Line data Source code
1 : /*-----------------------------------------------------------------------
2 : *
3 : * PostgreSQL locale utilities
4 : *
5 : * Portions Copyright (c) 2002-2025, PostgreSQL Global Development Group
6 : *
7 : * src/backend/utils/adt/pg_locale.c
8 : *
9 : *-----------------------------------------------------------------------
10 : */
11 :
12 : /*----------
13 : * Here is how the locale stuff is handled: LC_COLLATE and LC_CTYPE
14 : * are fixed at CREATE DATABASE time, stored in pg_database, and cannot
15 : * be changed. Thus, the effects of strcoll(), strxfrm(), isupper(),
16 : * toupper(), etc. are always in the same fixed locale.
17 : *
18 : * LC_MESSAGES is settable at run time and will take effect
19 : * immediately.
20 : *
21 : * The other categories, LC_MONETARY, LC_NUMERIC, and LC_TIME are
22 : * permanently set to "C", and then we use temporary locale_t
23 : * objects when we need to look up locale data based on the GUCs
24 : * of the same name. Information is cached when the GUCs change.
25 : * The cached information is only used by the formatting functions
26 : * (to_char, etc.) and the money type. For the user, this should all be
27 : * transparent.
28 : *----------
29 : */
30 :
31 :
32 : #include "postgres.h"
33 :
34 : #include <time.h>
35 :
36 : #include "access/htup_details.h"
37 : #include "catalog/pg_collation.h"
38 : #include "catalog/pg_database.h"
39 : #include "common/hashfn.h"
40 : #include "common/string.h"
41 : #include "mb/pg_wchar.h"
42 : #include "miscadmin.h"
43 : #include "utils/builtins.h"
44 : #include "utils/formatting.h"
45 : #include "utils/guc_hooks.h"
46 : #include "utils/lsyscache.h"
47 : #include "utils/memutils.h"
48 : #include "utils/pg_locale.h"
49 : #include "utils/syscache.h"
50 :
51 : #ifdef WIN32
52 : #include <shlwapi.h>
53 : #endif
54 :
55 : /* Error triggered for locale-sensitive subroutines */
56 : #define PGLOCALE_SUPPORT_ERROR(provider) \
57 : elog(ERROR, "unsupported collprovider for %s: %c", __func__, provider)
58 :
59 : /*
60 : * This should be large enough that most strings will fit, but small enough
61 : * that we feel comfortable putting it on the stack
62 : */
63 : #define TEXTBUFLEN 1024
64 :
65 : #define MAX_L10N_DATA 80
66 :
67 : /* pg_locale_builtin.c */
68 : extern pg_locale_t create_pg_locale_builtin(Oid collid, MemoryContext context);
69 : extern char *get_collation_actual_version_builtin(const char *collcollate);
70 :
71 : /* pg_locale_icu.c */
72 : #ifdef USE_ICU
73 : extern UCollator *pg_ucol_open(const char *loc_str);
74 : extern char *get_collation_actual_version_icu(const char *collcollate);
75 : #endif
76 : extern pg_locale_t create_pg_locale_icu(Oid collid, MemoryContext context);
77 :
78 : /* pg_locale_libc.c */
79 : extern pg_locale_t create_pg_locale_libc(Oid collid, MemoryContext context);
80 : extern char *get_collation_actual_version_libc(const char *collcollate);
81 :
82 : extern size_t strlower_builtin(char *dst, size_t dstsize, const char *src,
83 : ssize_t srclen, pg_locale_t locale);
84 : extern size_t strtitle_builtin(char *dst, size_t dstsize, const char *src,
85 : ssize_t srclen, pg_locale_t locale);
86 : extern size_t strupper_builtin(char *dst, size_t dstsize, const char *src,
87 : ssize_t srclen, pg_locale_t locale);
88 : extern size_t strfold_builtin(char *dst, size_t dstsize, const char *src,
89 : ssize_t srclen, pg_locale_t locale);
90 :
91 : extern size_t strlower_icu(char *dst, size_t dstsize, const char *src,
92 : ssize_t srclen, pg_locale_t locale);
93 : extern size_t strtitle_icu(char *dst, size_t dstsize, const char *src,
94 : ssize_t srclen, pg_locale_t locale);
95 : extern size_t strupper_icu(char *dst, size_t dstsize, const char *src,
96 : ssize_t srclen, pg_locale_t locale);
97 : extern size_t strfold_icu(char *dst, size_t dstsize, const char *src,
98 : ssize_t srclen, pg_locale_t locale);
99 :
100 : extern size_t strlower_libc(char *dst, size_t dstsize, const char *src,
101 : ssize_t srclen, pg_locale_t locale);
102 : extern size_t strtitle_libc(char *dst, size_t dstsize, const char *src,
103 : ssize_t srclen, pg_locale_t locale);
104 : extern size_t strupper_libc(char *dst, size_t dstsize, const char *src,
105 : ssize_t srclen, pg_locale_t locale);
106 :
107 : /* GUC settings */
108 : char *locale_messages;
109 : char *locale_monetary;
110 : char *locale_numeric;
111 : char *locale_time;
112 :
113 : int icu_validation_level = WARNING;
114 :
115 : /*
116 : * lc_time localization cache.
117 : *
118 : * We use only the first 7 or 12 entries of these arrays. The last array
119 : * element is left as NULL for the convenience of outside code that wants
120 : * to sequentially scan these arrays.
121 : */
122 : char *localized_abbrev_days[7 + 1];
123 : char *localized_full_days[7 + 1];
124 : char *localized_abbrev_months[12 + 1];
125 : char *localized_full_months[12 + 1];
126 :
127 : /* is the databases's LC_CTYPE the C locale? */
128 : bool database_ctype_is_c = false;
129 :
130 : static pg_locale_t default_locale = NULL;
131 :
132 : /* indicates whether locale information cache is valid */
133 : static bool CurrentLocaleConvValid = false;
134 : static bool CurrentLCTimeValid = false;
135 :
136 : /* Cache for collation-related knowledge */
137 :
138 : typedef struct
139 : {
140 : Oid collid; /* hash key: pg_collation OID */
141 : pg_locale_t locale; /* locale_t struct, or 0 if not valid */
142 :
143 : /* needed for simplehash */
144 : uint32 hash;
145 : char status;
146 : } collation_cache_entry;
147 :
148 : #define SH_PREFIX collation_cache
149 : #define SH_ELEMENT_TYPE collation_cache_entry
150 : #define SH_KEY_TYPE Oid
151 : #define SH_KEY collid
152 : #define SH_HASH_KEY(tb, key) murmurhash32((uint32) key)
153 : #define SH_EQUAL(tb, a, b) (a == b)
154 : #define SH_GET_HASH(tb, a) a->hash
155 : #define SH_SCOPE static inline
156 : #define SH_STORE_HASH
157 : #define SH_DECLARE
158 : #define SH_DEFINE
159 : #include "lib/simplehash.h"
160 :
161 : static MemoryContext CollationCacheContext = NULL;
162 : static collation_cache_hash *CollationCache = NULL;
163 :
164 : /*
165 : * The collation cache is often accessed repeatedly for the same collation, so
166 : * remember the last one used.
167 : */
168 : static Oid last_collation_cache_oid = InvalidOid;
169 : static pg_locale_t last_collation_cache_locale = NULL;
170 :
171 : #if defined(WIN32) && defined(LC_MESSAGES)
172 : static char *IsoLocaleName(const char *);
173 : #endif
174 :
175 : /*
176 : * pg_perm_setlocale
177 : *
178 : * This wraps the libc function setlocale(), with two additions. First, when
179 : * changing LC_CTYPE, update gettext's encoding for the current message
180 : * domain. GNU gettext automatically tracks LC_CTYPE on most platforms, but
181 : * not on Windows. Second, if the operation is successful, the corresponding
182 : * LC_XXX environment variable is set to match. By setting the environment
183 : * variable, we ensure that any subsequent use of setlocale(..., "") will
184 : * preserve the settings made through this routine. Of course, LC_ALL must
185 : * also be unset to fully ensure that, but that has to be done elsewhere after
186 : * all the individual LC_XXX variables have been set correctly. (Thank you
187 : * Perl for making this kluge necessary.)
188 : */
189 : char *
190 99528 : pg_perm_setlocale(int category, const char *locale)
191 : {
192 : char *result;
193 : const char *envvar;
194 :
195 : #ifndef WIN32
196 99528 : result = setlocale(category, locale);
197 : #else
198 :
199 : /*
200 : * On Windows, setlocale(LC_MESSAGES) does not work, so just assume that
201 : * the given value is good and set it in the environment variables. We
202 : * must ignore attempts to set to "", which means "keep using the old
203 : * environment value".
204 : */
205 : #ifdef LC_MESSAGES
206 : if (category == LC_MESSAGES)
207 : {
208 : result = (char *) locale;
209 : if (locale == NULL || locale[0] == '\0')
210 : return result;
211 : }
212 : else
213 : #endif
214 : result = setlocale(category, locale);
215 : #endif /* WIN32 */
216 :
217 99528 : if (result == NULL)
218 0 : return result; /* fall out immediately on failure */
219 :
220 : /*
221 : * Use the right encoding in translated messages. Under ENABLE_NLS, let
222 : * pg_bind_textdomain_codeset() figure it out. Under !ENABLE_NLS, message
223 : * format strings are ASCII, but database-encoding strings may enter the
224 : * message via %s. This makes the overall message encoding equal to the
225 : * database encoding.
226 : */
227 99528 : if (category == LC_CTYPE)
228 : {
229 : static char save_lc_ctype[LOCALE_NAME_BUFLEN];
230 :
231 : /* copy setlocale() return value before callee invokes it again */
232 33816 : strlcpy(save_lc_ctype, result, sizeof(save_lc_ctype));
233 33816 : result = save_lc_ctype;
234 :
235 : #ifdef ENABLE_NLS
236 33816 : SetMessageEncoding(pg_bind_textdomain_codeset(textdomain(NULL)));
237 : #else
238 : SetMessageEncoding(GetDatabaseEncoding());
239 : #endif
240 : }
241 :
242 99528 : switch (category)
243 : {
244 33816 : case LC_COLLATE:
245 33816 : envvar = "LC_COLLATE";
246 33816 : break;
247 33816 : case LC_CTYPE:
248 33816 : envvar = "LC_CTYPE";
249 33816 : break;
250 : #ifdef LC_MESSAGES
251 20904 : case LC_MESSAGES:
252 20904 : envvar = "LC_MESSAGES";
253 : #ifdef WIN32
254 : result = IsoLocaleName(locale);
255 : if (result == NULL)
256 : result = (char *) locale;
257 : elog(DEBUG3, "IsoLocaleName() executed; locale: \"%s\"", result);
258 : #endif /* WIN32 */
259 20904 : break;
260 : #endif /* LC_MESSAGES */
261 3664 : case LC_MONETARY:
262 3664 : envvar = "LC_MONETARY";
263 3664 : break;
264 3664 : case LC_NUMERIC:
265 3664 : envvar = "LC_NUMERIC";
266 3664 : break;
267 3664 : case LC_TIME:
268 3664 : envvar = "LC_TIME";
269 3664 : break;
270 0 : default:
271 0 : elog(FATAL, "unrecognized LC category: %d", category);
272 : return NULL; /* keep compiler quiet */
273 : }
274 :
275 99528 : if (setenv(envvar, result, 1) != 0)
276 0 : return NULL;
277 :
278 99528 : return result;
279 : }
280 :
281 :
282 : /*
283 : * Is the locale name valid for the locale category?
284 : *
285 : * If successful, and canonname isn't NULL, a palloc'd copy of the locale's
286 : * canonical name is stored there. This is especially useful for figuring out
287 : * what locale name "" means (ie, the server environment value). (Actually,
288 : * it seems that on most implementations that's the only thing it's good for;
289 : * we could wish that setlocale gave back a canonically spelled version of
290 : * the locale name, but typically it doesn't.)
291 : */
292 : bool
293 66318 : check_locale(int category, const char *locale, char **canonname)
294 : {
295 : char *save;
296 : char *res;
297 :
298 : /* Don't let Windows' non-ASCII locale names in. */
299 66318 : if (!pg_is_ascii(locale))
300 : {
301 0 : ereport(WARNING,
302 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
303 : errmsg("locale name \"%s\" contains non-ASCII characters",
304 : locale)));
305 0 : return false;
306 : }
307 :
308 66318 : if (canonname)
309 1542 : *canonname = NULL; /* in case of failure */
310 :
311 66318 : save = setlocale(category, NULL);
312 66318 : if (!save)
313 0 : return false; /* won't happen, we hope */
314 :
315 : /* save may be pointing at a modifiable scratch variable, see above. */
316 66318 : save = pstrdup(save);
317 :
318 : /* set the locale with setlocale, to see if it accepts it. */
319 66318 : res = setlocale(category, locale);
320 :
321 : /* save canonical name if requested. */
322 66318 : if (res && canonname)
323 1538 : *canonname = pstrdup(res);
324 :
325 : /* restore old value. */
326 66318 : if (!setlocale(category, save))
327 0 : elog(WARNING, "failed to restore old locale \"%s\"", save);
328 66318 : pfree(save);
329 :
330 : /* Don't let Windows' non-ASCII locale names out. */
331 66318 : if (canonname && *canonname && !pg_is_ascii(*canonname))
332 : {
333 0 : ereport(WARNING,
334 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
335 : errmsg("locale name \"%s\" contains non-ASCII characters",
336 : *canonname)));
337 0 : pfree(*canonname);
338 0 : *canonname = NULL;
339 0 : return false;
340 : }
341 :
342 66318 : return (res != NULL);
343 : }
344 :
345 :
346 : /*
347 : * GUC check/assign hooks
348 : *
349 : * For most locale categories, the assign hook doesn't actually set the locale
350 : * permanently, just reset flags so that the next use will cache the
351 : * appropriate values. (See explanation at the top of this file.)
352 : *
353 : * Note: we accept value = "" as selecting the postmaster's environment
354 : * value, whatever it was (so long as the environment setting is legal).
355 : * This will have been locked down by an earlier call to pg_perm_setlocale.
356 : */
357 : bool
358 17410 : check_locale_monetary(char **newval, void **extra, GucSource source)
359 : {
360 17410 : return check_locale(LC_MONETARY, *newval, NULL);
361 : }
362 :
363 : void
364 17208 : assign_locale_monetary(const char *newval, void *extra)
365 : {
366 17208 : CurrentLocaleConvValid = false;
367 17208 : }
368 :
369 : bool
370 17416 : check_locale_numeric(char **newval, void **extra, GucSource source)
371 : {
372 17416 : return check_locale(LC_NUMERIC, *newval, NULL);
373 : }
374 :
375 : void
376 17220 : assign_locale_numeric(const char *newval, void *extra)
377 : {
378 17220 : CurrentLocaleConvValid = false;
379 17220 : }
380 :
381 : bool
382 17410 : check_locale_time(char **newval, void **extra, GucSource source)
383 : {
384 17410 : return check_locale(LC_TIME, *newval, NULL);
385 : }
386 :
387 : void
388 17208 : assign_locale_time(const char *newval, void *extra)
389 : {
390 17208 : CurrentLCTimeValid = false;
391 17208 : }
392 :
393 : /*
394 : * We allow LC_MESSAGES to actually be set globally.
395 : *
396 : * Note: we normally disallow value = "" because it wouldn't have consistent
397 : * semantics (it'd effectively just use the previous value). However, this
398 : * is the value passed for PGC_S_DEFAULT, so don't complain in that case,
399 : * not even if the attempted setting fails due to invalid environment value.
400 : * The idea there is just to accept the environment setting *if possible*
401 : * during startup, until we can read the proper value from postgresql.conf.
402 : */
403 : bool
404 17444 : check_locale_messages(char **newval, void **extra, GucSource source)
405 : {
406 17444 : if (**newval == '\0')
407 : {
408 4904 : if (source == PGC_S_DEFAULT)
409 4904 : return true;
410 : else
411 0 : return false;
412 : }
413 :
414 : /*
415 : * LC_MESSAGES category does not exist everywhere, but accept it anyway
416 : *
417 : * On Windows, we can't even check the value, so accept blindly
418 : */
419 : #if defined(LC_MESSAGES) && !defined(WIN32)
420 12540 : return check_locale(LC_MESSAGES, *newval, NULL);
421 : #else
422 : return true;
423 : #endif
424 : }
425 :
426 : void
427 17240 : assign_locale_messages(const char *newval, void *extra)
428 : {
429 : /*
430 : * LC_MESSAGES category does not exist everywhere, but accept it anyway.
431 : * We ignore failure, as per comment above.
432 : */
433 : #ifdef LC_MESSAGES
434 17240 : (void) pg_perm_setlocale(LC_MESSAGES, newval);
435 : #endif
436 17240 : }
437 :
438 :
439 : /*
440 : * Frees the malloced content of a struct lconv. (But not the struct
441 : * itself.) It's important that this not throw elog(ERROR).
442 : */
443 : static void
444 6 : free_struct_lconv(struct lconv *s)
445 : {
446 6 : free(s->decimal_point);
447 6 : free(s->thousands_sep);
448 6 : free(s->grouping);
449 6 : free(s->int_curr_symbol);
450 6 : free(s->currency_symbol);
451 6 : free(s->mon_decimal_point);
452 6 : free(s->mon_thousands_sep);
453 6 : free(s->mon_grouping);
454 6 : free(s->positive_sign);
455 6 : free(s->negative_sign);
456 6 : }
457 :
458 : /*
459 : * Check that all fields of a struct lconv (or at least, the ones we care
460 : * about) are non-NULL. The field list must match free_struct_lconv().
461 : */
462 : static bool
463 64 : struct_lconv_is_valid(struct lconv *s)
464 : {
465 64 : if (s->decimal_point == NULL)
466 0 : return false;
467 64 : if (s->thousands_sep == NULL)
468 0 : return false;
469 64 : if (s->grouping == NULL)
470 0 : return false;
471 64 : if (s->int_curr_symbol == NULL)
472 0 : return false;
473 64 : if (s->currency_symbol == NULL)
474 0 : return false;
475 64 : if (s->mon_decimal_point == NULL)
476 0 : return false;
477 64 : if (s->mon_thousands_sep == NULL)
478 0 : return false;
479 64 : if (s->mon_grouping == NULL)
480 0 : return false;
481 64 : if (s->positive_sign == NULL)
482 0 : return false;
483 64 : if (s->negative_sign == NULL)
484 0 : return false;
485 64 : return true;
486 : }
487 :
488 :
489 : /*
490 : * Convert the strdup'd string at *str from the specified encoding to the
491 : * database encoding.
492 : */
493 : static void
494 512 : db_encoding_convert(int encoding, char **str)
495 : {
496 : char *pstr;
497 : char *mstr;
498 :
499 : /* convert the string to the database encoding */
500 512 : pstr = pg_any_to_server(*str, strlen(*str), encoding);
501 512 : if (pstr == *str)
502 512 : return; /* no conversion happened */
503 :
504 : /* need it malloc'd not palloc'd */
505 0 : mstr = strdup(pstr);
506 0 : if (mstr == NULL)
507 0 : ereport(ERROR,
508 : (errcode(ERRCODE_OUT_OF_MEMORY),
509 : errmsg("out of memory")));
510 :
511 : /* replace old string */
512 0 : free(*str);
513 0 : *str = mstr;
514 :
515 0 : pfree(pstr);
516 : }
517 :
518 :
519 : /*
520 : * Return the POSIX lconv struct (contains number/money formatting
521 : * information) with locale information for all categories.
522 : */
523 : struct lconv *
524 2982 : PGLC_localeconv(void)
525 : {
526 : static struct lconv CurrentLocaleConv;
527 : static bool CurrentLocaleConvAllocated = false;
528 : struct lconv *extlconv;
529 : struct lconv tmp;
530 2982 : struct lconv worklconv = {0};
531 :
532 : /* Did we do it already? */
533 2982 : if (CurrentLocaleConvValid)
534 2918 : return &CurrentLocaleConv;
535 :
536 : /* Free any already-allocated storage */
537 64 : if (CurrentLocaleConvAllocated)
538 : {
539 6 : free_struct_lconv(&CurrentLocaleConv);
540 6 : CurrentLocaleConvAllocated = false;
541 : }
542 :
543 : /*
544 : * Use thread-safe method of obtaining a copy of lconv from the operating
545 : * system.
546 : */
547 64 : if (pg_localeconv_r(locale_monetary,
548 : locale_numeric,
549 : &tmp) != 0)
550 0 : elog(ERROR,
551 : "could not get lconv for LC_MONETARY = \"%s\", LC_NUMERIC = \"%s\": %m",
552 : locale_monetary, locale_numeric);
553 :
554 : /* Must copy data now so we can re-encode it. */
555 64 : extlconv = &tmp;
556 64 : worklconv.decimal_point = strdup(extlconv->decimal_point);
557 64 : worklconv.thousands_sep = strdup(extlconv->thousands_sep);
558 64 : worklconv.grouping = strdup(extlconv->grouping);
559 64 : worklconv.int_curr_symbol = strdup(extlconv->int_curr_symbol);
560 64 : worklconv.currency_symbol = strdup(extlconv->currency_symbol);
561 64 : worklconv.mon_decimal_point = strdup(extlconv->mon_decimal_point);
562 64 : worklconv.mon_thousands_sep = strdup(extlconv->mon_thousands_sep);
563 64 : worklconv.mon_grouping = strdup(extlconv->mon_grouping);
564 64 : worklconv.positive_sign = strdup(extlconv->positive_sign);
565 64 : worklconv.negative_sign = strdup(extlconv->negative_sign);
566 : /* Copy scalar fields as well */
567 64 : worklconv.int_frac_digits = extlconv->int_frac_digits;
568 64 : worklconv.frac_digits = extlconv->frac_digits;
569 64 : worklconv.p_cs_precedes = extlconv->p_cs_precedes;
570 64 : worklconv.p_sep_by_space = extlconv->p_sep_by_space;
571 64 : worklconv.n_cs_precedes = extlconv->n_cs_precedes;
572 64 : worklconv.n_sep_by_space = extlconv->n_sep_by_space;
573 64 : worklconv.p_sign_posn = extlconv->p_sign_posn;
574 64 : worklconv.n_sign_posn = extlconv->n_sign_posn;
575 :
576 : /* Free the contents of the object populated by pg_localeconv_r(). */
577 64 : pg_localeconv_free(&tmp);
578 :
579 : /* If any of the preceding strdup calls failed, complain now. */
580 64 : if (!struct_lconv_is_valid(&worklconv))
581 0 : ereport(ERROR,
582 : (errcode(ERRCODE_OUT_OF_MEMORY),
583 : errmsg("out of memory")));
584 :
585 64 : PG_TRY();
586 : {
587 : int encoding;
588 :
589 : /*
590 : * Now we must perform encoding conversion from whatever's associated
591 : * with the locales into the database encoding. If we can't identify
592 : * the encoding implied by LC_NUMERIC or LC_MONETARY (ie we get -1),
593 : * use PG_SQL_ASCII, which will result in just validating that the
594 : * strings are OK in the database encoding.
595 : */
596 64 : encoding = pg_get_encoding_from_locale(locale_numeric, true);
597 64 : if (encoding < 0)
598 0 : encoding = PG_SQL_ASCII;
599 :
600 64 : db_encoding_convert(encoding, &worklconv.decimal_point);
601 64 : db_encoding_convert(encoding, &worklconv.thousands_sep);
602 : /* grouping is not text and does not require conversion */
603 :
604 64 : encoding = pg_get_encoding_from_locale(locale_monetary, true);
605 64 : if (encoding < 0)
606 0 : encoding = PG_SQL_ASCII;
607 :
608 64 : db_encoding_convert(encoding, &worklconv.int_curr_symbol);
609 64 : db_encoding_convert(encoding, &worklconv.currency_symbol);
610 64 : db_encoding_convert(encoding, &worklconv.mon_decimal_point);
611 64 : db_encoding_convert(encoding, &worklconv.mon_thousands_sep);
612 : /* mon_grouping is not text and does not require conversion */
613 64 : db_encoding_convert(encoding, &worklconv.positive_sign);
614 64 : db_encoding_convert(encoding, &worklconv.negative_sign);
615 : }
616 0 : PG_CATCH();
617 : {
618 0 : free_struct_lconv(&worklconv);
619 0 : PG_RE_THROW();
620 : }
621 64 : PG_END_TRY();
622 :
623 : /*
624 : * Everything is good, so save the results.
625 : */
626 64 : CurrentLocaleConv = worklconv;
627 64 : CurrentLocaleConvAllocated = true;
628 64 : CurrentLocaleConvValid = true;
629 64 : return &CurrentLocaleConv;
630 : }
631 :
632 : #ifdef WIN32
633 : /*
634 : * On Windows, strftime() returns its output in encoding CP_ACP (the default
635 : * operating system codepage for the computer), which is likely different
636 : * from SERVER_ENCODING. This is especially important in Japanese versions
637 : * of Windows which will use SJIS encoding, which we don't support as a
638 : * server encoding.
639 : *
640 : * So, instead of using strftime(), use wcsftime() to return the value in
641 : * wide characters (internally UTF16) and then convert to UTF8, which we
642 : * know how to handle directly.
643 : *
644 : * Note that this only affects the calls to strftime() in this file, which are
645 : * used to get the locale-aware strings. Other parts of the backend use
646 : * pg_strftime(), which isn't locale-aware and does not need to be replaced.
647 : */
648 : static size_t
649 : strftime_l_win32(char *dst, size_t dstlen,
650 : const char *format, const struct tm *tm, locale_t locale)
651 : {
652 : size_t len;
653 : wchar_t wformat[8]; /* formats used below need 3 chars */
654 : wchar_t wbuf[MAX_L10N_DATA];
655 :
656 : /*
657 : * Get a wchar_t version of the format string. We only actually use
658 : * plain-ASCII formats in this file, so we can say that they're UTF8.
659 : */
660 : len = MultiByteToWideChar(CP_UTF8, 0, format, -1,
661 : wformat, lengthof(wformat));
662 : if (len == 0)
663 : elog(ERROR, "could not convert format string from UTF-8: error code %lu",
664 : GetLastError());
665 :
666 : len = _wcsftime_l(wbuf, MAX_L10N_DATA, wformat, tm, locale);
667 : if (len == 0)
668 : {
669 : /*
670 : * wcsftime failed, possibly because the result would not fit in
671 : * MAX_L10N_DATA. Return 0 with the contents of dst unspecified.
672 : */
673 : return 0;
674 : }
675 :
676 : len = WideCharToMultiByte(CP_UTF8, 0, wbuf, len, dst, dstlen - 1,
677 : NULL, NULL);
678 : if (len == 0)
679 : elog(ERROR, "could not convert string to UTF-8: error code %lu",
680 : GetLastError());
681 :
682 : dst[len] = '\0';
683 :
684 : return len;
685 : }
686 :
687 : /* redefine strftime_l() */
688 : #define strftime_l(a,b,c,d,e) strftime_l_win32(a,b,c,d,e)
689 : #endif /* WIN32 */
690 :
691 : /*
692 : * Subroutine for cache_locale_time().
693 : * Convert the given string from encoding "encoding" to the database
694 : * encoding, and store the result at *dst, replacing any previous value.
695 : */
696 : static void
697 1748 : cache_single_string(char **dst, const char *src, int encoding)
698 : {
699 : char *ptr;
700 : char *olddst;
701 :
702 : /* Convert the string to the database encoding, or validate it's OK */
703 1748 : ptr = pg_any_to_server(src, strlen(src), encoding);
704 :
705 : /* Store the string in long-lived storage, replacing any previous value */
706 1748 : olddst = *dst;
707 1748 : *dst = MemoryContextStrdup(TopMemoryContext, ptr);
708 1748 : if (olddst)
709 0 : pfree(olddst);
710 :
711 : /* Might as well clean up any palloc'd conversion result, too */
712 1748 : if (ptr != src)
713 0 : pfree(ptr);
714 1748 : }
715 :
716 : /*
717 : * Update the lc_time localization cache variables if needed.
718 : */
719 : void
720 49546 : cache_locale_time(void)
721 : {
722 : char buf[(2 * 7 + 2 * 12) * MAX_L10N_DATA];
723 : char *bufptr;
724 : time_t timenow;
725 : struct tm *timeinfo;
726 : struct tm timeinfobuf;
727 49546 : bool strftimefail = false;
728 : int encoding;
729 : int i;
730 : locale_t locale;
731 :
732 : /* did we do this already? */
733 49546 : if (CurrentLCTimeValid)
734 49500 : return;
735 :
736 46 : elog(DEBUG3, "cache_locale_time() executed; locale: \"%s\"", locale_time);
737 :
738 46 : errno = ENOENT;
739 : #ifdef WIN32
740 : locale = _create_locale(LC_ALL, locale_time);
741 : if (locale == (locale_t) 0)
742 : _dosmaperr(GetLastError());
743 : #else
744 46 : locale = newlocale(LC_ALL_MASK, locale_time, (locale_t) 0);
745 : #endif
746 46 : if (!locale)
747 0 : report_newlocale_failure(locale_time);
748 :
749 : /* We use times close to current time as data for strftime(). */
750 46 : timenow = time(NULL);
751 46 : timeinfo = gmtime_r(&timenow, &timeinfobuf);
752 :
753 : /* Store the strftime results in MAX_L10N_DATA-sized portions of buf[] */
754 46 : bufptr = buf;
755 :
756 : /*
757 : * MAX_L10N_DATA is sufficient buffer space for every known locale, and
758 : * POSIX defines no strftime() errors. (Buffer space exhaustion is not an
759 : * error.) An implementation might report errors (e.g. ENOMEM) by
760 : * returning 0 (or, less plausibly, a negative value) and setting errno.
761 : * Report errno just in case the implementation did that, but clear it in
762 : * advance of the calls so we don't emit a stale, unrelated errno.
763 : */
764 46 : errno = 0;
765 :
766 : /* localized days */
767 368 : for (i = 0; i < 7; i++)
768 : {
769 322 : timeinfo->tm_wday = i;
770 322 : if (strftime_l(bufptr, MAX_L10N_DATA, "%a", timeinfo, locale) <= 0)
771 0 : strftimefail = true;
772 322 : bufptr += MAX_L10N_DATA;
773 322 : if (strftime_l(bufptr, MAX_L10N_DATA, "%A", timeinfo, locale) <= 0)
774 0 : strftimefail = true;
775 322 : bufptr += MAX_L10N_DATA;
776 : }
777 :
778 : /* localized months */
779 598 : for (i = 0; i < 12; i++)
780 : {
781 552 : timeinfo->tm_mon = i;
782 552 : timeinfo->tm_mday = 1; /* make sure we don't have invalid date */
783 552 : if (strftime_l(bufptr, MAX_L10N_DATA, "%b", timeinfo, locale) <= 0)
784 0 : strftimefail = true;
785 552 : bufptr += MAX_L10N_DATA;
786 552 : if (strftime_l(bufptr, MAX_L10N_DATA, "%B", timeinfo, locale) <= 0)
787 0 : strftimefail = true;
788 552 : bufptr += MAX_L10N_DATA;
789 : }
790 :
791 : #ifdef WIN32
792 : _free_locale(locale);
793 : #else
794 46 : freelocale(locale);
795 : #endif
796 :
797 : /*
798 : * At this point we've done our best to clean up, and can throw errors, or
799 : * call functions that might throw errors, with a clean conscience.
800 : */
801 46 : if (strftimefail)
802 0 : elog(ERROR, "strftime_l() failed");
803 :
804 : #ifndef WIN32
805 :
806 : /*
807 : * As in PGLC_localeconv(), we must convert strftime()'s output from the
808 : * encoding implied by LC_TIME to the database encoding. If we can't
809 : * identify the LC_TIME encoding, just perform encoding validation.
810 : */
811 46 : encoding = pg_get_encoding_from_locale(locale_time, true);
812 46 : if (encoding < 0)
813 0 : encoding = PG_SQL_ASCII;
814 :
815 : #else
816 :
817 : /*
818 : * On Windows, strftime_win32() always returns UTF8 data, so convert from
819 : * that if necessary.
820 : */
821 : encoding = PG_UTF8;
822 :
823 : #endif /* WIN32 */
824 :
825 46 : bufptr = buf;
826 :
827 : /* localized days */
828 368 : for (i = 0; i < 7; i++)
829 : {
830 322 : cache_single_string(&localized_abbrev_days[i], bufptr, encoding);
831 322 : bufptr += MAX_L10N_DATA;
832 322 : cache_single_string(&localized_full_days[i], bufptr, encoding);
833 322 : bufptr += MAX_L10N_DATA;
834 : }
835 46 : localized_abbrev_days[7] = NULL;
836 46 : localized_full_days[7] = NULL;
837 :
838 : /* localized months */
839 598 : for (i = 0; i < 12; i++)
840 : {
841 552 : cache_single_string(&localized_abbrev_months[i], bufptr, encoding);
842 552 : bufptr += MAX_L10N_DATA;
843 552 : cache_single_string(&localized_full_months[i], bufptr, encoding);
844 552 : bufptr += MAX_L10N_DATA;
845 : }
846 46 : localized_abbrev_months[12] = NULL;
847 46 : localized_full_months[12] = NULL;
848 :
849 46 : CurrentLCTimeValid = true;
850 : }
851 :
852 :
853 : #if defined(WIN32) && defined(LC_MESSAGES)
854 : /*
855 : * Convert a Windows setlocale() argument to a Unix-style one.
856 : *
857 : * Regardless of platform, we install message catalogs under a Unix-style
858 : * LL[_CC][.ENCODING][@VARIANT] naming convention. Only LC_MESSAGES settings
859 : * following that style will elicit localized interface strings.
860 : *
861 : * Before Visual Studio 2012 (msvcr110.dll), Windows setlocale() accepted "C"
862 : * (but not "c") and strings of the form <Language>[_<Country>][.<CodePage>],
863 : * case-insensitive. setlocale() returns the fully-qualified form; for
864 : * example, setlocale("thaI") returns "Thai_Thailand.874". Internally,
865 : * setlocale() and _create_locale() select a "locale identifier"[1] and store
866 : * it in an undocumented _locale_t field. From that LCID, we can retrieve the
867 : * ISO 639 language and the ISO 3166 country. Character encoding does not
868 : * matter, because the server and client encodings govern that.
869 : *
870 : * Windows Vista introduced the "locale name" concept[2], closely following
871 : * RFC 4646. Locale identifiers are now deprecated. Starting with Visual
872 : * Studio 2012, setlocale() accepts locale names in addition to the strings it
873 : * accepted historically. It does not standardize them; setlocale("Th-tH")
874 : * returns "Th-tH". setlocale(category, "") still returns a traditional
875 : * string. Furthermore, msvcr110.dll changed the undocumented _locale_t
876 : * content to carry locale names instead of locale identifiers.
877 : *
878 : * Visual Studio 2015 should still be able to do the same as Visual Studio
879 : * 2012, but the declaration of locale_name is missing in _locale_t, causing
880 : * this code compilation to fail, hence this falls back instead on to
881 : * enumerating all system locales by using EnumSystemLocalesEx to find the
882 : * required locale name. If the input argument is in Unix-style then we can
883 : * get ISO Locale name directly by using GetLocaleInfoEx() with LCType as
884 : * LOCALE_SNAME.
885 : *
886 : * This function returns a pointer to a static buffer bearing the converted
887 : * name or NULL if conversion fails.
888 : *
889 : * [1] https://docs.microsoft.com/en-us/windows/win32/intl/locale-identifiers
890 : * [2] https://docs.microsoft.com/en-us/windows/win32/intl/locale-names
891 : */
892 :
893 : /*
894 : * Callback function for EnumSystemLocalesEx() in get_iso_localename().
895 : *
896 : * This function enumerates all system locales, searching for one that matches
897 : * an input with the format: <Language>[_<Country>], e.g.
898 : * English[_United States]
899 : *
900 : * The input is a three wchar_t array as an LPARAM. The first element is the
901 : * locale_name we want to match, the second element is an allocated buffer
902 : * where the Unix-style locale is copied if a match is found, and the third
903 : * element is the search status, 1 if a match was found, 0 otherwise.
904 : */
905 : static BOOL CALLBACK
906 : search_locale_enum(LPWSTR pStr, DWORD dwFlags, LPARAM lparam)
907 : {
908 : wchar_t test_locale[LOCALE_NAME_MAX_LENGTH];
909 : wchar_t **argv;
910 :
911 : (void) (dwFlags);
912 :
913 : argv = (wchar_t **) lparam;
914 : *argv[2] = (wchar_t) 0;
915 :
916 : memset(test_locale, 0, sizeof(test_locale));
917 :
918 : /* Get the name of the <Language> in English */
919 : if (GetLocaleInfoEx(pStr, LOCALE_SENGLISHLANGUAGENAME,
920 : test_locale, LOCALE_NAME_MAX_LENGTH))
921 : {
922 : /*
923 : * If the enumerated locale does not have a hyphen ("en") OR the
924 : * locale_name input does not have an underscore ("English"), we only
925 : * need to compare the <Language> tags.
926 : */
927 : if (wcsrchr(pStr, '-') == NULL || wcsrchr(argv[0], '_') == NULL)
928 : {
929 : if (_wcsicmp(argv[0], test_locale) == 0)
930 : {
931 : wcscpy(argv[1], pStr);
932 : *argv[2] = (wchar_t) 1;
933 : return FALSE;
934 : }
935 : }
936 :
937 : /*
938 : * We have to compare a full <Language>_<Country> tag, so we append
939 : * the underscore and name of the country/region in English, e.g.
940 : * "English_United States".
941 : */
942 : else
943 : {
944 : size_t len;
945 :
946 : wcscat(test_locale, L"_");
947 : len = wcslen(test_locale);
948 : if (GetLocaleInfoEx(pStr, LOCALE_SENGLISHCOUNTRYNAME,
949 : test_locale + len,
950 : LOCALE_NAME_MAX_LENGTH - len))
951 : {
952 : if (_wcsicmp(argv[0], test_locale) == 0)
953 : {
954 : wcscpy(argv[1], pStr);
955 : *argv[2] = (wchar_t) 1;
956 : return FALSE;
957 : }
958 : }
959 : }
960 : }
961 :
962 : return TRUE;
963 : }
964 :
965 : /*
966 : * This function converts a Windows locale name to an ISO formatted version
967 : * for Visual Studio 2015 or greater.
968 : *
969 : * Returns NULL, if no valid conversion was found.
970 : */
971 : static char *
972 : get_iso_localename(const char *winlocname)
973 : {
974 : wchar_t wc_locale_name[LOCALE_NAME_MAX_LENGTH];
975 : wchar_t buffer[LOCALE_NAME_MAX_LENGTH];
976 : static char iso_lc_messages[LOCALE_NAME_MAX_LENGTH];
977 : char *period;
978 : int len;
979 : int ret_val;
980 :
981 : /*
982 : * Valid locales have the following syntax:
983 : * <Language>[_<Country>[.<CodePage>]]
984 : *
985 : * GetLocaleInfoEx can only take locale name without code-page and for the
986 : * purpose of this API the code-page doesn't matter.
987 : */
988 : period = strchr(winlocname, '.');
989 : if (period != NULL)
990 : len = period - winlocname;
991 : else
992 : len = pg_mbstrlen(winlocname);
993 :
994 : memset(wc_locale_name, 0, sizeof(wc_locale_name));
995 : memset(buffer, 0, sizeof(buffer));
996 : MultiByteToWideChar(CP_ACP, 0, winlocname, len, wc_locale_name,
997 : LOCALE_NAME_MAX_LENGTH);
998 :
999 : /*
1000 : * If the lc_messages is already a Unix-style string, we have a direct
1001 : * match with LOCALE_SNAME, e.g. en-US, en_US.
1002 : */
1003 : ret_val = GetLocaleInfoEx(wc_locale_name, LOCALE_SNAME, (LPWSTR) &buffer,
1004 : LOCALE_NAME_MAX_LENGTH);
1005 : if (!ret_val)
1006 : {
1007 : /*
1008 : * Search for a locale in the system that matches language and country
1009 : * name.
1010 : */
1011 : wchar_t *argv[3];
1012 :
1013 : argv[0] = wc_locale_name;
1014 : argv[1] = buffer;
1015 : argv[2] = (wchar_t *) &ret_val;
1016 : EnumSystemLocalesEx(search_locale_enum, LOCALE_WINDOWS, (LPARAM) argv,
1017 : NULL);
1018 : }
1019 :
1020 : if (ret_val)
1021 : {
1022 : size_t rc;
1023 : char *hyphen;
1024 :
1025 : /* Locale names use only ASCII, any conversion locale suffices. */
1026 : rc = wchar2char(iso_lc_messages, buffer, sizeof(iso_lc_messages), NULL);
1027 : if (rc == -1 || rc == sizeof(iso_lc_messages))
1028 : return NULL;
1029 :
1030 : /*
1031 : * Since the message catalogs sit on a case-insensitive filesystem, we
1032 : * need not standardize letter case here. So long as we do not ship
1033 : * message catalogs for which it would matter, we also need not
1034 : * translate the script/variant portion, e.g. uz-Cyrl-UZ to
1035 : * uz_UZ@cyrillic. Simply replace the hyphen with an underscore.
1036 : */
1037 : hyphen = strchr(iso_lc_messages, '-');
1038 : if (hyphen)
1039 : *hyphen = '_';
1040 : return iso_lc_messages;
1041 : }
1042 :
1043 : return NULL;
1044 : }
1045 :
1046 : static char *
1047 : IsoLocaleName(const char *winlocname)
1048 : {
1049 : static char iso_lc_messages[LOCALE_NAME_MAX_LENGTH];
1050 :
1051 : if (pg_strcasecmp("c", winlocname) == 0 ||
1052 : pg_strcasecmp("posix", winlocname) == 0)
1053 : {
1054 : strcpy(iso_lc_messages, "C");
1055 : return iso_lc_messages;
1056 : }
1057 : else
1058 : return get_iso_localename(winlocname);
1059 : }
1060 :
1061 : #endif /* WIN32 && LC_MESSAGES */
1062 :
1063 : /*
1064 : * Create a new pg_locale_t struct for the given collation oid.
1065 : */
1066 : static pg_locale_t
1067 4172 : create_pg_locale(Oid collid, MemoryContext context)
1068 : {
1069 : HeapTuple tp;
1070 : Form_pg_collation collform;
1071 : pg_locale_t result;
1072 : Datum datum;
1073 : bool isnull;
1074 :
1075 4172 : tp = SearchSysCache1(COLLOID, ObjectIdGetDatum(collid));
1076 4172 : if (!HeapTupleIsValid(tp))
1077 0 : elog(ERROR, "cache lookup failed for collation %u", collid);
1078 4172 : collform = (Form_pg_collation) GETSTRUCT(tp);
1079 :
1080 4172 : if (collform->collprovider == COLLPROVIDER_BUILTIN)
1081 56 : result = create_pg_locale_builtin(collid, context);
1082 4116 : else if (collform->collprovider == COLLPROVIDER_ICU)
1083 186 : result = create_pg_locale_icu(collid, context);
1084 3930 : else if (collform->collprovider == COLLPROVIDER_LIBC)
1085 3930 : result = create_pg_locale_libc(collid, context);
1086 : else
1087 : /* shouldn't happen */
1088 0 : PGLOCALE_SUPPORT_ERROR(collform->collprovider);
1089 :
1090 4166 : result->is_default = false;
1091 :
1092 : Assert((result->collate_is_c && result->collate == NULL) ||
1093 : (!result->collate_is_c && result->collate != NULL));
1094 :
1095 4166 : datum = SysCacheGetAttr(COLLOID, tp, Anum_pg_collation_collversion,
1096 : &isnull);
1097 4166 : if (!isnull)
1098 : {
1099 : char *actual_versionstr;
1100 : char *collversionstr;
1101 :
1102 236 : collversionstr = TextDatumGetCString(datum);
1103 :
1104 236 : if (collform->collprovider == COLLPROVIDER_LIBC)
1105 0 : datum = SysCacheGetAttrNotNull(COLLOID, tp, Anum_pg_collation_collcollate);
1106 : else
1107 236 : datum = SysCacheGetAttrNotNull(COLLOID, tp, Anum_pg_collation_colllocale);
1108 :
1109 236 : actual_versionstr = get_collation_actual_version(collform->collprovider,
1110 236 : TextDatumGetCString(datum));
1111 236 : if (!actual_versionstr)
1112 : {
1113 : /*
1114 : * This could happen when specifying a version in CREATE COLLATION
1115 : * but the provider does not support versioning, or manually
1116 : * creating a mess in the catalogs.
1117 : */
1118 0 : ereport(ERROR,
1119 : (errmsg("collation \"%s\" has no actual version, but a version was recorded",
1120 : NameStr(collform->collname))));
1121 : }
1122 :
1123 236 : if (strcmp(actual_versionstr, collversionstr) != 0)
1124 0 : ereport(WARNING,
1125 : (errmsg("collation \"%s\" has version mismatch",
1126 : NameStr(collform->collname)),
1127 : errdetail("The collation in the database was created using version %s, "
1128 : "but the operating system provides version %s.",
1129 : collversionstr, actual_versionstr),
1130 : errhint("Rebuild all objects affected by this collation and run "
1131 : "ALTER COLLATION %s REFRESH VERSION, "
1132 : "or build PostgreSQL with the right library version.",
1133 : quote_qualified_identifier(get_namespace_name(collform->collnamespace),
1134 : NameStr(collform->collname)))));
1135 : }
1136 :
1137 4166 : ReleaseSysCache(tp);
1138 :
1139 4166 : return result;
1140 : }
1141 :
1142 : /*
1143 : * Initialize default_locale with database locale settings.
1144 : */
1145 : void
1146 30152 : init_database_collation(void)
1147 : {
1148 : HeapTuple tup;
1149 : Form_pg_database dbform;
1150 : pg_locale_t result;
1151 :
1152 : Assert(default_locale == NULL);
1153 :
1154 : /* Fetch our pg_database row normally, via syscache */
1155 30152 : tup = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(MyDatabaseId));
1156 30152 : if (!HeapTupleIsValid(tup))
1157 0 : elog(ERROR, "cache lookup failed for database %u", MyDatabaseId);
1158 30152 : dbform = (Form_pg_database) GETSTRUCT(tup);
1159 :
1160 30152 : if (dbform->datlocprovider == COLLPROVIDER_BUILTIN)
1161 1738 : result = create_pg_locale_builtin(DEFAULT_COLLATION_OID,
1162 : TopMemoryContext);
1163 28414 : else if (dbform->datlocprovider == COLLPROVIDER_ICU)
1164 26 : result = create_pg_locale_icu(DEFAULT_COLLATION_OID,
1165 : TopMemoryContext);
1166 28388 : else if (dbform->datlocprovider == COLLPROVIDER_LIBC)
1167 28388 : result = create_pg_locale_libc(DEFAULT_COLLATION_OID,
1168 : TopMemoryContext);
1169 : else
1170 : /* shouldn't happen */
1171 0 : PGLOCALE_SUPPORT_ERROR(dbform->datlocprovider);
1172 :
1173 30148 : result->is_default = true;
1174 30148 : ReleaseSysCache(tup);
1175 :
1176 30148 : default_locale = result;
1177 30148 : }
1178 :
1179 : /*
1180 : * Create a pg_locale_t from a collation OID. Results are cached for the
1181 : * lifetime of the backend. Thus, do not free the result with freelocale().
1182 : *
1183 : * For simplicity, we always generate COLLATE + CTYPE even though we
1184 : * might only need one of them. Since this is called only once per session,
1185 : * it shouldn't cost much.
1186 : */
1187 : pg_locale_t
1188 28734146 : pg_newlocale_from_collation(Oid collid)
1189 : {
1190 : collation_cache_entry *cache_entry;
1191 : bool found;
1192 :
1193 28734146 : if (collid == DEFAULT_COLLATION_OID)
1194 24707738 : return default_locale;
1195 :
1196 4026408 : if (!OidIsValid(collid))
1197 0 : elog(ERROR, "cache lookup failed for collation %u", collid);
1198 :
1199 4026408 : AssertCouldGetRelation();
1200 :
1201 4026408 : if (last_collation_cache_oid == collid)
1202 4020296 : return last_collation_cache_locale;
1203 :
1204 6112 : if (CollationCache == NULL)
1205 : {
1206 3860 : CollationCacheContext = AllocSetContextCreate(TopMemoryContext,
1207 : "collation cache",
1208 : ALLOCSET_DEFAULT_SIZES);
1209 3860 : CollationCache = collation_cache_create(CollationCacheContext,
1210 : 16, NULL);
1211 : }
1212 :
1213 6112 : cache_entry = collation_cache_insert(CollationCache, collid, &found);
1214 6112 : if (!found)
1215 : {
1216 : /*
1217 : * Make sure cache entry is marked invalid, in case we fail before
1218 : * setting things.
1219 : */
1220 4172 : cache_entry->locale = 0;
1221 : }
1222 :
1223 6112 : if (cache_entry->locale == 0)
1224 : {
1225 4172 : cache_entry->locale = create_pg_locale(collid, CollationCacheContext);
1226 : }
1227 :
1228 6106 : last_collation_cache_oid = collid;
1229 6106 : last_collation_cache_locale = cache_entry->locale;
1230 :
1231 6106 : return cache_entry->locale;
1232 : }
1233 :
1234 : /*
1235 : * Get provider-specific collation version string for the given collation from
1236 : * the operating system/library.
1237 : */
1238 : char *
1239 103748 : get_collation_actual_version(char collprovider, const char *collcollate)
1240 : {
1241 103748 : char *collversion = NULL;
1242 :
1243 103748 : if (collprovider == COLLPROVIDER_BUILTIN)
1244 1868 : collversion = get_collation_actual_version_builtin(collcollate);
1245 : #ifdef USE_ICU
1246 101880 : else if (collprovider == COLLPROVIDER_ICU)
1247 74068 : collversion = get_collation_actual_version_icu(collcollate);
1248 : #endif
1249 27812 : else if (collprovider == COLLPROVIDER_LIBC)
1250 27812 : collversion = get_collation_actual_version_libc(collcollate);
1251 :
1252 103748 : return collversion;
1253 : }
1254 :
1255 : size_t
1256 436428 : pg_strlower(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1257 : pg_locale_t locale)
1258 : {
1259 436428 : if (locale->provider == COLLPROVIDER_BUILTIN)
1260 13022 : return strlower_builtin(dst, dstsize, src, srclen, locale);
1261 : #ifdef USE_ICU
1262 423406 : else if (locale->provider == COLLPROVIDER_ICU)
1263 528 : return strlower_icu(dst, dstsize, src, srclen, locale);
1264 : #endif
1265 422878 : else if (locale->provider == COLLPROVIDER_LIBC)
1266 422878 : return strlower_libc(dst, dstsize, src, srclen, locale);
1267 : else
1268 : /* shouldn't happen */
1269 0 : PGLOCALE_SUPPORT_ERROR(locale->provider);
1270 :
1271 : return 0; /* keep compiler quiet */
1272 : }
1273 :
1274 : size_t
1275 232 : pg_strtitle(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1276 : pg_locale_t locale)
1277 : {
1278 232 : if (locale->provider == COLLPROVIDER_BUILTIN)
1279 194 : return strtitle_builtin(dst, dstsize, src, srclen, locale);
1280 : #ifdef USE_ICU
1281 38 : else if (locale->provider == COLLPROVIDER_ICU)
1282 30 : return strtitle_icu(dst, dstsize, src, srclen, locale);
1283 : #endif
1284 8 : else if (locale->provider == COLLPROVIDER_LIBC)
1285 8 : return strtitle_libc(dst, dstsize, src, srclen, locale);
1286 : else
1287 : /* shouldn't happen */
1288 0 : PGLOCALE_SUPPORT_ERROR(locale->provider);
1289 :
1290 : return 0; /* keep compiler quiet */
1291 : }
1292 :
1293 : size_t
1294 1034612 : pg_strupper(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1295 : pg_locale_t locale)
1296 : {
1297 1034612 : if (locale->provider == COLLPROVIDER_BUILTIN)
1298 316882 : return strupper_builtin(dst, dstsize, src, srclen, locale);
1299 : #ifdef USE_ICU
1300 717730 : else if (locale->provider == COLLPROVIDER_ICU)
1301 54 : return strupper_icu(dst, dstsize, src, srclen, locale);
1302 : #endif
1303 717676 : else if (locale->provider == COLLPROVIDER_LIBC)
1304 717676 : return strupper_libc(dst, dstsize, src, srclen, locale);
1305 : else
1306 : /* shouldn't happen */
1307 0 : PGLOCALE_SUPPORT_ERROR(locale->provider);
1308 :
1309 : return 0; /* keep compiler quiet */
1310 : }
1311 :
1312 : size_t
1313 24 : pg_strfold(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1314 : pg_locale_t locale)
1315 : {
1316 24 : if (locale->provider == COLLPROVIDER_BUILTIN)
1317 12 : return strfold_builtin(dst, dstsize, src, srclen, locale);
1318 : #ifdef USE_ICU
1319 12 : else if (locale->provider == COLLPROVIDER_ICU)
1320 12 : return strfold_icu(dst, dstsize, src, srclen, locale);
1321 : #endif
1322 : /* for libc, just use strlower */
1323 0 : else if (locale->provider == COLLPROVIDER_LIBC)
1324 0 : return strlower_libc(dst, dstsize, src, srclen, locale);
1325 : else
1326 : /* shouldn't happen */
1327 0 : PGLOCALE_SUPPORT_ERROR(locale->provider);
1328 :
1329 : return 0; /* keep compiler quiet */
1330 : }
1331 :
1332 : /*
1333 : * pg_strcoll
1334 : *
1335 : * Like pg_strncoll for NUL-terminated input strings.
1336 : */
1337 : int
1338 25161872 : pg_strcoll(const char *arg1, const char *arg2, pg_locale_t locale)
1339 : {
1340 25161872 : return locale->collate->strncoll(arg1, -1, arg2, -1, locale);
1341 : }
1342 :
1343 : /*
1344 : * pg_strncoll
1345 : *
1346 : * Call ucol_strcollUTF8(), ucol_strcoll(), strcoll_l() or wcscoll_l() as
1347 : * appropriate for the given locale, platform, and database encoding. If the
1348 : * locale is not specified, use the database collation.
1349 : *
1350 : * The input strings must be encoded in the database encoding. If an input
1351 : * string is NUL-terminated, its length may be specified as -1.
1352 : *
1353 : * The caller is responsible for breaking ties if the collation is
1354 : * deterministic; this maintains consistency with pg_strnxfrm(), which cannot
1355 : * easily account for deterministic collations.
1356 : */
1357 : int
1358 4444076 : pg_strncoll(const char *arg1, ssize_t len1, const char *arg2, ssize_t len2,
1359 : pg_locale_t locale)
1360 : {
1361 4444076 : return locale->collate->strncoll(arg1, len1, arg2, len2, locale);
1362 : }
1363 :
1364 : /*
1365 : * Return true if the collation provider supports pg_strxfrm() and
1366 : * pg_strnxfrm(); otherwise false.
1367 : *
1368 : *
1369 : * No similar problem is known for the ICU provider.
1370 : */
1371 : bool
1372 45472 : pg_strxfrm_enabled(pg_locale_t locale)
1373 : {
1374 : /*
1375 : * locale->collate->strnxfrm is still a required method, even if it may
1376 : * have the wrong behavior, because the planner uses it for estimates in
1377 : * some cases.
1378 : */
1379 45472 : return locale->collate->strxfrm_is_safe;
1380 : }
1381 :
1382 : /*
1383 : * pg_strxfrm
1384 : *
1385 : * Like pg_strnxfrm for a NUL-terminated input string.
1386 : */
1387 : size_t
1388 144 : pg_strxfrm(char *dest, const char *src, size_t destsize, pg_locale_t locale)
1389 : {
1390 144 : return locale->collate->strnxfrm(dest, destsize, src, -1, locale);
1391 : }
1392 :
1393 : /*
1394 : * pg_strnxfrm
1395 : *
1396 : * Transforms 'src' to a nul-terminated string stored in 'dest' such that
1397 : * ordinary strcmp() on transformed strings is equivalent to pg_strcoll() on
1398 : * untransformed strings.
1399 : *
1400 : * The input string must be encoded in the database encoding. If the input
1401 : * string is NUL-terminated, its length may be specified as -1. If 'destsize'
1402 : * is zero, 'dest' may be NULL.
1403 : *
1404 : * Not all providers support pg_strnxfrm() safely. The caller should check
1405 : * pg_strxfrm_enabled() first, otherwise this function may return wrong
1406 : * results or an error.
1407 : *
1408 : * Returns the number of bytes needed (or more) to store the transformed
1409 : * string, excluding the terminating nul byte. If the value returned is
1410 : * 'destsize' or greater, the resulting contents of 'dest' are undefined.
1411 : */
1412 : size_t
1413 10020 : pg_strnxfrm(char *dest, size_t destsize, const char *src, ssize_t srclen,
1414 : pg_locale_t locale)
1415 : {
1416 10020 : return locale->collate->strnxfrm(dest, destsize, src, srclen, locale);
1417 : }
1418 :
1419 : /*
1420 : * Return true if the collation provider supports pg_strxfrm_prefix() and
1421 : * pg_strnxfrm_prefix(); otherwise false.
1422 : */
1423 : bool
1424 1668 : pg_strxfrm_prefix_enabled(pg_locale_t locale)
1425 : {
1426 1668 : return (locale->collate->strnxfrm_prefix != NULL);
1427 : }
1428 :
1429 : /*
1430 : * pg_strxfrm_prefix
1431 : *
1432 : * Like pg_strnxfrm_prefix for a NUL-terminated input string.
1433 : */
1434 : size_t
1435 1668 : pg_strxfrm_prefix(char *dest, const char *src, size_t destsize,
1436 : pg_locale_t locale)
1437 : {
1438 1668 : return locale->collate->strnxfrm_prefix(dest, destsize, src, -1, locale);
1439 : }
1440 :
1441 : /*
1442 : * pg_strnxfrm_prefix
1443 : *
1444 : * Transforms 'src' to a byte sequence stored in 'dest' such that ordinary
1445 : * memcmp() on the byte sequence is equivalent to pg_strncoll() on
1446 : * untransformed strings. The result is not nul-terminated.
1447 : *
1448 : * The input string must be encoded in the database encoding. If the input
1449 : * string is NUL-terminated, its length may be specified as -1.
1450 : *
1451 : * Not all providers support pg_strnxfrm_prefix() safely. The caller should
1452 : * check pg_strxfrm_prefix_enabled() first, otherwise this function may return
1453 : * wrong results or an error.
1454 : *
1455 : * If destsize is not large enough to hold the resulting byte sequence, stores
1456 : * only the first destsize bytes in 'dest'. Returns the number of bytes
1457 : * actually copied to 'dest'.
1458 : */
1459 : size_t
1460 0 : pg_strnxfrm_prefix(char *dest, size_t destsize, const char *src,
1461 : ssize_t srclen, pg_locale_t locale)
1462 : {
1463 0 : return locale->collate->strnxfrm_prefix(dest, destsize, src, srclen, locale);
1464 : }
1465 :
1466 : /*
1467 : * Return required encoding ID for the given locale, or -1 if any encoding is
1468 : * valid for the locale.
1469 : */
1470 : int
1471 1930 : builtin_locale_encoding(const char *locale)
1472 : {
1473 1930 : if (strcmp(locale, "C") == 0)
1474 64 : return -1;
1475 1866 : else if (strcmp(locale, "C.UTF-8") == 0)
1476 1830 : return PG_UTF8;
1477 36 : else if (strcmp(locale, "PG_UNICODE_FAST") == 0)
1478 36 : return PG_UTF8;
1479 :
1480 :
1481 0 : ereport(ERROR,
1482 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1483 : errmsg("invalid locale name \"%s\" for builtin provider",
1484 : locale)));
1485 :
1486 : return 0; /* keep compiler quiet */
1487 : }
1488 :
1489 :
1490 : /*
1491 : * Validate the locale and encoding combination, and return the canonical form
1492 : * of the locale name.
1493 : */
1494 : const char *
1495 1910 : builtin_validate_locale(int encoding, const char *locale)
1496 : {
1497 1910 : const char *canonical_name = NULL;
1498 : int required_encoding;
1499 :
1500 1910 : if (strcmp(locale, "C") == 0)
1501 52 : canonical_name = "C";
1502 1858 : else if (strcmp(locale, "C.UTF-8") == 0 || strcmp(locale, "C.UTF8") == 0)
1503 1814 : canonical_name = "C.UTF-8";
1504 44 : else if (strcmp(locale, "PG_UNICODE_FAST") == 0)
1505 26 : canonical_name = "PG_UNICODE_FAST";
1506 :
1507 1910 : if (!canonical_name)
1508 18 : ereport(ERROR,
1509 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1510 : errmsg("invalid locale name \"%s\" for builtin provider",
1511 : locale)));
1512 :
1513 1892 : required_encoding = builtin_locale_encoding(canonical_name);
1514 1892 : if (required_encoding >= 0 && encoding != required_encoding)
1515 2 : ereport(ERROR,
1516 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1517 : errmsg("encoding \"%s\" does not match locale \"%s\"",
1518 : pg_encoding_to_char(encoding), locale)));
1519 :
1520 1890 : return canonical_name;
1521 : }
1522 :
1523 :
1524 :
1525 : /*
1526 : * Return the BCP47 language tag representation of the requested locale.
1527 : *
1528 : * This function should be called before passing the string to ucol_open(),
1529 : * because conversion to a language tag also performs "level 2
1530 : * canonicalization". In addition to producing a consistent format, level 2
1531 : * canonicalization is able to more accurately interpret different input
1532 : * locale string formats, such as POSIX and .NET IDs.
1533 : */
1534 : char *
1535 73762 : icu_language_tag(const char *loc_str, int elevel)
1536 : {
1537 : #ifdef USE_ICU
1538 : UErrorCode status;
1539 : char *langtag;
1540 73762 : size_t buflen = 32; /* arbitrary starting buffer size */
1541 73762 : const bool strict = true;
1542 :
1543 : /*
1544 : * A BCP47 language tag doesn't have a clearly-defined upper limit (cf.
1545 : * RFC5646 section 4.4). Additionally, in older ICU versions,
1546 : * uloc_toLanguageTag() doesn't always return the ultimate length on the
1547 : * first call, necessitating a loop.
1548 : */
1549 73762 : langtag = palloc(buflen);
1550 : while (true)
1551 : {
1552 73762 : status = U_ZERO_ERROR;
1553 73762 : uloc_toLanguageTag(loc_str, langtag, buflen, strict, &status);
1554 :
1555 : /* try again if the buffer is not large enough */
1556 73762 : if ((status == U_BUFFER_OVERFLOW_ERROR ||
1557 73762 : status == U_STRING_NOT_TERMINATED_WARNING) &&
1558 : buflen < MaxAllocSize)
1559 : {
1560 0 : buflen = Min(buflen * 2, MaxAllocSize);
1561 0 : langtag = repalloc(langtag, buflen);
1562 0 : continue;
1563 : }
1564 :
1565 73762 : break;
1566 : }
1567 :
1568 73762 : if (U_FAILURE(status))
1569 : {
1570 18 : pfree(langtag);
1571 :
1572 18 : if (elevel > 0)
1573 14 : ereport(elevel,
1574 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1575 : errmsg("could not convert locale name \"%s\" to language tag: %s",
1576 : loc_str, u_errorName(status))));
1577 12 : return NULL;
1578 : }
1579 :
1580 73744 : return langtag;
1581 : #else /* not USE_ICU */
1582 : ereport(ERROR,
1583 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1584 : errmsg("ICU is not supported in this build")));
1585 : return NULL; /* keep compiler quiet */
1586 : #endif /* not USE_ICU */
1587 : }
1588 :
1589 : /*
1590 : * Perform best-effort check that the locale is a valid one.
1591 : */
1592 : void
1593 168 : icu_validate_locale(const char *loc_str)
1594 : {
1595 : #ifdef USE_ICU
1596 : UCollator *collator;
1597 : UErrorCode status;
1598 : char lang[ULOC_LANG_CAPACITY];
1599 168 : bool found = false;
1600 168 : int elevel = icu_validation_level;
1601 :
1602 : /* no validation */
1603 168 : if (elevel < 0)
1604 12 : return;
1605 :
1606 : /* downgrade to WARNING during pg_upgrade */
1607 156 : if (IsBinaryUpgrade && elevel > WARNING)
1608 0 : elevel = WARNING;
1609 :
1610 : /* validate that we can extract the language */
1611 156 : status = U_ZERO_ERROR;
1612 156 : uloc_getLanguage(loc_str, lang, ULOC_LANG_CAPACITY, &status);
1613 156 : if (U_FAILURE(status) || status == U_STRING_NOT_TERMINATED_WARNING)
1614 : {
1615 0 : ereport(elevel,
1616 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1617 : errmsg("could not get language from ICU locale \"%s\": %s",
1618 : loc_str, u_errorName(status)),
1619 : errhint("To disable ICU locale validation, set the parameter \"%s\" to \"%s\".",
1620 : "icu_validation_level", "disabled")));
1621 0 : return;
1622 : }
1623 :
1624 : /* check for special language name */
1625 156 : if (strcmp(lang, "") == 0 ||
1626 46 : strcmp(lang, "root") == 0 || strcmp(lang, "und") == 0)
1627 110 : found = true;
1628 :
1629 : /* search for matching language within ICU */
1630 15140 : for (int32_t i = 0; !found && i < uloc_countAvailable(); i++)
1631 : {
1632 14984 : const char *otherloc = uloc_getAvailable(i);
1633 : char otherlang[ULOC_LANG_CAPACITY];
1634 :
1635 14984 : status = U_ZERO_ERROR;
1636 14984 : uloc_getLanguage(otherloc, otherlang, ULOC_LANG_CAPACITY, &status);
1637 14984 : if (U_FAILURE(status) || status == U_STRING_NOT_TERMINATED_WARNING)
1638 0 : continue;
1639 :
1640 14984 : if (strcmp(lang, otherlang) == 0)
1641 32 : found = true;
1642 : }
1643 :
1644 156 : if (!found)
1645 14 : ereport(elevel,
1646 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1647 : errmsg("ICU locale \"%s\" has unknown language \"%s\"",
1648 : loc_str, lang),
1649 : errhint("To disable ICU locale validation, set the parameter \"%s\" to \"%s\".",
1650 : "icu_validation_level", "disabled")));
1651 :
1652 : /* check that it can be opened */
1653 150 : collator = pg_ucol_open(loc_str);
1654 142 : ucol_close(collator);
1655 : #else /* not USE_ICU */
1656 : /* could get here if a collation was created by a build with ICU */
1657 : ereport(ERROR,
1658 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1659 : errmsg("ICU is not supported in this build")));
1660 : #endif /* not USE_ICU */
1661 : }
|
