_base)); |
17f41037 KW |
4328 | PL_collxfrm_base = new_b; |
4329 | } | |
4330 | } | |
4331 | ||
4ebeff16 KW |
4332 | break; |
4333 | } | |
bb0f664e | 4334 | |
c7202dee KW |
4335 | if (UNLIKELY(*xlen >= PERL_INT_MAX)) { |
4336 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
4337 | "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n", | |
4338 | *xlen, PERL_INT_MAX)); | |
4ebeff16 | 4339 | goto bad; |
c7202dee | 4340 | } |
d35fca5f | 4341 | |
c664130f | 4342 | /* A well-behaved strxfrm() returns exactly how much space it needs |
1adab0a7 KW |
4343 | * (usually not including the trailing NUL) when it fails due to not |
4344 | * enough space being provided. Assume that this is the case unless | |
4345 | * it's been proven otherwise */ | |
c664130f | 4346 | if (LIKELY(PL_strxfrm_is_behaved) && first_time) { |
55e5378d | 4347 | xAlloc = *xlen + COLLXFRM_HDR_LEN + 1; |
c664130f KW |
4348 | } |
4349 | else { /* Here, either: | |
4350 | * 1) The strxfrm() has previously shown bad behavior; or | |
4351 | * 2) It isn't the first time through the loop, which means | |
4352 | * that the strxfrm() is now showing bad behavior, because | |
4353 | * we gave it what it said was needed in the previous | |
4354 | * iteration, and it came back saying it needed still more. | |
4355 | * (Many versions of cygwin fit this. When the buffer size | |
4356 | * isn't sufficient, they return the input size instead of | |
4357 | * how much is needed.) | |
d4ff9586 KW |
4358 | * Increase the buffer size by a fixed percentage and try again. |
4359 | * */ | |
6ddd902c | 4360 | xAlloc += (xAlloc / 4) + 1; |
c664130f | 4361 | PL_strxfrm_is_behaved = FALSE; |
c664130f | 4362 | |
7d4bcc4a KW |
4363 | # ifdef DEBUGGING |
4364 | ||
58eebef2 KW |
4365 | if (DEBUG_Lv_TEST || debug_initialization) { |
4366 | PerlIO_printf(Perl_debug_log, | |
4367 | "_mem_collxfrm required more space than previously calculated" | |
b07929e4 | 4368 | " for locale %s, trying again with new guess=%d+%zu\n", |
58eebef2 | 4369 | PL_collation_name, (int) COLLXFRM_HDR_LEN, |
b07929e4 | 4370 | xAlloc - COLLXFRM_HDR_LEN); |
58eebef2 | 4371 | } |
7d4bcc4a KW |
4372 | |
4373 | # endif | |
4374 | ||
58eebef2 | 4375 | } |
c664130f | 4376 | |
4ebeff16 | 4377 | Renew(xbuf, xAlloc, char); |
c7202dee KW |
4378 | if (UNLIKELY(! xbuf)) { |
4379 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
4380 | "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc)); | |
4ebeff16 | 4381 | goto bad; |
c7202dee | 4382 | } |
c664130f KW |
4383 | |
4384 | first_time = FALSE; | |
4ebeff16 | 4385 | } |
98994639 | 4386 | |
6696cfa7 | 4387 | |
7d4bcc4a KW |
4388 | # ifdef DEBUGGING |
4389 | ||
58eebef2 | 4390 | if (DEBUG_Lv_TEST || debug_initialization) { |
c7202dee KW |
4391 | |
4392 | print_collxfrm_input_and_return(s, s + len, xlen, utf8); | |
4393 | PerlIO_printf(Perl_debug_log, "Its xfrm is:"); | |
7e2f38b2 KW |
4394 | PerlIO_printf(Perl_debug_log, "%s\n", |
4395 | _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN, | |
4396 | *xlen, 1)); | |
58eebef2 | 4397 | } |
7d4bcc4a KW |
4398 | |
4399 | # endif | |
58eebef2 | 4400 | |
3c5f993e | 4401 | /* Free up unneeded space; retain ehough for trailing NUL */ |
55e5378d | 4402 | Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char); |
98994639 | 4403 | |
6696cfa7 KW |
4404 | if (s != input_string) { |
4405 | Safefree(s); | |
98994639 HS |
4406 | } |
4407 | ||
98994639 HS |
4408 | return xbuf; |
4409 | ||
4410 | bad: | |
7d4bcc4a KW |
4411 | |
4412 | # ifdef DEBUGGING | |
4413 | ||
58eebef2 | 4414 | if (DEBUG_Lv_TEST || debug_initialization) { |
c7202dee | 4415 | print_collxfrm_input_and_return(s, s + len, NULL, utf8); |
58eebef2 | 4416 | } |
7d4bcc4a KW |
4417 | |
4418 | # endif | |
4419 | ||
21dce8f4 KW |
4420 | Safefree(xbuf); |
4421 | if (s != input_string) { | |
4422 | Safefree(s); | |
4423 | } | |
4424 | *xlen = 0; | |
4425 | ||
98994639 HS |
4426 | return NULL; |
4427 | } | |
4428 | ||
7d4bcc4a | 4429 | # ifdef DEBUGGING |
c7202dee | 4430 | |
4cbaac56 | 4431 | STATIC void |
c7202dee KW |
4432 | S_print_collxfrm_input_and_return(pTHX_ |
4433 | const char * const s, | |
4434 | const char * const e, | |
4435 | const STRLEN * const xlen, | |
4436 | const bool is_utf8) | |
4437 | { | |
c7202dee KW |
4438 | |
4439 | PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN; | |
4440 | ||
511e4ff7 DM |
4441 | PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ", |
4442 | (UV)PL_collation_ix); | |
c7202dee | 4443 | if (xlen) { |
08b6dc1d | 4444 | PerlIO_printf(Perl_debug_log, "%zu", *xlen); |
c7202dee KW |
4445 | } |
4446 | else { | |
4447 | PerlIO_printf(Perl_debug_log, "NULL"); | |
4448 | } | |
4449 | PerlIO_printf(Perl_debug_log, " for locale '%s', string='", | |
4450 | PL_collation_name); | |
9c8a6dc2 KW |
4451 | print_bytes_for_locale(s, e, is_utf8); |
4452 | ||
4453 | PerlIO_printf(Perl_debug_log, "'\n"); | |
4454 | } | |
4455 | ||
4d9252fd KW |
4456 | # endif /* DEBUGGING */ |
4457 | #endif /* USE_LOCALE_COLLATE */ | |
1c0e67b5 KW |
4458 | #ifdef USE_LOCALE |
4459 | # ifdef DEBUGGING | |
4d9252fd | 4460 | |
9c8a6dc2 KW |
4461 | STATIC void |
4462 | S_print_bytes_for_locale(pTHX_ | |
4463 | const char * const s, | |
4464 | const char * const e, | |
4465 | const bool is_utf8) | |
4466 | { | |
4467 | const char * t = s; | |
4468 | bool prev_was_printable = TRUE; | |
4469 | bool first_time = TRUE; | |
4470 | ||
4471 | PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE; | |
c7202dee KW |
4472 | |
4473 | while (t < e) { | |
4474 | UV cp = (is_utf8) | |
4475 | ? utf8_to_uvchr_buf((U8 *) t, e, NULL) | |
4476 | : * (U8 *) t; | |
4477 | if (isPRINT(cp)) { | |
4478 | if (! prev_was_printable) { | |
4479 | PerlIO_printf(Perl_debug_log, " "); | |
4480 | } | |
4481 | PerlIO_printf(Perl_debug_log, "%c", (U8) cp); | |
4482 | prev_was_printable = TRUE; | |
4483 | } | |
4484 | else { | |
4485 | if (! first_time) { | |
4486 | PerlIO_printf(Perl_debug_log, " "); | |
4487 | } | |
147e3846 | 4488 | PerlIO_printf(Perl_debug_log, "%02" UVXf, cp); |
c7202dee KW |
4489 | prev_was_printable = FALSE; |
4490 | } | |
4491 | t += (is_utf8) ? UTF8SKIP(t) : 1; | |
4492 | first_time = FALSE; | |
4493 | } | |
c7202dee KW |
4494 | } |
4495 | ||
1c0e67b5 | 4496 | # endif /* #ifdef DEBUGGING */ |
8ef6e574 | 4497 | |
962aa53f KW |
4498 | STATIC const char * |
4499 | S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale) | |
4500 | { | |
4501 | /* Changes the locale for LC_'switch_category" to that of | |
4502 | * LC_'template_category', if they aren't already the same. If not NULL, | |
4503 | * 'template_locale' is the locale that 'template_category' is in. | |
4504 | * | |
9de6fe47 KW |
4505 | * Returns a copy of the name of the original locale for 'switch_category' |
4506 | * so can be switched back to with the companion function | |
4507 | * restore_switched_locale(), (NULL if no restoral is necessary.) */ | |
962aa53f KW |
4508 | |
4509 | char * restore_to_locale = NULL; | |
4510 | ||
4511 | if (switch_category == template_category) { /* No changes needed */ | |
4512 | return NULL; | |
4513 | } | |
4514 | ||
4515 | /* Find the original locale of the category we may need to change, so that | |
4516 | * it can be restored to later */ | |
d6189047 KW |
4517 | restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category, |
4518 | NULL))); | |
962aa53f KW |
4519 | if (! restore_to_locale) { |
4520 | Perl_croak(aTHX_ | |
4521 | "panic: %s: %d: Could not find current %s locale, errno=%d\n", | |
4522 | __FILE__, __LINE__, category_name(switch_category), errno); | |
4523 | } | |
962aa53f KW |
4524 | |
4525 | /* If the locale of the template category wasn't passed in, find it now */ | |
4526 | if (template_locale == NULL) { | |
4527 | template_locale = do_setlocale_r(template_category, NULL); | |
4528 | if (! template_locale) { | |
4529 | Perl_croak(aTHX_ | |
4530 | "panic: %s: %d: Could not find current %s locale, errno=%d\n", | |
4531 | __FILE__, __LINE__, category_name(template_category), errno); | |
4532 | } | |
4533 | } | |
4534 | ||
4535 | /* It the locales are the same, there's nothing to do */ | |
4536 | if (strEQ(restore_to_locale, template_locale)) { | |
4537 | Safefree(restore_to_locale); | |
4538 | ||
4539 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n", | |
4540 | category_name(switch_category), restore_to_locale)); | |
4541 | ||
4542 | return NULL; | |
4543 | } | |
4544 | ||
4545 | /* Finally, change the locale to the template one */ | |
4546 | if (! do_setlocale_r(switch_category, template_locale)) { | |
4547 | Perl_croak(aTHX_ | |
4548 | "panic: %s: %d: Could not change %s locale to %s, errno=%d\n", | |
4549 | __FILE__, __LINE__, category_name(switch_category), | |
4550 | template_locale, errno); | |
4551 | } | |
4552 | ||
4553 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n", | |
4554 | category_name(switch_category), template_locale)); | |
4555 | ||
4556 | return restore_to_locale; | |
4557 | } | |
4558 | ||
4559 | STATIC void | |
4560 | S_restore_switched_locale(pTHX_ const int category, const char * const original_locale) | |
4561 | { | |
9de6fe47 KW |
4562 | /* Restores the locale for LC_'category' to 'original_locale' (which is a |
4563 | * copy that will be freed by this function), or do nothing if the latter | |
4564 | * parameter is NULL */ | |
962aa53f KW |
4565 | |
4566 | if (original_locale == NULL) { | |
4567 | return; | |
4568 | } | |
4569 | ||
4570 | if (! do_setlocale_r(category, original_locale)) { | |
4571 | Perl_croak(aTHX_ | |
4572 | "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n", | |
4573 | __FILE__, __LINE__, | |
4574 | category_name(category), original_locale, errno); | |
4575 | } | |
4576 | ||
4577 | Safefree(original_locale); | |
4578 | } | |
4579 | ||
51332242 N |
4580 | /* is_cur_LC_category_utf8 uses a small char buffer to avoid malloc/free */ |
4581 | #define CUR_LC_BUFFER_SIZE 64 | |
4582 | ||
c1284011 KW |
4583 | bool |
4584 | Perl__is_cur_LC_category_utf8(pTHX_ int category) | |
7d74bb61 KW |
4585 | { |
4586 | /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE | |
4587 | * otherwise. 'category' may not be LC_ALL. If the platform doesn't have | |
119ee68b | 4588 | * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence |
609548d2 KW |
4589 | * could give the wrong result. The result will very likely be correct for |
4590 | * languages that have commonly used non-ASCII characters, but for notably | |
4591 | * English, it comes down to if the locale's name ends in something like | |
9de6fe47 KW |
4592 | * "UTF-8". It errs on the side of not being a UTF-8 locale. |
4593 | * | |
4594 | * If the platform is early C89, not containing mbtowc(), or we are | |
4595 | * compiled to not pay attention to LC_CTYPE, this employs heuristics. | |
4596 | * These work very well for non-Latin locales or those whose currency | |
4597 | * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at | |
4598 | * least MB_CUR_MAX, English locales with an ASCII currency symbol depend | |
4599 | * on the name containing UTF-8 or not. */ | |
7d74bb61 | 4600 | |
47280b20 | 4601 | /* Name of current locale corresponding to the input category */ |
8de4332b | 4602 | const char *save_input_locale = NULL; |
47280b20 KW |
4603 | |
4604 | bool is_utf8 = FALSE; /* The return value */ | |
7d74bb61 | 4605 | |
47280b20 KW |
4606 | /* The variables below are for the cache of previous lookups using this |
4607 | * function. The cache is a C string, described at the definition for | |
4608 | * 'C_and_POSIX_utf8ness'. | |
4609 | * | |
4610 | * The first part of the cache is fixed, for the C and POSIX locales. The | |
4611 | * varying part starts just after them. */ | |
4612 | char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness); | |
4613 | ||
4614 | Size_t utf8ness_cache_size; /* Size of the varying portion */ | |
4615 | Size_t input_name_len; /* Length in bytes of save_input_locale */ | |
4616 | Size_t input_name_len_with_overhead; /* plus extra chars used to store | |
4617 | the name in the cache */ | |
4618 | char * delimited; /* The name plus the delimiters used to store | |
4619 | it in the cache */ | |
51332242 | 4620 | char buffer[CUR_LC_BUFFER_SIZE]; /* small buffer */ |
47280b20 KW |
4621 | char * name_pos; /* position of 'delimited' in the cache, or 0 |
4622 | if not there */ | |
4623 | ||
4624 | ||
7d4bcc4a KW |
4625 | # ifdef LC_ALL |
4626 | ||
7d74bb61 | 4627 | assert(category != LC_ALL); |
7d4bcc4a KW |
4628 | |
4629 | # endif | |
7d74bb61 | 4630 | |
47280b20 | 4631 | /* Get the desired category's locale */ |
d6189047 | 4632 | save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL))); |
7d74bb61 | 4633 | if (! save_input_locale) { |
47280b20 | 4634 | Perl_croak(aTHX_ |
d707d779 KW |
4635 | "panic: %s: %d: Could not find current %s locale, errno=%d\n", |
4636 | __FILE__, __LINE__, category_name(category), errno); | |
7d74bb61 | 4637 | } |
47280b20 | 4638 | |
47280b20 KW |
4639 | DEBUG_L(PerlIO_printf(Perl_debug_log, |
4640 | "Current locale for %s is %s\n", | |
4641 | category_name(category), save_input_locale)); | |
4642 | ||
4643 | input_name_len = strlen(save_input_locale); | |
4644 | ||
4645 | /* In our cache, each name is accompanied by two delimiters and a single | |
4646 | * utf8ness digit */ | |
4647 | input_name_len_with_overhead = input_name_len + 3; | |
4648 | ||
51332242 N |
4649 | if ( input_name_len_with_overhead <= CUR_LC_BUFFER_SIZE ) { |
4650 | /* we can use the buffer, avoid a malloc */ | |
4651 | delimited = buffer; | |
4652 | } else { /* need a malloc */ | |
4653 | /* Allocate and populate space for a copy of the name surrounded by the | |
4654 | * delimiters */ | |
4655 | Newx(delimited, input_name_len_with_overhead, char); | |
4656 | } | |
4657 | ||
47280b20 KW |
4658 | delimited[0] = UTF8NESS_SEP[0]; |
4659 | Copy(save_input_locale, delimited + 1, input_name_len, char); | |
4660 | delimited[input_name_len+1] = UTF8NESS_PREFIX[0]; | |
4661 | delimited[input_name_len+2] = '\0'; | |
4662 | ||
4663 | /* And see if that is in the cache */ | |
4664 | name_pos = instr(PL_locale_utf8ness, delimited); | |
4665 | if (name_pos) { | |
4666 | is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0'; | |
4667 | ||
4668 | # ifdef DEBUGGING | |
4669 | ||
4670 | if (DEBUG_Lv_TEST || debug_initialization) { | |
4671 | PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n", | |
4672 | save_input_locale, is_utf8); | |
4673 | } | |
4674 | ||
4675 | # endif | |
4676 | ||
4677 | /* And, if not already in that position, move it to the beginning of | |
4678 | * the non-constant portion of the list, since it is the most recently | |
4679 | * used. (We don't have to worry about overflow, since just moving | |
4680 | * existing names around) */ | |
4681 | if (name_pos > utf8ness_cache) { | |
4682 | Move(utf8ness_cache, | |
4683 | utf8ness_cache + input_name_len_with_overhead, | |
4684 | name_pos - utf8ness_cache, char); | |
4685 | Copy(delimited, | |
4686 | utf8ness_cache, | |
4687 | input_name_len_with_overhead - 1, char); | |
4688 | utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0'; | |
4689 | } | |
4690 | ||
51332242 N |
4691 | /* free only when not using the buffer */ |
4692 | if ( delimited != buffer ) Safefree(delimited); | |
b07fffd1 | 4693 | Safefree(save_input_locale); |
47280b20 | 4694 | return is_utf8; |
7d74bb61 KW |
4695 | } |
4696 | ||
47280b20 KW |
4697 | /* Here we don't have stored the utf8ness for the input locale. We have to |
4698 | * calculate it */ | |
4699 | ||
94646a69 | 4700 | # if defined(USE_LOCALE_CTYPE) \ |
6201c220 | 4701 | && ( defined(HAS_NL_LANGINFO) \ |
94646a69 | 4702 | || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC))) |
7d74bb61 | 4703 | |
0dec74cd | 4704 | { |
962aa53f KW |
4705 | const char *original_ctype_locale |
4706 | = switch_category_locale_to_template(LC_CTYPE, | |
4707 | category, | |
4708 | save_input_locale); | |
69014004 | 4709 | |
7d74bb61 | 4710 | /* Here the current LC_CTYPE is set to the locale of the category whose |
0dec74cd | 4711 | * information is desired. This means that nl_langinfo() and mbtowc() |
1d958db2 | 4712 | * should give the correct results */ |
119ee68b | 4713 | |
0dec74cd KW |
4714 | # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding |
4715 | calling the functions if we have this */ | |
4716 | ||
4717 | /* Standard UTF-8 needs at least 4 bytes to represent the maximum | |
4718 | * Unicode code point. */ | |
4719 | ||
de3a3072 KW |
4720 | DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n", |
4721 | __FILE__, __LINE__, (int) MB_CUR_MAX)); | |
0dec74cd KW |
4722 | if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) { |
4723 | is_utf8 = FALSE; | |
b5b2847c KW |
4724 | restore_switched_locale(LC_CTYPE, original_ctype_locale); |
4725 | goto finish_and_return; | |
0dec74cd KW |
4726 | } |
4727 | ||
4728 | # endif | |
6201c220 | 4729 | # if defined(HAS_NL_LANGINFO) |
7d4bcc4a | 4730 | |
0dec74cd KW |
4731 | { /* The task is easiest if the platform has this POSIX 2001 function. |
4732 | Except on some platforms it can wrongly return "", so have to have | |
4733 | a fallback. And it can return that it's UTF-8, even if there are | |
4734 | variances from that. For example, Turkish locales may use the | |
4735 | alternate dotted I rules, and sometimes it appears to be a | |
4736 | defective locale definition. XXX We should probably check for | |
9de6fe47 KW |
4737 | these in the Latin1 range and warn (but on glibc, requires |
4738 | iswalnum() etc. due to their not handling 80-FF correctly */ | |
4e6826bf | 4739 | const char *codeset = my_nl_langinfo(CODESET, FALSE); |
c70a3e68 | 4740 | /* FALSE => already in dest locale */ |
119ee68b | 4741 | |
af84e886 | 4742 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
c70a3e68 KW |
4743 | "\tnllanginfo returned CODESET '%s'\n", codeset)); |
4744 | ||
4745 | if (codeset && strNE(codeset, "")) { | |
1d958db2 | 4746 | |
1d075173 KW |
4747 | /* If the implementation of foldEQ() somehow were |
4748 | * to change to not go byte-by-byte, this could | |
4749 | * read past end of string, as only one length is | |
4750 | * checked. But currently, a premature NUL will | |
4751 | * compare false, and it will stop there */ | |
4752 | is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8")) | |
4753 | || foldEQ(codeset, STR_WITH_LEN("UTF8"))); | |
1d958db2 | 4754 | |
69014004 KW |
4755 | DEBUG_L(PerlIO_printf(Perl_debug_log, |
4756 | "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n", | |
4757 | codeset, is_utf8)); | |
b5b2847c KW |
4758 | restore_switched_locale(LC_CTYPE, original_ctype_locale); |
4759 | goto finish_and_return; | |
1d958db2 | 4760 | } |
119ee68b KW |
4761 | } |
4762 | ||
7d4bcc4a | 4763 | # endif |
94646a69 | 4764 | # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC) |
0dec74cd KW |
4765 | /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a |
4766 | * late adder to C89, so very likely to have it. However, testing has | |
4767 | * shown that, like nl_langinfo() above, there are locales that are not | |
4768 | * strictly UTF-8 that this will return that they are */ | |
69014004 | 4769 | |
0dec74cd | 4770 | { |
119ee68b | 4771 | wchar_t wc; |
51fc4b19 | 4772 | int len; |
48015184 | 4773 | dSAVEDERRNO; |
51fc4b19 | 4774 | |
94646a69 KW |
4775 | # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS) |
4776 | ||
4777 | mbstate_t ps; | |
4778 | ||
4779 | # endif | |
4780 | ||
4781 | /* mbrtowc() and mbtowc() convert a byte string to a wide | |
4782 | * character. Feed a byte string to one of them and check that the | |
4783 | * result is the expected Unicode code point */ | |
4784 | ||
4785 | # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS) | |
4786 | /* Prefer this function if available, as it's reentrant */ | |
4787 | ||
4788 | memset(&ps, 0, sizeof(ps));; | |
4789 | PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift | |
4790 | state */ | |
48015184 | 4791 | SETERRNO(0, 0); |
94646a69 | 4792 | len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps); |
48015184 | 4793 | SAVE_ERRNO; |
94646a69 KW |
4794 | |
4795 | # else | |
0dec74cd | 4796 | |
5a91a58b | 4797 | LOCALE_LOCK; |
856b881c | 4798 | PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */ |
48015184 | 4799 | SETERRNO(0, 0); |
b1d4925c | 4800 | len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8)); |
48015184 | 4801 | SAVE_ERRNO; |
5a91a58b | 4802 | LOCALE_UNLOCK; |
51fc4b19 | 4803 | |
94646a69 KW |
4804 | # endif |
4805 | ||
48015184 | 4806 | RESTORE_ERRNO; |
af84e886 | 4807 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
0dec74cd | 4808 | "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n", |
48015184 | 4809 | len, (unsigned int) wc, GET_ERRNO)); |
b1d4925c | 4810 | |
0dec74cd KW |
4811 | is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8) |
4812 | && wc == (wchar_t) UNICODE_REPLACEMENT); | |
119ee68b | 4813 | } |
7d4bcc4a | 4814 | |
17dd77cd FP |
4815 | # endif |
4816 | ||
962aa53f | 4817 | restore_switched_locale(LC_CTYPE, original_ctype_locale); |
47280b20 | 4818 | goto finish_and_return; |
7d74bb61 | 4819 | } |
119ee68b | 4820 | |
0dec74cd | 4821 | # else |
7d74bb61 | 4822 | |
0dec74cd KW |
4823 | /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next |
4824 | * try looking at the currency symbol to see if it disambiguates | |
4825 | * things. Often that will be in the native script, and if the symbol | |
4826 | * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII | |
4827 | * UTF-8, we infer that the locale is too, as the odds of a non-UTF8 | |
4828 | * string being valid UTF-8 are quite small */ | |
fa9b773e | 4829 | |
660ee27b KW |
4830 | # ifdef USE_LOCALE_MONETARY |
4831 | ||
4832 | /* If have LC_MONETARY, we can look at the currency symbol. Often that | |
4833 | * will be in the native script. We do this one first because there is | |
4834 | * just one string to examine, so potentially avoids work */ | |
7d4bcc4a | 4835 | |
9db16864 KW |
4836 | { |
4837 | const char *original_monetary_locale | |
660ee27b KW |
4838 | = switch_category_locale_to_template(LC_MONETARY, |
4839 | category, | |
4840 | save_input_locale); | |
9db16864 | 4841 | bool only_ascii = FALSE; |
660ee27b | 4842 | const U8 * currency_string |
4e6826bf | 4843 | = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE); |
660ee27b KW |
4844 | /* 2nd param not relevant for this item */ |
4845 | const U8 * first_variant; | |
fa9b773e | 4846 | |
660ee27b KW |
4847 | assert( *currency_string == '-' |
4848 | || *currency_string == '+' | |
4849 | || *currency_string == '.'); | |
97f4de96 | 4850 | |
660ee27b | 4851 | currency_string++; |
fa9b773e | 4852 | |
660ee27b | 4853 | if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant)) |
9db16864 KW |
4854 | { |
4855 | 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)); | |
4856 | only_ascii = TRUE; | |
4857 | } | |
4858 | else { | |
660ee27b | 4859 | is_utf8 = is_strict_utf8_string(first_variant, 0); |
9db16864 | 4860 | } |
fa9b773e | 4861 | |
9db16864 | 4862 | restore_switched_locale(LC_MONETARY, original_monetary_locale); |
fa9b773e | 4863 | |
9db16864 | 4864 | if (! only_ascii) { |
fa9b773e | 4865 | |
9db16864 KW |
4866 | /* It isn't a UTF-8 locale if the symbol is not legal UTF-8; |
4867 | * otherwise assume the locale is UTF-8 if and only if the symbol | |
4868 | * is non-ascii UTF-8. */ | |
af84e886 | 4869 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n", |
9db16864 KW |
4870 | save_input_locale, is_utf8)); |
4871 | goto finish_and_return; | |
4872 | } | |
13542a67 | 4873 | } |
fa9b773e | 4874 | |
660ee27b | 4875 | # endif /* USE_LOCALE_MONETARY */ |
7d4bcc4a | 4876 | # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME) |
15f7e74e | 4877 | |
9db16864 KW |
4878 | /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try |
4879 | * the names of the months and weekdays, timezone, and am/pm indicator */ | |
4880 | { | |
4881 | const char *original_time_locale | |
4882 | = switch_category_locale_to_template(LC_TIME, | |
4883 | category, | |
4884 | save_input_locale); | |
4885 | int hour = 10; | |
4886 | bool is_dst = FALSE; | |
4887 | int dom = 1; | |
4888 | int month = 0; | |
4889 | int i; | |
4890 | char * formatted_time; | |
4891 | ||
4892 | /* Here the current LC_TIME is set to the locale of the category | |
4893 | * whose information is desired. Look at all the days of the week and | |
4894 | * month names, and the timezone and am/pm indicator for UTF-8 variant | |
4895 | * characters. The first such a one found will tell us if the locale | |
4896 | * is UTF-8 or not */ | |
4897 | ||
4898 | for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */ | |
4899 | formatted_time = my_strftime("%A %B %Z %p", | |
4900 | 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst); | |
4901 | if ( ! formatted_time | |
4902 | || is_utf8_invariant_string((U8 *) formatted_time, 0)) | |
4903 | { | |
15f7e74e | 4904 | |
9db16864 KW |
4905 | /* Here, we didn't find a non-ASCII. Try the next time through |
4906 | * with the complemented dst and am/pm, and try with the next | |
4907 | * weekday. After we have gotten all weekdays, try the next | |
4908 | * month */ | |
4909 | is_dst = ! is_dst; | |
4910 | hour = (hour + 12) % 24; | |
4911 | dom++; | |
4912 | if (i > 6) { | |
4913 | month++; | |
4914 | } | |
4915 | continue; | |
15f7e74e | 4916 | } |
9db16864 KW |
4917 | |
4918 | /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8; | |
4919 | * false otherwise. But first, restore LC_TIME to its original | |
4920 | * locale if we changed it */ | |
4921 | restore_switched_locale(LC_TIME, original_time_locale); | |
4922 | ||
af84e886 | 4923 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n", |
9db16864 KW |
4924 | save_input_locale, |
4925 | is_utf8_string((U8 *) formatted_time, 0))); | |
4926 | is_utf8 = is_utf8_string((U8 *) formatted_time, 0); | |
4927 | goto finish_and_return; | |
15f7e74e KW |
4928 | } |
4929 | ||
9db16864 KW |
4930 | /* Falling off the end of the loop indicates all the names were just |
4931 | * ASCII. Go on to the next test. If we changed it, restore LC_TIME | |
4932 | * to its original locale */ | |
962aa53f | 4933 | restore_switched_locale(LC_TIME, original_time_locale); |
af84e886 | 4934 | 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)); |
15f7e74e KW |
4935 | } |
4936 | ||
7d4bcc4a | 4937 | # endif |
15f7e74e | 4938 | |
7d4bcc4a | 4939 | # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST) |
855aeb93 | 4940 | |
9db16864 KW |
4941 | /* This code is ifdefd out because it was found to not be necessary in testing |
4942 | * on our dromedary test machine, which has over 700 locales. There, this | |
4943 | * added no value to looking at the currency symbol and the time strings. I | |
4944 | * left it in so as to avoid rewriting it if real-world experience indicates | |
4945 | * that dromedary is an outlier. Essentially, instead of returning abpve if we | |
4946 | * haven't found illegal utf8, we continue on and examine all the strerror() | |
4947 | * messages on the platform for utf8ness. If all are ASCII, we still don't | |
4948 | * know the answer; but otherwise we have a pretty good indication of the | |
4949 | * utf8ness. The reason this doesn't help much is that the messages may not | |
4950 | * have been translated into the locale. The currency symbol and time strings | |
4951 | * are much more likely to have been translated. */ | |
4952 | { | |
4953 | int e; | |
4954 | bool non_ascii = FALSE; | |
4955 | const char *original_messages_locale | |
4956 | = switch_category_locale_to_template(LC_MESSAGES, | |
4957 | category, | |
4958 | save_input_locale); | |
4959 | const char * errmsg = NULL; | |
4960 | ||
4961 | /* Here the current LC_MESSAGES is set to the locale of the category | |
4962 | * whose information is desired. Look through all the messages. We | |
4963 | * can't use Strerror() here because it may expand to code that | |
4964 | * segfaults in miniperl */ | |
4965 | ||
4966 | for (e = 0; e <= sys_nerr; e++) { | |
4967 | errno = 0; | |
4968 | errmsg = sys_errlist[e]; | |
4969 | if (errno || !errmsg) { | |
4970 | break; | |
4971 | } | |
4972 | errmsg = savepv(errmsg); | |
4973 | if (! is_utf8_invariant_string((U8 *) errmsg, 0)) { | |
4974 | non_ascii = TRUE; | |
4975 | is_utf8 = is_utf8_string((U8 *) errmsg, 0); | |
4976 | break; | |
4977 | } | |
855aeb93 | 4978 | } |
9db16864 | 4979 | Safefree(errmsg); |
855aeb93 | 4980 | |
9db16864 | 4981 | restore_switched_locale(LC_MESSAGES, original_messages_locale); |
855aeb93 | 4982 | |
9db16864 | 4983 | if (non_ascii) { |
5857e934 | 4984 | |
9db16864 KW |
4985 | /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid, |
4986 | * any non-ascii means it is one; otherwise we assume it isn't */ | |
af84e886 | 4987 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n", |
9db16864 KW |
4988 | save_input_locale, |
4989 | is_utf8)); | |
4990 | goto finish_and_return; | |
4991 | } | |
855aeb93 | 4992 | |
9db16864 KW |
4993 | 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)); |
4994 | } | |
855aeb93 | 4995 | |
7d4bcc4a | 4996 | # endif |
0dec74cd | 4997 | # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a |
92c0a900 | 4998 | UTF-8 locale */ |
7d4bcc4a | 4999 | |
97f4de96 KW |
5000 | /* As a last resort, look at the locale name to see if it matches |
5001 | * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the | |
5002 | * return of setlocale(), is actually defined to be opaque, so we can't | |
5003 | * really rely on the absence of various substrings in the name to indicate | |
5004 | * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to | |
5005 | * be a UTF-8 locale. Similarly for the other common names */ | |
5006 | ||
0dec74cd | 5007 | { |
c36d8df8 | 5008 | const Size_t final_pos = strlen(save_input_locale) - 1; |
0dec74cd | 5009 | |
c36d8df8 KW |
5010 | if (final_pos >= 3) { |
5011 | const char *name = save_input_locale; | |
97f4de96 | 5012 | |
c36d8df8 KW |
5013 | /* Find next 'U' or 'u' and look from there */ |
5014 | while ((name += strcspn(name, "Uu") + 1) | |
5015 | <= save_input_locale + final_pos - 2) | |
97f4de96 | 5016 | { |
c36d8df8 KW |
5017 | if ( isALPHA_FOLD_NE(*name, 't') |
5018 | || isALPHA_FOLD_NE(*(name + 1), 'f')) | |
5019 | { | |
5020 | continue; | |
5021 | } | |
5022 | name += 2; | |
5023 | if (*(name) == '-') { | |
5024 | if ((name > save_input_locale + final_pos - 1)) { | |
5025 | break; | |
5026 | } | |
5027 | name++; | |
5028 | } | |
5029 | if (*(name) == '8') { | |
5030 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
5031 | "Locale %s ends with UTF-8 in name\n", | |
5032 | save_input_locale)); | |
5033 | is_utf8 = TRUE; | |
5034 | goto finish_and_return; | |
97f4de96 | 5035 | } |
97f4de96 | 5036 | } |
c36d8df8 KW |
5037 | DEBUG_L(PerlIO_printf(Perl_debug_log, |
5038 | "Locale %s doesn't end with UTF-8 in name\n", | |
5039 | save_input_locale)); | |
97f4de96 | 5040 | } |
97f4de96 | 5041 | |
4d8d465a | 5042 | # ifdef WIN32 |
7d4bcc4a | 5043 | |
c36d8df8 KW |
5044 | /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */ |
5045 | if (memENDs(save_input_locale, final_pos, "65001")) { | |
5046 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
8a0832a1 | 5047 | "Locale %s ends with 65001 in name, is UTF-8 locale\n", |
97f4de96 | 5048 | save_input_locale)); |
c36d8df8 KW |
5049 | is_utf8 = TRUE; |
5050 | goto finish_and_return; | |
5051 | } | |
7d4bcc4a | 5052 | |
4d8d465a | 5053 | # endif |
17dd77cd | 5054 | } |
4d8d465a | 5055 | # endif |
97f4de96 KW |
5056 | |
5057 | /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But | |
5058 | * since we are about to return FALSE anyway, there is no point in doing | |
5059 | * this extra work */ | |
7d4bcc4a | 5060 | |
0dec74cd | 5061 | # if 0 |
97f4de96 KW |
5062 | if (instr(save_input_locale, "8859")) { |
5063 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
5064 | "Locale %s has 8859 in name, not UTF-8 locale\n", | |
5065 | save_input_locale)); | |
47280b20 KW |
5066 | is_utf8 = FALSE; |
5067 | goto finish_and_return; | |
97f4de96 | 5068 | } |
0dec74cd | 5069 | # endif |
97f4de96 | 5070 | |
69014004 KW |
5071 | DEBUG_L(PerlIO_printf(Perl_debug_log, |
5072 | "Assuming locale %s is not a UTF-8 locale\n", | |
5073 | save_input_locale)); | |
47280b20 KW |
5074 | is_utf8 = FALSE; |
5075 | ||
0dec74cd KW |
5076 | # endif /* the code that is compiled when no modern LC_CTYPE */ |
5077 | ||
47280b20 KW |
5078 | finish_and_return: |
5079 | ||
5080 | /* Cache this result so we don't have to go through all this next time. */ | |
5081 | utf8ness_cache_size = sizeof(PL_locale_utf8ness) | |
5082 | - (utf8ness_cache - PL_locale_utf8ness); | |
5083 | ||
5084 | /* But we can't save it if it is too large for the total space available */ | |
5085 | if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) { | |
5086 | Size_t utf8ness_cache_len = strlen(utf8ness_cache); | |
5087 | ||
5088 | /* Here it can fit, but we may need to clear out the oldest cached | |
5089 | * result(s) to do so. Check */ | |
5090 | if (utf8ness_cache_len + input_name_len_with_overhead | |
5091 | >= utf8ness_cache_size) | |
5092 | { | |
5093 | /* Here we have to clear something out to make room for this. | |
5094 | * Start looking at the rightmost place where it could fit and find | |
5095 | * the beginning of the entry that extends past that. */ | |
5096 | char * cutoff = (char *) my_memrchr(utf8ness_cache, | |
5097 | UTF8NESS_SEP[0], | |
5098 | utf8ness_cache_size | |
5099 | - input_name_len_with_overhead); | |
5100 | ||
5101 | assert(cutoff); | |
5102 | assert(cutoff >= utf8ness_cache); | |
5103 | ||
5104 | /* This and all subsequent entries must be removed */ | |
5105 | *cutoff = '\0'; | |
5106 | utf8ness_cache_len = strlen(utf8ness_cache); | |
5107 | } | |
5108 | ||
5109 | /* Make space for the new entry */ | |
5110 | Move(utf8ness_cache, | |
5111 | utf8ness_cache + input_name_len_with_overhead, | |
5112 | utf8ness_cache_len + 1 /* Incl. trailing NUL */, char); | |
5113 | ||
5114 | /* And insert it */ | |
5115 | Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char); | |
5116 | utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0'; | |
5117 | ||
72299e00 | 5118 | if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1] & ~1) != '0') { |
47280b20 | 5119 | Perl_croak(aTHX_ |
7fcf9272 KW |
5120 | "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu," |
5121 | " inserted_name=%s, its_len=%zu\n", | |
47280b20 KW |
5122 | __FILE__, __LINE__, |
5123 | PL_locale_utf8ness, strlen(PL_locale_utf8ness), | |
5124 | delimited, input_name_len_with_overhead); | |
5125 | } | |
5126 | } | |
5127 | ||
5128 | # ifdef DEBUGGING | |
5129 | ||
de3a3072 KW |
5130 | if (DEBUG_Lv_TEST) { |
5131 | const char * s = PL_locale_utf8ness; | |
5132 | ||
5133 | /* Audit the structure */ | |
5134 | while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) { | |
5135 | const char *e; | |
5136 | ||
5137 | if (*s != UTF8NESS_SEP[0]) { | |
5138 | Perl_croak(aTHX_ | |
5139 | "panic: %s: %d: Corrupt utf8ness_cache: missing" | |
5140 | " separator %.*s<-- HERE %s\n", | |
5141 | __FILE__, __LINE__, | |
fe301c93 | 5142 | (int) (s - PL_locale_utf8ness), PL_locale_utf8ness, |
de3a3072 KW |
5143 | s); |
5144 | } | |
5145 | s++; | |
5146 | e = strchr(s, UTF8NESS_PREFIX[0]); | |
5147 | if (! e) { | |
01b91050 | 5148 | e = PL_locale_utf8ness + strlen(PL_locale_utf8ness); |
de3a3072 KW |
5149 | Perl_croak(aTHX_ |
5150 | "panic: %s: %d: Corrupt utf8ness_cache: missing" | |
5151 | " separator %.*s<-- HERE %s\n", | |
5152 | __FILE__, __LINE__, | |
fe301c93 | 5153 | (int) (e - PL_locale_utf8ness), PL_locale_utf8ness, |
de3a3072 KW |
5154 | e); |
5155 | } | |
5156 | e++; | |
5157 | if (*e != '0' && *e != '1') { | |
5158 | Perl_croak(aTHX_ | |
5159 | "panic: %s: %d: Corrupt utf8ness_cache: utf8ness" | |
5160 | " must be [01] %.*s<-- HERE %s\n", | |
5161 | __FILE__, __LINE__, | |
fe301c93 KW |
5162 | (int) (e + 1 - PL_locale_utf8ness), |
5163 | PL_locale_utf8ness, e + 1); | |
de3a3072 KW |
5164 | } |
5165 | if (ninstr(PL_locale_utf8ness, s, s-1, e)) { | |
5166 | Perl_croak(aTHX_ | |
5167 | "panic: %s: %d: Corrupt utf8ness_cache: entry" | |
5168 | " has duplicate %.*s<-- HERE %s\n", | |
5169 | __FILE__, __LINE__, | |
fe301c93 | 5170 | (int) (e - PL_locale_utf8ness), PL_locale_utf8ness, |
de3a3072 KW |
5171 | e); |
5172 | } | |
5173 | s = e + 1; | |
5174 | } | |
5175 | } | |
5176 | ||
47280b20 | 5177 | if (DEBUG_Lv_TEST || debug_initialization) { |
de3a3072 | 5178 | |
47280b20 KW |
5179 | PerlIO_printf(Perl_debug_log, |
5180 | "PL_locale_utf8ness is now %s; returning %d\n", | |
5181 | PL_locale_utf8ness, is_utf8); | |
5182 | } | |
5183 | ||
5184 | # endif | |
5185 | ||
51332242 N |
5186 | /* free only when not using the buffer */ |
5187 | if ( delimited != buffer ) Safefree(delimited); | |
fa9b773e | 5188 | Safefree(save_input_locale); |
47280b20 | 5189 | return is_utf8; |
7d74bb61 KW |
5190 | } |
5191 | ||
8ef6e574 | 5192 | #endif |
7d74bb61 | 5193 | |
d6ded950 KW |
5194 | bool |
5195 | Perl__is_in_locale_category(pTHX_ const bool compiling, const int category) | |
5196 | { | |
1a4f13e1 | 5197 | dVAR; |
d6ded950 KW |
5198 | /* Internal function which returns if we are in the scope of a pragma that |
5199 | * enables the locale category 'category'. 'compiling' should indicate if | |
5200 | * this is during the compilation phase (TRUE) or not (FALSE). */ | |
5201 | ||
5202 | const COP * const cop = (compiling) ? &PL_compiling : PL_curcop; | |
5203 | ||
363d7d4a JK |
5204 | SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0); |
5205 | if (! these_categories || these_categories == &PL_sv_placeholder) { | |
d6ded950 KW |
5206 | return FALSE; |
5207 | } | |
5208 | ||
5209 | /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get | |
5210 | * a valid unsigned */ | |
5211 | assert(category >= -1); | |
363d7d4a | 5212 | return cBOOL(SvUV(these_categories) & (1U << (category + 1))); |
d6ded950 KW |
5213 | } |
5214 | ||
2c6ee1a7 | 5215 | char * |
6ebbc862 KW |
5216 | Perl_my_strerror(pTHX_ const int errnum) |
5217 | { | |
5218 | /* Returns a mortalized copy of the text of the error message associated | |
5219 | * with 'errnum'. It uses the current locale's text unless the platform | |
5220 | * doesn't have the LC_MESSAGES category or we are not being called from | |
5221 | * within the scope of 'use locale'. In the former case, it uses whatever | |
5222 | * strerror returns; in the latter case it uses the text from the C locale. | |
5223 | * | |
5224 | * The function just calls strerror(), but temporarily switches, if needed, | |
5225 | * to the C locale */ | |
5226 | ||
5227 | char *errstr; | |
52770946 | 5228 | dVAR; |
6ebbc862 | 5229 | |
52770946 | 5230 | #ifndef USE_LOCALE_MESSAGES |
6ebbc862 | 5231 | |
52770946 KW |
5232 | /* If platform doesn't have messages category, we don't do any switching to |
5233 | * the C locale; we just use whatever strerror() returns */ | |
5234 | ||
5235 | errstr = savepv(Strerror(errnum)); | |
5236 | ||
5237 | #else /* Has locale messages */ | |
5238 | ||
5239 | const bool within_locale_scope = IN_LC(LC_MESSAGES); | |
2c6ee1a7 | 5240 | |
39e69e77 | 5241 | # ifndef USE_ITHREADS |
7aaa36b1 | 5242 | |
39e69e77 KW |
5243 | /* This function is trivial without threads. */ |
5244 | if (within_locale_scope) { | |
5245 | errstr = savepv(strerror(errnum)); | |
5246 | } | |
5247 | else { | |
b0bde642 | 5248 | const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL)); |
39e69e77 KW |
5249 | |
5250 | do_setlocale_c(LC_MESSAGES, "C"); | |
5251 | errstr = savepv(strerror(errnum)); | |
5252 | do_setlocale_c(LC_MESSAGES, save_locale); | |
b0bde642 | 5253 | Safefree(save_locale); |
39e69e77 KW |
5254 | } |
5255 | ||
80306bb6 KW |
5256 | # elif defined(HAS_POSIX_2008_LOCALE) \ |
5257 | && defined(HAS_STRERROR_L) \ | |
5258 | && defined(HAS_DUPLOCALE) | |
39e69e77 KW |
5259 | |
5260 | /* This function is also trivial if we don't have to worry about thread | |
5261 | * safety and have strerror_l(), as it handles the switch of locales so we | |
5262 | * don't have to deal with that. We don't have to worry about thread | |
5263 | * safety if strerror_r() is also available. Both it and strerror_l() are | |
5264 | * thread-safe. Plain strerror() isn't thread safe. But on threaded | |
5265 | * builds when strerror_r() is available, the apparent call to strerror() | |
5266 | * below is actually a macro that behind-the-scenes calls strerror_r(). */ | |
43cb6651 | 5267 | |
39e69e77 | 5268 | # ifdef HAS_STRERROR_R |
7aaa36b1 KW |
5269 | |
5270 | if (within_locale_scope) { | |
4eb27fc5 | 5271 | errstr = savepv(strerror(errnum)); |
7aaa36b1 KW |
5272 | } |
5273 | else { | |
4eb27fc5 | 5274 | errstr = savepv(strerror_l(errnum, PL_C_locale_obj)); |
7aaa36b1 KW |
5275 | } |
5276 | ||
43cb6651 KW |
5277 | # else |
5278 | ||
5279 | /* Here we have strerror_l(), but not strerror_r() and we are on a | |
5280 | * threaded-build. We use strerror_l() for everything, constructing a | |
5281 | * locale to pass to it if necessary */ | |
5282 | ||
5283 | bool do_free = FALSE; | |
5284 | locale_t locale_to_use; | |
5285 | ||
5286 | if (within_locale_scope) { | |
5287 | locale_to_use = uselocale((locale_t) 0); | |
5288 | if (locale_to_use == LC_GLOBAL_LOCALE) { | |
5289 | locale_to_use = duplocale(LC_GLOBAL_LOCALE); | |
5290 | do_free = TRUE; | |
5291 | } | |
5292 | } | |
5293 | else { /* Use C locale if not within 'use locale' scope */ | |
5294 | locale_to_use = PL_C_locale_obj; | |
5295 | } | |
5296 | ||
5297 | errstr = savepv(strerror_l(errnum, locale_to_use)); | |
5298 | ||
5299 | if (do_free) { | |
5300 | freelocale(locale_to_use); | |
5301 | } | |
5302 | ||
5303 | # endif | |
5304 | # else /* Doesn't have strerror_l() */ | |
7aaa36b1 | 5305 | |
8de4332b | 5306 | const char * save_locale = NULL; |
c9dda6da | 5307 | bool locale_is_C = FALSE; |
2c6ee1a7 | 5308 | |
9de6fe47 | 5309 | /* We have a critical section to prevent another thread from executing this |
e9bc6d6b | 5310 | * same code at the same time. (On thread-safe perls, the LOCK is a |
9de6fe47 KW |
5311 | * no-op.) Since this is the only place in core that changes LC_MESSAGES |
5312 | * (unless the user has called setlocale(), this works to prevent races. */ | |
6ebbc862 KW |
5313 | LOCALE_LOCK; |
5314 | ||
9c8a6dc2 KW |
5315 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
5316 | "my_strerror called with errnum %d\n", errnum)); | |
6ebbc862 | 5317 | if (! within_locale_scope) { |
837ce802 | 5318 | save_locale = do_setlocale_c(LC_MESSAGES, NULL); |
c9dda6da | 5319 | if (! save_locale) { |
d707d779 KW |
5320 | Perl_croak(aTHX_ |
5321 | "panic: %s: %d: Could not find current LC_MESSAGES locale," | |
5322 | " errno=%d\n", __FILE__, __LINE__, errno); | |
c9dda6da KW |
5323 | } |
5324 | else { | |
5325 | locale_is_C = isNAME_C_OR_POSIX(save_locale); | |
2c6ee1a7 | 5326 | |
c9dda6da KW |
5327 | /* Switch to the C locale if not already in it */ |
5328 | if (! locale_is_C) { | |
2c6ee1a7 | 5329 | |
c9dda6da KW |
5330 | /* The setlocale() just below likely will zap 'save_locale', so |
5331 | * create a copy. */ | |
5332 | save_locale = savepv(save_locale); | |
837ce802 | 5333 | do_setlocale_c(LC_MESSAGES, "C"); |
c9dda6da | 5334 | } |
6ebbc862 | 5335 | } |
6ebbc862 | 5336 | } /* end of ! within_locale_scope */ |
9c8a6dc2 KW |
5337 | else { |
5338 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n", | |
5339 | __FILE__, __LINE__)); | |
5340 | } | |
a0b53297 | 5341 | |
9c8a6dc2 KW |
5342 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
5343 | "Any locale change has been done; about to call Strerror\n")); | |
52770946 | 5344 | errstr = savepv(Strerror(errnum)); |
6ebbc862 KW |
5345 | |
5346 | if (! within_locale_scope) { | |
c9dda6da | 5347 | if (save_locale && ! locale_is_C) { |
837ce802 | 5348 | if (! do_setlocale_c(LC_MESSAGES, save_locale)) { |
d707d779 KW |
5349 | Perl_croak(aTHX_ |
5350 | "panic: %s: %d: setlocale restore failed, errno=%d\n", | |
5351 | __FILE__, __LINE__, errno); | |
c9dda6da | 5352 | } |
6ebbc862 KW |
5353 | Safefree(save_locale); |
5354 | } | |
5355 | } | |
5356 | ||
5357 | LOCALE_UNLOCK; | |
5358 | ||
7aaa36b1 | 5359 | # endif /* End of doesn't have strerror_l */ |
d36adde0 | 5360 | # ifdef DEBUGGING |
6affbbf0 KW |
5361 | |
5362 | if (DEBUG_Lv_TEST) { | |
5363 | PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '"); | |
5364 | print_bytes_for_locale(errstr, errstr + strlen(errstr), 0); | |
5365 | PerlIO_printf(Perl_debug_log, "'\n"); | |
5366 | } | |
5367 | ||
d36adde0 KW |
5368 | # endif |
5369 | #endif /* End of does have locale messages */ | |
2c6ee1a7 | 5370 | |
52770946 | 5371 | SAVEFREEPV(errstr); |
6ebbc862 | 5372 | return errstr; |
2c6ee1a7 KW |
5373 | } |
5374 | ||
66610fdd | 5375 | /* |
747c467a | 5376 | |
58e641fb KW |
5377 | =for apidoc switch_to_global_locale |
5378 | ||
75920755 | 5379 | On systems without locale support, or on typical single-threaded builds, or on |
58e641fb KW |
5380 | platforms that do not support per-thread locale operations, this function does |
5381 | nothing. On such systems that do have locale support, only a locale global to | |
5382 | the whole program is available. | |
5383 | ||
5384 | On multi-threaded builds on systems that do have per-thread locale operations, | |
5385 | this function converts the thread it is running in to use the global locale. | |
5386 | This is for code that has not yet or cannot be updated to handle multi-threaded | |
5387 | locale operation. As long as only a single thread is so-converted, everything | |
5388 | works fine, as all the other threads continue to ignore the global one, so only | |
5389 | this thread looks at it. | |
5390 | ||
69c5e0db KW |
5391 | However, on Windows systems this isn't quite true prior to Visual Studio 15, |
5392 | at which point Microsoft fixed a bug. A race can occur if you use the | |
5393 | following operations on earlier Windows platforms: | |
5394 | ||
5395 | =over | |
5396 | ||
5397 | =item L<POSIX::localeconv|POSIX/localeconv> | |
5398 | ||
5399 | =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP> | |
5400 | ||
5401 | =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP> | |
5402 | ||
5403 | =back | |
5404 | ||
5405 | The first item is not fixable (except by upgrading to a later Visual Studio | |
5406 | release), but it would be possible to work around the latter two items by using | |
5407 | the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches | |
5408 | welcome. | |
5409 | ||
58e641fb KW |
5410 | Without this function call, threads that use the L<C<setlocale(3)>> system |
5411 | function will not work properly, as all the locale-sensitive functions will | |
5412 | look at the per-thread locale, and C<setlocale> will have no effect on this | |
5413 | thread. | |
5414 | ||
5415 | Perl code should convert to either call | |
5416 | L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system | |
5417 | C<setlocale>) or use the methods given in L<perlcall> to call | |
5418 | L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly | |
5419 | handle all cases of single- vs multi-thread, POSIX 2008-supported or not. | |
5420 | ||
5421 | Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can | |
5422 | continue to work if this function is called before transferring control to the | |
5423 | library. | |
5424 | ||
5425 | Upon return from the code that needs to use the global locale, | |
5426 | L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe | |
5427 | multi-thread operation. | |
5428 | ||
5429 | =cut | |
5430 | */ | |
5431 | ||
5432 | void | |
5433 | Perl_switch_to_global_locale() | |
5434 | { | |
5435 | ||
5436 | #ifdef USE_THREAD_SAFE_LOCALE | |
5437 | # ifdef WIN32 | |
5438 | ||
5439 | _configthreadlocale(_DISABLE_PER_THREAD_LOCALE); | |
5440 | ||
5441 | # else | |
5442 | # ifdef HAS_QUERYLOCALE | |
5443 | ||
5444 | setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0))); | |
5445 | ||
5446 | # else | |
5447 | ||
5448 | { | |
5449 | unsigned int i; | |
5450 | ||
5451 | for (i = 0; i < LC_ALL_INDEX; i++) { | |
5452 | setlocale(categories[i], do_setlocale_r(categories[i], NULL)); | |
5453 | } | |
5454 | } | |
5455 | ||
5456 | # endif | |
5457 | ||
5458 | uselocale(LC_GLOBAL_LOCALE); | |
5459 | ||
5460 | # endif | |
5461 | #endif | |
5462 | ||
5463 | } | |
5464 | ||
5465 | /* | |
5466 | ||
747c467a KW |
5467 | =for apidoc sync_locale |
5468 | ||
b2e9ba0c KW |
5469 | L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or |
5470 | change the locale (though changing the locale is antisocial and dangerous on | |
5471 | multi-threaded systems that don't have multi-thread safe locale operations. | |
5472 | (See L<perllocale/Multi-threaded operation>). Using the system | |
5473 | L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries | |
5474 | called from XS, such as C<Gtk> do so, and this can't be changed. When the | |
5475 | locale is changed by XS code that didn't use | |
5476 | L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the | |
5477 | locale has changed. Use this function to do so, before returning to Perl. | |
5478 | ||
5479 | The return value is a boolean: TRUE if the global locale at the time of call | |
5480 | was in effect; and FALSE if a per-thread locale was in effect. This can be | |
5481 | used by the caller that needs to restore things as-they-were to decide whether | |
5482 | or not to call | |
5483 | L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>. | |
747c467a KW |
5484 | |
5485 | =cut | |
5486 | */ | |
5487 | ||
b2e9ba0c KW |
5488 | bool |
5489 | Perl_sync_locale() | |
747c467a | 5490 | { |
d36adde0 KW |
5491 | |
5492 | #ifndef USE_LOCALE | |
5493 | ||
5494 | return TRUE; | |
5495 | ||
5496 | #else | |
5497 | ||
e9bc6d6b | 5498 | const char * newlocale; |
b2e9ba0c KW |
5499 | dTHX; |
5500 | ||
d36adde0 | 5501 | # ifdef USE_POSIX_2008_LOCALE |
b2e9ba0c KW |
5502 | |
5503 | bool was_in_global_locale = FALSE; | |
5504 | locale_t cur_obj = uselocale((locale_t) 0); | |
5505 | ||
5506 | /* On Windows, unless the foreign code has turned off the thread-safe | |
5507 | * locale setting, any plain setlocale() will have affected what we see, so | |
5508 | * no need to worry. Otherwise, If the foreign code has done a plain | |
5509 | * setlocale(), it will only affect the global locale on POSIX systems, but | |
5510 | * will affect the */ | |
5511 | if (cur_obj == LC_GLOBAL_LOCALE) { | |
5512 | ||
d36adde0 | 5513 | # ifdef HAS_QUERY_LOCALE |
b2e9ba0c KW |
5514 | |
5515 | do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL)); | |
5516 | ||
d36adde0 | 5517 | # else |
b2e9ba0c KW |
5518 | |
5519 | unsigned int i; | |
5520 | ||
5521 | /* We can't trust that we can read the LC_ALL format on the | |
5522 | * platform, so do them individually */ | |
5523 | for (i = 0; i < LC_ALL_INDEX; i++) { | |
5524 | do_setlocale_r(categories[i], setlocale(categories[i], NULL)); | |
5525 | } | |
747c467a | 5526 | |
d36adde0 | 5527 | # endif |
b2e9ba0c KW |
5528 | |
5529 | was_in_global_locale = TRUE; | |
5530 | } | |
5531 | ||
d36adde0 | 5532 | # else |
b2e9ba0c KW |
5533 | |
5534 | bool was_in_global_locale = TRUE; | |
5535 | ||
d36adde0 KW |
5536 | # endif |
5537 | # ifdef USE_LOCALE_CTYPE | |
7d4bcc4a | 5538 | |
b0bde642 | 5539 | newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL)); |
9f82ea3e KW |
5540 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
5541 | "%s:%d: %s\n", __FILE__, __LINE__, | |
5542 | setlocale_debug_string(LC_CTYPE, NULL, newlocale))); | |
5543 | new_ctype(newlocale); | |
b0bde642 | 5544 | Safefree(newlocale); |
747c467a | 5545 | |
d36adde0 KW |
5546 | # endif /* USE_LOCALE_CTYPE */ |
5547 | # ifdef USE_LOCALE_COLLATE | |
7d4bcc4a | 5548 | |
b0bde642 | 5549 | newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL)); |
9f82ea3e KW |
5550 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
5551 | "%s:%d: %s\n", __FILE__, __LINE__, | |
5552 | setlocale_debug_string(LC_COLLATE, NULL, newlocale))); | |
5553 | new_collate(newlocale); | |
b0bde642 | 5554 | Safefree(newlocale); |
747c467a | 5555 | |
d36adde0 KW |
5556 | # endif |
5557 | # ifdef USE_LOCALE_NUMERIC | |
7d4bcc4a | 5558 | |
b0bde642 | 5559 | newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL)); |
9f82ea3e KW |
5560 | DEBUG_Lv(PerlIO_printf(Perl_debug_log, |
5561 | "%s:%d: %s\n", __FILE__, __LINE__, | |
5562 | setlocale_debug_string(LC_NUMERIC, NULL, newlocale))); | |
5563 | new_numeric(newlocale); | |
b0bde642 | 5564 | Safefree(newlocale); |
7d4bcc4a | 5565 | |
d36adde0 | 5566 | # endif /* USE_LOCALE_NUMERIC */ |
747c467a | 5567 | |
b2e9ba0c | 5568 | return was_in_global_locale; |
d36adde0 KW |
5569 | |
5570 | #endif | |
5571 | ||
747c467a KW |
5572 | } |
5573 | ||
5d1187d1 KW |
5574 | #if defined(DEBUGGING) && defined(USE_LOCALE) |
5575 | ||
a4f00dcc KW |
5576 | STATIC char * |
5577 | S_setlocale_debug_string(const int category, /* category number, | |
5d1187d1 KW |
5578 | like LC_ALL */ |
5579 | const char* const locale, /* locale name */ | |
5580 | ||
5581 | /* return value from setlocale() when attempting to | |
5582 | * set 'category' to 'locale' */ | |
5583 | const char* const retval) | |
5584 | { | |
5585 | /* Returns a pointer to a NUL-terminated string in static storage with | |
5586 | * added text about the info passed in. This is not thread safe and will | |
5587 | * be overwritten by the next call, so this should be used just to | |
fa07b8e5 | 5588 | * formulate a string to immediately print or savepv() on. */ |
5d1187d1 | 5589 | |
398a990f DM |
5590 | /* initialise to a non-null value to keep it out of BSS and so keep |
5591 | * -DPERL_GLOBAL_STRUCT_PRIVATE happy */ | |
8773126a | 5592 | static char ret[256] = "If you can read this, thank your buggy C" |
60b45a7d KW |
5593 | " library strlcpy(), and change your hints file" |
5594 | " to undef it"; | |
7d4bcc4a | 5595 | |
e5f10d49 | 5596 | my_strlcpy(ret, "setlocale(", sizeof(ret)); |
b09aaf40 | 5597 | my_strlcat(ret, category_name(category), sizeof(ret)); |
fa07b8e5 | 5598 | my_strlcat(ret, ", ", sizeof(ret)); |
5d1187d1 KW |
5599 | |
5600 | if (locale) { | |
fa07b8e5 KW |
5601 | my_strlcat(ret, "\"", sizeof(ret)); |
5602 | my_strlcat(ret, locale, sizeof(ret)); | |
5603 | my_strlcat(ret, "\"", sizeof(ret)); | |
5d1187d1 KW |
5604 | } |
5605 | else { | |
fa07b8e5 | 5606 | my_strlcat(ret, "NULL", sizeof(ret)); |
5d1187d1 KW |
5607 | } |
5608 | ||
fa07b8e5 | 5609 | my_strlcat(ret, ") returned ", sizeof(ret)); |
5d1187d1 KW |
5610 | |
5611 | if (retval) { | |
fa07b8e5 KW |
5612 | my_strlcat(ret, "\"", sizeof(ret)); |
5613 | my_strlcat(ret, retval, sizeof(ret)); | |
5614 | my_strlcat(ret, "\"", sizeof(ret)); | |
5d1187d1 KW |
5615 | } |
5616 | else { | |
fa07b8e5 | 5617 | my_strlcat(ret, "NULL", sizeof(ret)); |
5d1187d1 KW |
5618 | } |
5619 | ||
5620 | assert(strlen(ret) < sizeof(ret)); | |
5621 | ||
5622 | return ret; | |
5623 | } | |
5624 | ||
5625 | #endif | |
747c467a | 5626 | |
e9bc6d6b KW |
5627 | void |
5628 | Perl_thread_locale_init() | |
5629 | { | |
5630 | /* Called from a thread on startup*/ | |
5631 | ||
5632 | #ifdef USE_THREAD_SAFE_LOCALE | |
5633 | ||
5634 | dTHX_DEBUGGING; | |
5635 | ||
5636 | /* C starts the new thread in the global C locale. If we are thread-safe, | |
5637 | * we want to not be in the global locale */ | |
5638 | ||
5639 | DEBUG_L(PerlIO_printf(Perl_debug_log, | |
5640 | "%s:%d: new thread, initial locale is %s; calling setlocale\n", | |
5641 | __FILE__, __LINE__, setlocale(LC_ALL, NULL))); | |
5642 | ||
5643 | # ifdef WIN32 | |
5644 | ||
5645 | _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); | |
5646 | ||
5647 | # else | |
5648 | ||
5649 | Perl_setlocale(LC_ALL, "C"); | |
5650 | ||
5651 | # endif | |
5652 | #endif | |
5653 | ||
5654 | } | |
5655 | ||
5656 | void | |
5657 | Perl_thread_locale_term() | |
5658 | { | |
5659 | /* Called from a thread as it gets ready to terminate */ | |
5660 | ||
5661 | #ifdef USE_THREAD_SAFE_LOCALE | |
5662 | ||
5663 | /* C starts the new thread in the global C locale. If we are thread-safe, | |
5664 | * we want to not be in the global locale */ | |
5665 | ||
5666 | # ifndef WIN32 | |
5667 | ||
5668 | { /* Free up */ | |
e7b55bf0 | 5669 | dVAR; |
e9bc6d6b | 5670 | locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE); |
e72200e7 | 5671 | if (cur_obj != LC_GLOBAL_LOCALE && cur_obj != PL_C_locale_obj) { |
e9bc6d6b KW |
5672 | freelocale(cur_obj); |
5673 | } | |
5674 | } | |
5675 | ||
5676 | # endif | |
5677 | #endif | |
5678 | ||
5679 | } | |
747c467a KW |
5680 | |
5681 | /* | |
14d04a33 | 5682 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 5683 | */ |