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/guc_hooks.h"
45 : #include "utils/lsyscache.h"
46 : #include "utils/memutils.h"
47 : #include "utils/pg_locale.h"
48 : #include "utils/pg_locale_c.h"
49 : #include "utils/relcache.h"
50 : #include "utils/syscache.h"
51 :
52 : #ifdef WIN32
53 : #include <shlwapi.h>
54 : #endif
55 :
56 : /* Error triggered for locale-sensitive subroutines */
57 : #define PGLOCALE_SUPPORT_ERROR(provider) \
58 : elog(ERROR, "unsupported collprovider for %s: %c", __func__, provider)
59 :
60 : /*
61 : * This should be large enough that most strings will fit, but small enough
62 : * that we feel comfortable putting it on the stack
63 : */
64 : #define TEXTBUFLEN 1024
65 :
66 : #define MAX_L10N_DATA 80
67 :
68 : /* pg_locale_builtin.c */
69 : extern pg_locale_t create_pg_locale_builtin(Oid collid, MemoryContext context);
70 : extern char *get_collation_actual_version_builtin(const char *collcollate);
71 :
72 : /* pg_locale_icu.c */
73 : #ifdef USE_ICU
74 : extern UCollator *pg_ucol_open(const char *loc_str);
75 : extern char *get_collation_actual_version_icu(const char *collcollate);
76 : #endif
77 : extern pg_locale_t create_pg_locale_icu(Oid collid, MemoryContext context);
78 :
79 : /* pg_locale_libc.c */
80 : extern pg_locale_t create_pg_locale_libc(Oid collid, MemoryContext context);
81 : extern char *get_collation_actual_version_libc(const char *collcollate);
82 :
83 : /* GUC settings */
84 : char *locale_messages;
85 : char *locale_monetary;
86 : char *locale_numeric;
87 : char *locale_time;
88 :
89 : int icu_validation_level = WARNING;
90 :
91 : /*
92 : * lc_time localization cache.
93 : *
94 : * We use only the first 7 or 12 entries of these arrays. The last array
95 : * element is left as NULL for the convenience of outside code that wants
96 : * to sequentially scan these arrays.
97 : */
98 : char *localized_abbrev_days[7 + 1];
99 : char *localized_full_days[7 + 1];
100 : char *localized_abbrev_months[12 + 1];
101 : char *localized_full_months[12 + 1];
102 :
103 : static pg_locale_t default_locale = NULL;
104 :
105 : /* indicates whether locale information cache is valid */
106 : static bool CurrentLocaleConvValid = false;
107 : static bool CurrentLCTimeValid = false;
108 :
109 : static struct pg_locale_struct c_locale = {
110 : .deterministic = true,
111 : .collate_is_c = true,
112 : .ctype_is_c = true,
113 : };
114 :
115 : /* Cache for collation-related knowledge */
116 :
117 : typedef struct
118 : {
119 : Oid collid; /* hash key: pg_collation OID */
120 : pg_locale_t locale; /* locale_t struct, or 0 if not valid */
121 :
122 : /* needed for simplehash */
123 : uint32 hash;
124 : char status;
125 : } collation_cache_entry;
126 :
127 : #define SH_PREFIX collation_cache
128 : #define SH_ELEMENT_TYPE collation_cache_entry
129 : #define SH_KEY_TYPE Oid
130 : #define SH_KEY collid
131 : #define SH_HASH_KEY(tb, key) murmurhash32((uint32) key)
132 : #define SH_EQUAL(tb, a, b) (a == b)
133 : #define SH_GET_HASH(tb, a) a->hash
134 : #define SH_SCOPE static inline
135 : #define SH_STORE_HASH
136 : #define SH_DECLARE
137 : #define SH_DEFINE
138 : #include "lib/simplehash.h"
139 :
140 : static MemoryContext CollationCacheContext = NULL;
141 : static collation_cache_hash *CollationCache = NULL;
142 :
143 : /*
144 : * The collation cache is often accessed repeatedly for the same collation, so
145 : * remember the last one used.
146 : */
147 : static Oid last_collation_cache_oid = InvalidOid;
148 : static pg_locale_t last_collation_cache_locale = NULL;
149 :
150 : #if defined(WIN32) && defined(LC_MESSAGES)
151 : static char *IsoLocaleName(const char *);
152 : #endif
153 :
154 : /*
155 : * pg_perm_setlocale
156 : *
157 : * This wraps the libc function setlocale(), with two additions. First, when
158 : * changing LC_CTYPE, update gettext's encoding for the current message
159 : * domain. GNU gettext automatically tracks LC_CTYPE on most platforms, but
160 : * not on Windows. Second, if the operation is successful, the corresponding
161 : * LC_XXX environment variable is set to match. By setting the environment
162 : * variable, we ensure that any subsequent use of setlocale(..., "") will
163 : * preserve the settings made through this routine. Of course, LC_ALL must
164 : * also be unset to fully ensure that, but that has to be done elsewhere after
165 : * all the individual LC_XXX variables have been set correctly. (Thank you
166 : * Perl for making this kluge necessary.)
167 : */
168 : char *
169 75300 : pg_perm_setlocale(int category, const char *locale)
170 : {
171 : char *result;
172 : const char *envvar;
173 :
174 : #ifndef WIN32
175 75300 : result = setlocale(category, locale);
176 : #else
177 :
178 : /*
179 : * On Windows, setlocale(LC_MESSAGES) does not work, so just assume that
180 : * the given value is good and set it in the environment variables. We
181 : * must ignore attempts to set to "", which means "keep using the old
182 : * environment value".
183 : */
184 : #ifdef LC_MESSAGES
185 : if (category == LC_MESSAGES)
186 : {
187 : result = (char *) locale;
188 : if (locale == NULL || locale[0] == '\0')
189 : return result;
190 : }
191 : else
192 : #endif
193 : result = setlocale(category, locale);
194 : #endif /* WIN32 */
195 :
196 75300 : if (result == NULL)
197 0 : return result; /* fall out immediately on failure */
198 :
199 : /*
200 : * Use the right encoding in translated messages. Under ENABLE_NLS, let
201 : * pg_bind_textdomain_codeset() figure it out. Under !ENABLE_NLS, message
202 : * format strings are ASCII, but database-encoding strings may enter the
203 : * message via %s. This makes the overall message encoding equal to the
204 : * database encoding.
205 : */
206 75300 : if (category == LC_CTYPE)
207 : {
208 : static char save_lc_ctype[LOCALE_NAME_BUFLEN];
209 :
210 : /* copy setlocale() return value before callee invokes it again */
211 37480 : strlcpy(save_lc_ctype, result, sizeof(save_lc_ctype));
212 37480 : result = save_lc_ctype;
213 :
214 : #ifdef ENABLE_NLS
215 37480 : SetMessageEncoding(pg_bind_textdomain_codeset(textdomain(NULL)));
216 : #else
217 : SetMessageEncoding(GetDatabaseEncoding());
218 : #endif
219 : }
220 :
221 75300 : switch (category)
222 : {
223 3846 : case LC_COLLATE:
224 3846 : envvar = "LC_COLLATE";
225 3846 : break;
226 37480 : case LC_CTYPE:
227 37480 : envvar = "LC_CTYPE";
228 37480 : break;
229 : #ifdef LC_MESSAGES
230 22436 : case LC_MESSAGES:
231 22436 : envvar = "LC_MESSAGES";
232 : #ifdef WIN32
233 : result = IsoLocaleName(locale);
234 : if (result == NULL)
235 : result = (char *) locale;
236 : elog(DEBUG3, "IsoLocaleName() executed; locale: \"%s\"", result);
237 : #endif /* WIN32 */
238 22436 : break;
239 : #endif /* LC_MESSAGES */
240 3846 : case LC_MONETARY:
241 3846 : envvar = "LC_MONETARY";
242 3846 : break;
243 3846 : case LC_NUMERIC:
244 3846 : envvar = "LC_NUMERIC";
245 3846 : break;
246 3846 : case LC_TIME:
247 3846 : envvar = "LC_TIME";
248 3846 : break;
249 0 : default:
250 0 : elog(FATAL, "unrecognized LC category: %d", category);
251 : return NULL; /* keep compiler quiet */
252 : }
253 :
254 75300 : if (setenv(envvar, result, 1) != 0)
255 0 : return NULL;
256 :
257 75300 : return result;
258 : }
259 :
260 :
261 : /*
262 : * Is the locale name valid for the locale category?
263 : *
264 : * If successful, and canonname isn't NULL, a palloc'd copy of the locale's
265 : * canonical name is stored there. This is especially useful for figuring out
266 : * what locale name "" means (ie, the server environment value). (Actually,
267 : * it seems that on most implementations that's the only thing it's good for;
268 : * we could wish that setlocale gave back a canonically spelled version of
269 : * the locale name, but typically it doesn't.)
270 : */
271 : bool
272 71560 : check_locale(int category, const char *locale, char **canonname)
273 : {
274 : char *save;
275 : char *res;
276 :
277 : /* Don't let Windows' non-ASCII locale names in. */
278 71560 : if (!pg_is_ascii(locale))
279 : {
280 0 : ereport(WARNING,
281 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
282 : errmsg("locale name \"%s\" contains non-ASCII characters",
283 : locale)));
284 0 : return false;
285 : }
286 :
287 71560 : if (canonname)
288 1590 : *canonname = NULL; /* in case of failure */
289 :
290 71560 : save = setlocale(category, NULL);
291 71560 : if (!save)
292 0 : return false; /* won't happen, we hope */
293 :
294 : /* save may be pointing at a modifiable scratch variable, see above. */
295 71560 : save = pstrdup(save);
296 :
297 : /* set the locale with setlocale, to see if it accepts it. */
298 71560 : res = setlocale(category, locale);
299 :
300 : /* save canonical name if requested. */
301 71560 : if (res && canonname)
302 1586 : *canonname = pstrdup(res);
303 :
304 : /* restore old value. */
305 71560 : if (!setlocale(category, save))
306 0 : elog(WARNING, "failed to restore old locale \"%s\"", save);
307 71560 : pfree(save);
308 :
309 : /* Don't let Windows' non-ASCII locale names out. */
310 71560 : if (canonname && *canonname && !pg_is_ascii(*canonname))
311 : {
312 0 : ereport(WARNING,
313 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
314 : errmsg("locale name \"%s\" contains non-ASCII characters",
315 : *canonname)));
316 0 : pfree(*canonname);
317 0 : *canonname = NULL;
318 0 : return false;
319 : }
320 :
321 71560 : return (res != NULL);
322 : }
323 :
324 :
325 : /*
326 : * GUC check/assign hooks
327 : *
328 : * For most locale categories, the assign hook doesn't actually set the locale
329 : * permanently, just reset flags so that the next use will cache the
330 : * appropriate values. (See explanation at the top of this file.)
331 : *
332 : * Note: we accept value = "" as selecting the postmaster's environment
333 : * value, whatever it was (so long as the environment setting is legal).
334 : * This will have been locked down by an earlier call to pg_perm_setlocale.
335 : */
336 : bool
337 18772 : check_locale_monetary(char **newval, void **extra, GucSource source)
338 : {
339 18772 : return check_locale(LC_MONETARY, *newval, NULL);
340 : }
341 :
342 : void
343 18558 : assign_locale_monetary(const char *newval, void *extra)
344 : {
345 18558 : CurrentLocaleConvValid = false;
346 18558 : }
347 :
348 : bool
349 18778 : check_locale_numeric(char **newval, void **extra, GucSource source)
350 : {
351 18778 : return check_locale(LC_NUMERIC, *newval, NULL);
352 : }
353 :
354 : void
355 18570 : assign_locale_numeric(const char *newval, void *extra)
356 : {
357 18570 : CurrentLocaleConvValid = false;
358 18570 : }
359 :
360 : bool
361 18772 : check_locale_time(char **newval, void **extra, GucSource source)
362 : {
363 18772 : return check_locale(LC_TIME, *newval, NULL);
364 : }
365 :
366 : void
367 18558 : assign_locale_time(const char *newval, void *extra)
368 : {
369 18558 : CurrentLCTimeValid = false;
370 18558 : }
371 :
372 : /*
373 : * We allow LC_MESSAGES to actually be set globally.
374 : *
375 : * Note: we normally disallow value = "" because it wouldn't have consistent
376 : * semantics (it'd effectively just use the previous value). However, this
377 : * is the value passed for PGC_S_DEFAULT, so don't complain in that case,
378 : * not even if the attempted setting fails due to invalid environment value.
379 : * The idea there is just to accept the environment setting *if possible*
380 : * during startup, until we can read the proper value from postgresql.conf.
381 : */
382 : bool
383 18806 : check_locale_messages(char **newval, void **extra, GucSource source)
384 : {
385 18806 : if (**newval == '\0')
386 : {
387 5158 : if (source == PGC_S_DEFAULT)
388 5158 : return true;
389 : else
390 0 : return false;
391 : }
392 :
393 : /*
394 : * LC_MESSAGES category does not exist everywhere, but accept it anyway
395 : *
396 : * On Windows, we can't even check the value, so accept blindly
397 : */
398 : #if defined(LC_MESSAGES) && !defined(WIN32)
399 13648 : return check_locale(LC_MESSAGES, *newval, NULL);
400 : #else
401 : return true;
402 : #endif
403 : }
404 :
405 : void
406 18590 : assign_locale_messages(const char *newval, void *extra)
407 : {
408 : /*
409 : * LC_MESSAGES category does not exist everywhere, but accept it anyway.
410 : * We ignore failure, as per comment above.
411 : */
412 : #ifdef LC_MESSAGES
413 18590 : (void) pg_perm_setlocale(LC_MESSAGES, newval);
414 : #endif
415 18590 : }
416 :
417 :
418 : /*
419 : * Frees the malloced content of a struct lconv. (But not the struct
420 : * itself.) It's important that this not throw elog(ERROR).
421 : */
422 : static void
423 6 : free_struct_lconv(struct lconv *s)
424 : {
425 6 : free(s->decimal_point);
426 6 : free(s->thousands_sep);
427 6 : free(s->grouping);
428 6 : free(s->int_curr_symbol);
429 6 : free(s->currency_symbol);
430 6 : free(s->mon_decimal_point);
431 6 : free(s->mon_thousands_sep);
432 6 : free(s->mon_grouping);
433 6 : free(s->positive_sign);
434 6 : free(s->negative_sign);
435 6 : }
436 :
437 : /*
438 : * Check that all fields of a struct lconv (or at least, the ones we care
439 : * about) are non-NULL. The field list must match free_struct_lconv().
440 : */
441 : static bool
442 56 : struct_lconv_is_valid(struct lconv *s)
443 : {
444 56 : if (s->decimal_point == NULL)
445 0 : return false;
446 56 : if (s->thousands_sep == NULL)
447 0 : return false;
448 56 : if (s->grouping == NULL)
449 0 : return false;
450 56 : if (s->int_curr_symbol == NULL)
451 0 : return false;
452 56 : if (s->currency_symbol == NULL)
453 0 : return false;
454 56 : if (s->mon_decimal_point == NULL)
455 0 : return false;
456 56 : if (s->mon_thousands_sep == NULL)
457 0 : return false;
458 56 : if (s->mon_grouping == NULL)
459 0 : return false;
460 56 : if (s->positive_sign == NULL)
461 0 : return false;
462 56 : if (s->negative_sign == NULL)
463 0 : return false;
464 56 : return true;
465 : }
466 :
467 :
468 : /*
469 : * Convert the strdup'd string at *str from the specified encoding to the
470 : * database encoding.
471 : */
472 : static void
473 448 : db_encoding_convert(int encoding, char **str)
474 : {
475 : char *pstr;
476 : char *mstr;
477 :
478 : /* convert the string to the database encoding */
479 448 : pstr = pg_any_to_server(*str, strlen(*str), encoding);
480 448 : if (pstr == *str)
481 448 : return; /* no conversion happened */
482 :
483 : /* need it malloc'd not palloc'd */
484 0 : mstr = strdup(pstr);
485 0 : if (mstr == NULL)
486 0 : ereport(ERROR,
487 : (errcode(ERRCODE_OUT_OF_MEMORY),
488 : errmsg("out of memory")));
489 :
490 : /* replace old string */
491 0 : free(*str);
492 0 : *str = mstr;
493 :
494 0 : pfree(pstr);
495 : }
496 :
497 :
498 : /*
499 : * Return the POSIX lconv struct (contains number/money formatting
500 : * information) with locale information for all categories.
501 : */
502 : struct lconv *
503 2966 : PGLC_localeconv(void)
504 : {
505 : static struct lconv CurrentLocaleConv;
506 : static bool CurrentLocaleConvAllocated = false;
507 : struct lconv *extlconv;
508 : struct lconv tmp;
509 2966 : struct lconv worklconv = {0};
510 :
511 : /* Did we do it already? */
512 2966 : if (CurrentLocaleConvValid)
513 2910 : return &CurrentLocaleConv;
514 :
515 : /* Free any already-allocated storage */
516 56 : if (CurrentLocaleConvAllocated)
517 : {
518 6 : free_struct_lconv(&CurrentLocaleConv);
519 6 : CurrentLocaleConvAllocated = false;
520 : }
521 :
522 : /*
523 : * Use thread-safe method of obtaining a copy of lconv from the operating
524 : * system.
525 : */
526 56 : if (pg_localeconv_r(locale_monetary,
527 : locale_numeric,
528 : &tmp) != 0)
529 0 : elog(ERROR,
530 : "could not get lconv for LC_MONETARY = \"%s\", LC_NUMERIC = \"%s\": %m",
531 : locale_monetary, locale_numeric);
532 :
533 : /* Must copy data now so we can re-encode it. */
534 56 : extlconv = &tmp;
535 56 : worklconv.decimal_point = strdup(extlconv->decimal_point);
536 56 : worklconv.thousands_sep = strdup(extlconv->thousands_sep);
537 56 : worklconv.grouping = strdup(extlconv->grouping);
538 56 : worklconv.int_curr_symbol = strdup(extlconv->int_curr_symbol);
539 56 : worklconv.currency_symbol = strdup(extlconv->currency_symbol);
540 56 : worklconv.mon_decimal_point = strdup(extlconv->mon_decimal_point);
541 56 : worklconv.mon_thousands_sep = strdup(extlconv->mon_thousands_sep);
542 56 : worklconv.mon_grouping = strdup(extlconv->mon_grouping);
543 56 : worklconv.positive_sign = strdup(extlconv->positive_sign);
544 56 : worklconv.negative_sign = strdup(extlconv->negative_sign);
545 : /* Copy scalar fields as well */
546 56 : worklconv.int_frac_digits = extlconv->int_frac_digits;
547 56 : worklconv.frac_digits = extlconv->frac_digits;
548 56 : worklconv.p_cs_precedes = extlconv->p_cs_precedes;
549 56 : worklconv.p_sep_by_space = extlconv->p_sep_by_space;
550 56 : worklconv.n_cs_precedes = extlconv->n_cs_precedes;
551 56 : worklconv.n_sep_by_space = extlconv->n_sep_by_space;
552 56 : worklconv.p_sign_posn = extlconv->p_sign_posn;
553 56 : worklconv.n_sign_posn = extlconv->n_sign_posn;
554 :
555 : /* Free the contents of the object populated by pg_localeconv_r(). */
556 56 : pg_localeconv_free(&tmp);
557 :
558 : /* If any of the preceding strdup calls failed, complain now. */
559 56 : if (!struct_lconv_is_valid(&worklconv))
560 0 : ereport(ERROR,
561 : (errcode(ERRCODE_OUT_OF_MEMORY),
562 : errmsg("out of memory")));
563 :
564 56 : PG_TRY();
565 : {
566 : int encoding;
567 :
568 : /*
569 : * Now we must perform encoding conversion from whatever's associated
570 : * with the locales into the database encoding. If we can't identify
571 : * the encoding implied by LC_NUMERIC or LC_MONETARY (ie we get -1),
572 : * use PG_SQL_ASCII, which will result in just validating that the
573 : * strings are OK in the database encoding.
574 : */
575 56 : encoding = pg_get_encoding_from_locale(locale_numeric, true);
576 56 : if (encoding < 0)
577 0 : encoding = PG_SQL_ASCII;
578 :
579 56 : db_encoding_convert(encoding, &worklconv.decimal_point);
580 56 : db_encoding_convert(encoding, &worklconv.thousands_sep);
581 : /* grouping is not text and does not require conversion */
582 :
583 56 : encoding = pg_get_encoding_from_locale(locale_monetary, true);
584 56 : if (encoding < 0)
585 0 : encoding = PG_SQL_ASCII;
586 :
587 56 : db_encoding_convert(encoding, &worklconv.int_curr_symbol);
588 56 : db_encoding_convert(encoding, &worklconv.currency_symbol);
589 56 : db_encoding_convert(encoding, &worklconv.mon_decimal_point);
590 56 : db_encoding_convert(encoding, &worklconv.mon_thousands_sep);
591 : /* mon_grouping is not text and does not require conversion */
592 56 : db_encoding_convert(encoding, &worklconv.positive_sign);
593 56 : db_encoding_convert(encoding, &worklconv.negative_sign);
594 : }
595 0 : PG_CATCH();
596 : {
597 0 : free_struct_lconv(&worklconv);
598 0 : PG_RE_THROW();
599 : }
600 56 : PG_END_TRY();
601 :
602 : /*
603 : * Everything is good, so save the results.
604 : */
605 56 : CurrentLocaleConv = worklconv;
606 56 : CurrentLocaleConvAllocated = true;
607 56 : CurrentLocaleConvValid = true;
608 56 : return &CurrentLocaleConv;
609 : }
610 :
611 : #ifdef WIN32
612 : /*
613 : * On Windows, strftime() returns its output in encoding CP_ACP (the default
614 : * operating system codepage for the computer), which is likely different
615 : * from SERVER_ENCODING. This is especially important in Japanese versions
616 : * of Windows which will use SJIS encoding, which we don't support as a
617 : * server encoding.
618 : *
619 : * So, instead of using strftime(), use wcsftime() to return the value in
620 : * wide characters (internally UTF16) and then convert to UTF8, which we
621 : * know how to handle directly.
622 : *
623 : * Note that this only affects the calls to strftime() in this file, which are
624 : * used to get the locale-aware strings. Other parts of the backend use
625 : * pg_strftime(), which isn't locale-aware and does not need to be replaced.
626 : */
627 : static size_t
628 : strftime_l_win32(char *dst, size_t dstlen,
629 : const char *format, const struct tm *tm, locale_t locale)
630 : {
631 : size_t len;
632 : wchar_t wformat[8]; /* formats used below need 3 chars */
633 : wchar_t wbuf[MAX_L10N_DATA];
634 :
635 : /*
636 : * Get a wchar_t version of the format string. We only actually use
637 : * plain-ASCII formats in this file, so we can say that they're UTF8.
638 : */
639 : len = MultiByteToWideChar(CP_UTF8, 0, format, -1,
640 : wformat, lengthof(wformat));
641 : if (len == 0)
642 : elog(ERROR, "could not convert format string from UTF-8: error code %lu",
643 : GetLastError());
644 :
645 : len = _wcsftime_l(wbuf, MAX_L10N_DATA, wformat, tm, locale);
646 : if (len == 0)
647 : {
648 : /*
649 : * wcsftime failed, possibly because the result would not fit in
650 : * MAX_L10N_DATA. Return 0 with the contents of dst unspecified.
651 : */
652 : return 0;
653 : }
654 :
655 : len = WideCharToMultiByte(CP_UTF8, 0, wbuf, len, dst, dstlen - 1,
656 : NULL, NULL);
657 : if (len == 0)
658 : elog(ERROR, "could not convert string to UTF-8: error code %lu",
659 : GetLastError());
660 :
661 : dst[len] = '\0';
662 :
663 : return len;
664 : }
665 :
666 : /* redefine strftime_l() */
667 : #define strftime_l(a,b,c,d,e) strftime_l_win32(a,b,c,d,e)
668 : #endif /* WIN32 */
669 :
670 : /*
671 : * Subroutine for cache_locale_time().
672 : * Convert the given string from encoding "encoding" to the database
673 : * encoding, and store the result at *dst, replacing any previous value.
674 : */
675 : static void
676 1900 : cache_single_string(char **dst, const char *src, int encoding)
677 : {
678 : char *ptr;
679 : char *olddst;
680 :
681 : /* Convert the string to the database encoding, or validate it's OK */
682 1900 : ptr = pg_any_to_server(src, strlen(src), encoding);
683 :
684 : /* Store the string in long-lived storage, replacing any previous value */
685 1900 : olddst = *dst;
686 1900 : *dst = MemoryContextStrdup(TopMemoryContext, ptr);
687 1900 : if (olddst)
688 0 : pfree(olddst);
689 :
690 : /* Might as well clean up any palloc'd conversion result, too */
691 1900 : if (ptr != src)
692 0 : pfree(ptr);
693 1900 : }
694 :
695 : /*
696 : * Update the lc_time localization cache variables if needed.
697 : */
698 : void
699 50034 : cache_locale_time(void)
700 : {
701 : char buf[(2 * 7 + 2 * 12) * MAX_L10N_DATA];
702 : char *bufptr;
703 : time_t timenow;
704 : struct tm *timeinfo;
705 : struct tm timeinfobuf;
706 50034 : bool strftimefail = false;
707 : int encoding;
708 : int i;
709 : locale_t locale;
710 :
711 : /* did we do this already? */
712 50034 : if (CurrentLCTimeValid)
713 49984 : return;
714 :
715 50 : elog(DEBUG3, "cache_locale_time() executed; locale: \"%s\"", locale_time);
716 :
717 50 : errno = ENOENT;
718 : #ifdef WIN32
719 : locale = _create_locale(LC_ALL, locale_time);
720 : if (locale == (locale_t) 0)
721 : _dosmaperr(GetLastError());
722 : #else
723 50 : locale = newlocale(LC_ALL_MASK, locale_time, (locale_t) 0);
724 : #endif
725 50 : if (!locale)
726 0 : report_newlocale_failure(locale_time);
727 :
728 : /* We use times close to current time as data for strftime(). */
729 50 : timenow = time(NULL);
730 50 : timeinfo = gmtime_r(&timenow, &timeinfobuf);
731 :
732 : /* Store the strftime results in MAX_L10N_DATA-sized portions of buf[] */
733 50 : bufptr = buf;
734 :
735 : /*
736 : * MAX_L10N_DATA is sufficient buffer space for every known locale, and
737 : * POSIX defines no strftime() errors. (Buffer space exhaustion is not an
738 : * error.) An implementation might report errors (e.g. ENOMEM) by
739 : * returning 0 (or, less plausibly, a negative value) and setting errno.
740 : * Report errno just in case the implementation did that, but clear it in
741 : * advance of the calls so we don't emit a stale, unrelated errno.
742 : */
743 50 : errno = 0;
744 :
745 : /* localized days */
746 400 : for (i = 0; i < 7; i++)
747 : {
748 350 : timeinfo->tm_wday = i;
749 350 : if (strftime_l(bufptr, MAX_L10N_DATA, "%a", timeinfo, locale) <= 0)
750 0 : strftimefail = true;
751 350 : bufptr += MAX_L10N_DATA;
752 350 : if (strftime_l(bufptr, MAX_L10N_DATA, "%A", timeinfo, locale) <= 0)
753 0 : strftimefail = true;
754 350 : bufptr += MAX_L10N_DATA;
755 : }
756 :
757 : /* localized months */
758 650 : for (i = 0; i < 12; i++)
759 : {
760 600 : timeinfo->tm_mon = i;
761 600 : timeinfo->tm_mday = 1; /* make sure we don't have invalid date */
762 600 : if (strftime_l(bufptr, MAX_L10N_DATA, "%b", timeinfo, locale) <= 0)
763 0 : strftimefail = true;
764 600 : bufptr += MAX_L10N_DATA;
765 600 : if (strftime_l(bufptr, MAX_L10N_DATA, "%B", timeinfo, locale) <= 0)
766 0 : strftimefail = true;
767 600 : bufptr += MAX_L10N_DATA;
768 : }
769 :
770 : #ifdef WIN32
771 : _free_locale(locale);
772 : #else
773 50 : freelocale(locale);
774 : #endif
775 :
776 : /*
777 : * At this point we've done our best to clean up, and can throw errors, or
778 : * call functions that might throw errors, with a clean conscience.
779 : */
780 50 : if (strftimefail)
781 0 : elog(ERROR, "strftime_l() failed");
782 :
783 : #ifndef WIN32
784 :
785 : /*
786 : * As in PGLC_localeconv(), we must convert strftime()'s output from the
787 : * encoding implied by LC_TIME to the database encoding. If we can't
788 : * identify the LC_TIME encoding, just perform encoding validation.
789 : */
790 50 : encoding = pg_get_encoding_from_locale(locale_time, true);
791 50 : if (encoding < 0)
792 0 : encoding = PG_SQL_ASCII;
793 :
794 : #else
795 :
796 : /*
797 : * On Windows, strftime_win32() always returns UTF8 data, so convert from
798 : * that if necessary.
799 : */
800 : encoding = PG_UTF8;
801 :
802 : #endif /* WIN32 */
803 :
804 50 : bufptr = buf;
805 :
806 : /* localized days */
807 400 : for (i = 0; i < 7; i++)
808 : {
809 350 : cache_single_string(&localized_abbrev_days[i], bufptr, encoding);
810 350 : bufptr += MAX_L10N_DATA;
811 350 : cache_single_string(&localized_full_days[i], bufptr, encoding);
812 350 : bufptr += MAX_L10N_DATA;
813 : }
814 50 : localized_abbrev_days[7] = NULL;
815 50 : localized_full_days[7] = NULL;
816 :
817 : /* localized months */
818 650 : for (i = 0; i < 12; i++)
819 : {
820 600 : cache_single_string(&localized_abbrev_months[i], bufptr, encoding);
821 600 : bufptr += MAX_L10N_DATA;
822 600 : cache_single_string(&localized_full_months[i], bufptr, encoding);
823 600 : bufptr += MAX_L10N_DATA;
824 : }
825 50 : localized_abbrev_months[12] = NULL;
826 50 : localized_full_months[12] = NULL;
827 :
828 50 : CurrentLCTimeValid = true;
829 : }
830 :
831 :
832 : #if defined(WIN32) && defined(LC_MESSAGES)
833 : /*
834 : * Convert a Windows setlocale() argument to a Unix-style one.
835 : *
836 : * Regardless of platform, we install message catalogs under a Unix-style
837 : * LL[_CC][.ENCODING][@VARIANT] naming convention. Only LC_MESSAGES settings
838 : * following that style will elicit localized interface strings.
839 : *
840 : * Before Visual Studio 2012 (msvcr110.dll), Windows setlocale() accepted "C"
841 : * (but not "c") and strings of the form <Language>[_<Country>][.<CodePage>],
842 : * case-insensitive. setlocale() returns the fully-qualified form; for
843 : * example, setlocale("thaI") returns "Thai_Thailand.874". Internally,
844 : * setlocale() and _create_locale() select a "locale identifier"[1] and store
845 : * it in an undocumented _locale_t field. From that LCID, we can retrieve the
846 : * ISO 639 language and the ISO 3166 country. Character encoding does not
847 : * matter, because the server and client encodings govern that.
848 : *
849 : * Windows Vista introduced the "locale name" concept[2], closely following
850 : * RFC 4646. Locale identifiers are now deprecated. Starting with Visual
851 : * Studio 2012, setlocale() accepts locale names in addition to the strings it
852 : * accepted historically. It does not standardize them; setlocale("Th-tH")
853 : * returns "Th-tH". setlocale(category, "") still returns a traditional
854 : * string. Furthermore, msvcr110.dll changed the undocumented _locale_t
855 : * content to carry locale names instead of locale identifiers.
856 : *
857 : * Visual Studio 2015 should still be able to do the same as Visual Studio
858 : * 2012, but the declaration of locale_name is missing in _locale_t, causing
859 : * this code compilation to fail, hence this falls back instead on to
860 : * enumerating all system locales by using EnumSystemLocalesEx to find the
861 : * required locale name. If the input argument is in Unix-style then we can
862 : * get ISO Locale name directly by using GetLocaleInfoEx() with LCType as
863 : * LOCALE_SNAME.
864 : *
865 : * This function returns a pointer to a static buffer bearing the converted
866 : * name or NULL if conversion fails.
867 : *
868 : * [1] https://docs.microsoft.com/en-us/windows/win32/intl/locale-identifiers
869 : * [2] https://docs.microsoft.com/en-us/windows/win32/intl/locale-names
870 : */
871 :
872 : /*
873 : * Callback function for EnumSystemLocalesEx() in get_iso_localename().
874 : *
875 : * This function enumerates all system locales, searching for one that matches
876 : * an input with the format: <Language>[_<Country>], e.g.
877 : * English[_United States]
878 : *
879 : * The input is a three wchar_t array as an LPARAM. The first element is the
880 : * locale_name we want to match, the second element is an allocated buffer
881 : * where the Unix-style locale is copied if a match is found, and the third
882 : * element is the search status, 1 if a match was found, 0 otherwise.
883 : */
884 : static BOOL CALLBACK
885 : search_locale_enum(LPWSTR pStr, DWORD dwFlags, LPARAM lparam)
886 : {
887 : wchar_t test_locale[LOCALE_NAME_MAX_LENGTH];
888 : wchar_t **argv;
889 :
890 : (void) (dwFlags);
891 :
892 : argv = (wchar_t **) lparam;
893 : *argv[2] = (wchar_t) 0;
894 :
895 : memset(test_locale, 0, sizeof(test_locale));
896 :
897 : /* Get the name of the <Language> in English */
898 : if (GetLocaleInfoEx(pStr, LOCALE_SENGLISHLANGUAGENAME,
899 : test_locale, LOCALE_NAME_MAX_LENGTH))
900 : {
901 : /*
902 : * If the enumerated locale does not have a hyphen ("en") OR the
903 : * locale_name input does not have an underscore ("English"), we only
904 : * need to compare the <Language> tags.
905 : */
906 : if (wcsrchr(pStr, '-') == NULL || wcsrchr(argv[0], '_') == NULL)
907 : {
908 : if (_wcsicmp(argv[0], test_locale) == 0)
909 : {
910 : wcscpy(argv[1], pStr);
911 : *argv[2] = (wchar_t) 1;
912 : return FALSE;
913 : }
914 : }
915 :
916 : /*
917 : * We have to compare a full <Language>_<Country> tag, so we append
918 : * the underscore and name of the country/region in English, e.g.
919 : * "English_United States".
920 : */
921 : else
922 : {
923 : size_t len;
924 :
925 : wcscat(test_locale, L"_");
926 : len = wcslen(test_locale);
927 : if (GetLocaleInfoEx(pStr, LOCALE_SENGLISHCOUNTRYNAME,
928 : test_locale + len,
929 : LOCALE_NAME_MAX_LENGTH - len))
930 : {
931 : if (_wcsicmp(argv[0], test_locale) == 0)
932 : {
933 : wcscpy(argv[1], pStr);
934 : *argv[2] = (wchar_t) 1;
935 : return FALSE;
936 : }
937 : }
938 : }
939 : }
940 :
941 : return TRUE;
942 : }
943 :
944 : /*
945 : * This function converts a Windows locale name to an ISO formatted version
946 : * for Visual Studio 2015 or greater.
947 : *
948 : * Returns NULL, if no valid conversion was found.
949 : */
950 : static char *
951 : get_iso_localename(const char *winlocname)
952 : {
953 : wchar_t wc_locale_name[LOCALE_NAME_MAX_LENGTH];
954 : wchar_t buffer[LOCALE_NAME_MAX_LENGTH];
955 : static char iso_lc_messages[LOCALE_NAME_MAX_LENGTH];
956 : char *period;
957 : int len;
958 : int ret_val;
959 :
960 : /*
961 : * Valid locales have the following syntax:
962 : * <Language>[_<Country>[.<CodePage>]]
963 : *
964 : * GetLocaleInfoEx can only take locale name without code-page and for the
965 : * purpose of this API the code-page doesn't matter.
966 : */
967 : period = strchr(winlocname, '.');
968 : if (period != NULL)
969 : len = period - winlocname;
970 : else
971 : len = pg_mbstrlen(winlocname);
972 :
973 : memset(wc_locale_name, 0, sizeof(wc_locale_name));
974 : memset(buffer, 0, sizeof(buffer));
975 : MultiByteToWideChar(CP_ACP, 0, winlocname, len, wc_locale_name,
976 : LOCALE_NAME_MAX_LENGTH);
977 :
978 : /*
979 : * If the lc_messages is already a Unix-style string, we have a direct
980 : * match with LOCALE_SNAME, e.g. en-US, en_US.
981 : */
982 : ret_val = GetLocaleInfoEx(wc_locale_name, LOCALE_SNAME, (LPWSTR) &buffer,
983 : LOCALE_NAME_MAX_LENGTH);
984 : if (!ret_val)
985 : {
986 : /*
987 : * Search for a locale in the system that matches language and country
988 : * name.
989 : */
990 : wchar_t *argv[3];
991 :
992 : argv[0] = wc_locale_name;
993 : argv[1] = buffer;
994 : argv[2] = (wchar_t *) &ret_val;
995 : EnumSystemLocalesEx(search_locale_enum, LOCALE_WINDOWS, (LPARAM) argv,
996 : NULL);
997 : }
998 :
999 : if (ret_val)
1000 : {
1001 : size_t rc;
1002 : char *hyphen;
1003 :
1004 : /* Locale names use only ASCII, any conversion locale suffices. */
1005 : rc = wchar2char(iso_lc_messages, buffer, sizeof(iso_lc_messages), NULL);
1006 : if (rc == -1 || rc == sizeof(iso_lc_messages))
1007 : return NULL;
1008 :
1009 : /*
1010 : * Since the message catalogs sit on a case-insensitive filesystem, we
1011 : * need not standardize letter case here. So long as we do not ship
1012 : * message catalogs for which it would matter, we also need not
1013 : * translate the script/variant portion, e.g. uz-Cyrl-UZ to
1014 : * uz_UZ@cyrillic. Simply replace the hyphen with an underscore.
1015 : */
1016 : hyphen = strchr(iso_lc_messages, '-');
1017 : if (hyphen)
1018 : *hyphen = '_';
1019 : return iso_lc_messages;
1020 : }
1021 :
1022 : return NULL;
1023 : }
1024 :
1025 : static char *
1026 : IsoLocaleName(const char *winlocname)
1027 : {
1028 : static char iso_lc_messages[LOCALE_NAME_MAX_LENGTH];
1029 :
1030 : if (pg_strcasecmp("c", winlocname) == 0 ||
1031 : pg_strcasecmp("posix", winlocname) == 0)
1032 : {
1033 : strcpy(iso_lc_messages, "C");
1034 : return iso_lc_messages;
1035 : }
1036 : else
1037 : return get_iso_localename(winlocname);
1038 : }
1039 :
1040 : #endif /* WIN32 && LC_MESSAGES */
1041 :
1042 : /*
1043 : * Create a new pg_locale_t struct for the given collation oid.
1044 : */
1045 : static pg_locale_t
1046 326 : create_pg_locale(Oid collid, MemoryContext context)
1047 : {
1048 : HeapTuple tp;
1049 : Form_pg_collation collform;
1050 : pg_locale_t result;
1051 : Datum datum;
1052 : bool isnull;
1053 :
1054 326 : tp = SearchSysCache1(COLLOID, ObjectIdGetDatum(collid));
1055 326 : if (!HeapTupleIsValid(tp))
1056 0 : elog(ERROR, "cache lookup failed for collation %u", collid);
1057 326 : collform = (Form_pg_collation) GETSTRUCT(tp);
1058 :
1059 326 : if (collform->collprovider == COLLPROVIDER_BUILTIN)
1060 52 : result = create_pg_locale_builtin(collid, context);
1061 274 : else if (collform->collprovider == COLLPROVIDER_ICU)
1062 184 : result = create_pg_locale_icu(collid, context);
1063 90 : else if (collform->collprovider == COLLPROVIDER_LIBC)
1064 90 : result = create_pg_locale_libc(collid, context);
1065 : else
1066 : /* shouldn't happen */
1067 0 : PGLOCALE_SUPPORT_ERROR(collform->collprovider);
1068 :
1069 320 : result->is_default = false;
1070 :
1071 : Assert((result->collate_is_c && result->collate == NULL) ||
1072 : (!result->collate_is_c && result->collate != NULL));
1073 :
1074 : Assert((result->ctype_is_c && result->ctype == NULL) ||
1075 : (!result->ctype_is_c && result->ctype != NULL));
1076 :
1077 320 : datum = SysCacheGetAttr(COLLOID, tp, Anum_pg_collation_collversion,
1078 : &isnull);
1079 320 : if (!isnull)
1080 : {
1081 : char *actual_versionstr;
1082 : char *collversionstr;
1083 :
1084 230 : collversionstr = TextDatumGetCString(datum);
1085 :
1086 230 : if (collform->collprovider == COLLPROVIDER_LIBC)
1087 0 : datum = SysCacheGetAttrNotNull(COLLOID, tp, Anum_pg_collation_collcollate);
1088 : else
1089 230 : datum = SysCacheGetAttrNotNull(COLLOID, tp, Anum_pg_collation_colllocale);
1090 :
1091 230 : actual_versionstr = get_collation_actual_version(collform->collprovider,
1092 230 : TextDatumGetCString(datum));
1093 230 : if (!actual_versionstr)
1094 : {
1095 : /*
1096 : * This could happen when specifying a version in CREATE COLLATION
1097 : * but the provider does not support versioning, or manually
1098 : * creating a mess in the catalogs.
1099 : */
1100 0 : ereport(ERROR,
1101 : (errmsg("collation \"%s\" has no actual version, but a version was recorded",
1102 : NameStr(collform->collname))));
1103 : }
1104 :
1105 230 : if (strcmp(actual_versionstr, collversionstr) != 0)
1106 0 : ereport(WARNING,
1107 : (errmsg("collation \"%s\" has version mismatch",
1108 : NameStr(collform->collname)),
1109 : errdetail("The collation in the database was created using version %s, "
1110 : "but the operating system provides version %s.",
1111 : collversionstr, actual_versionstr),
1112 : errhint("Rebuild all objects affected by this collation and run "
1113 : "ALTER COLLATION %s REFRESH VERSION, "
1114 : "or build PostgreSQL with the right library version.",
1115 : quote_qualified_identifier(get_namespace_name(collform->collnamespace),
1116 : NameStr(collform->collname)))));
1117 : }
1118 :
1119 320 : ReleaseSysCache(tp);
1120 :
1121 320 : return result;
1122 : }
1123 :
1124 : /*
1125 : * Initialize default_locale with database locale settings.
1126 : */
1127 : void
1128 33634 : init_database_collation(void)
1129 : {
1130 : HeapTuple tup;
1131 : Form_pg_database dbform;
1132 : pg_locale_t result;
1133 :
1134 : Assert(default_locale == NULL);
1135 :
1136 : /* Fetch our pg_database row normally, via syscache */
1137 33634 : tup = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(MyDatabaseId));
1138 33634 : if (!HeapTupleIsValid(tup))
1139 0 : elog(ERROR, "cache lookup failed for database %u", MyDatabaseId);
1140 33634 : dbform = (Form_pg_database) GETSTRUCT(tup);
1141 :
1142 33634 : if (dbform->datlocprovider == COLLPROVIDER_BUILTIN)
1143 1784 : result = create_pg_locale_builtin(DEFAULT_COLLATION_OID,
1144 : TopMemoryContext);
1145 31850 : else if (dbform->datlocprovider == COLLPROVIDER_ICU)
1146 26 : result = create_pg_locale_icu(DEFAULT_COLLATION_OID,
1147 : TopMemoryContext);
1148 31824 : else if (dbform->datlocprovider == COLLPROVIDER_LIBC)
1149 31824 : result = create_pg_locale_libc(DEFAULT_COLLATION_OID,
1150 : TopMemoryContext);
1151 : else
1152 : /* shouldn't happen */
1153 0 : PGLOCALE_SUPPORT_ERROR(dbform->datlocprovider);
1154 :
1155 33630 : result->is_default = true;
1156 :
1157 : Assert((result->collate_is_c && result->collate == NULL) ||
1158 : (!result->collate_is_c && result->collate != NULL));
1159 :
1160 : Assert((result->ctype_is_c && result->ctype == NULL) ||
1161 : (!result->ctype_is_c && result->ctype != NULL));
1162 :
1163 33630 : ReleaseSysCache(tup);
1164 :
1165 33630 : default_locale = result;
1166 33630 : }
1167 :
1168 : /*
1169 : * Get database default locale.
1170 : */
1171 : pg_locale_t
1172 3091552 : pg_database_locale(void)
1173 : {
1174 3091552 : return pg_newlocale_from_collation(DEFAULT_COLLATION_OID);
1175 : }
1176 :
1177 : /*
1178 : * Create a pg_locale_t from a collation OID. Results are cached for the
1179 : * lifetime of the backend. Thus, do not free the result with freelocale().
1180 : *
1181 : * For simplicity, we always generate COLLATE + CTYPE even though we
1182 : * might only need one of them. Since this is called only once per session,
1183 : * it shouldn't cost much.
1184 : */
1185 : pg_locale_t
1186 32096808 : pg_newlocale_from_collation(Oid collid)
1187 : {
1188 : collation_cache_entry *cache_entry;
1189 : bool found;
1190 :
1191 32096808 : if (collid == DEFAULT_COLLATION_OID)
1192 27843054 : return default_locale;
1193 :
1194 : /*
1195 : * Some callers expect C_COLLATION_OID to succeed even without catalog
1196 : * access.
1197 : */
1198 4253754 : if (collid == C_COLLATION_OID)
1199 4218076 : return &c_locale;
1200 :
1201 35678 : if (!OidIsValid(collid))
1202 0 : elog(ERROR, "cache lookup failed for collation %u", collid);
1203 :
1204 35678 : AssertCouldGetRelation();
1205 :
1206 35678 : if (last_collation_cache_oid == collid)
1207 34644 : return last_collation_cache_locale;
1208 :
1209 1034 : if (CollationCache == NULL)
1210 : {
1211 78 : CollationCacheContext = AllocSetContextCreate(TopMemoryContext,
1212 : "collation cache",
1213 : ALLOCSET_DEFAULT_SIZES);
1214 78 : CollationCache = collation_cache_create(CollationCacheContext,
1215 : 16, NULL);
1216 : }
1217 :
1218 1034 : cache_entry = collation_cache_insert(CollationCache, collid, &found);
1219 1034 : if (!found)
1220 : {
1221 : /*
1222 : * Make sure cache entry is marked invalid, in case we fail before
1223 : * setting things.
1224 : */
1225 326 : cache_entry->locale = NULL;
1226 : }
1227 :
1228 1034 : if (cache_entry->locale == NULL)
1229 : {
1230 326 : cache_entry->locale = create_pg_locale(collid, CollationCacheContext);
1231 : }
1232 :
1233 1028 : last_collation_cache_oid = collid;
1234 1028 : last_collation_cache_locale = cache_entry->locale;
1235 :
1236 1028 : return cache_entry->locale;
1237 : }
1238 :
1239 : /*
1240 : * Get provider-specific collation version string for the given collation from
1241 : * the operating system/library.
1242 : */
1243 : char *
1244 110790 : get_collation_actual_version(char collprovider, const char *collcollate)
1245 : {
1246 110790 : char *collversion = NULL;
1247 :
1248 110790 : if (collprovider == COLLPROVIDER_BUILTIN)
1249 1904 : collversion = get_collation_actual_version_builtin(collcollate);
1250 : #ifdef USE_ICU
1251 108886 : else if (collprovider == COLLPROVIDER_ICU)
1252 77840 : collversion = get_collation_actual_version_icu(collcollate);
1253 : #endif
1254 31046 : else if (collprovider == COLLPROVIDER_LIBC)
1255 31046 : collversion = get_collation_actual_version_libc(collcollate);
1256 :
1257 110790 : return collversion;
1258 : }
1259 :
1260 : /* lowercasing/casefolding in C locale */
1261 : static size_t
1262 0 : strlower_c(char *dst, size_t dstsize, const char *src, ssize_t srclen)
1263 : {
1264 : int i;
1265 :
1266 0 : srclen = (srclen >= 0) ? srclen : strlen(src);
1267 0 : for (i = 0; i < srclen && i < dstsize; i++)
1268 0 : dst[i] = pg_ascii_tolower(src[i]);
1269 0 : if (i < dstsize)
1270 0 : dst[i] = '\0';
1271 0 : return srclen;
1272 : }
1273 :
1274 : /* titlecasing in C locale */
1275 : static size_t
1276 0 : strtitle_c(char *dst, size_t dstsize, const char *src, ssize_t srclen)
1277 : {
1278 0 : bool wasalnum = false;
1279 : int i;
1280 :
1281 0 : srclen = (srclen >= 0) ? srclen : strlen(src);
1282 0 : for (i = 0; i < srclen && i < dstsize; i++)
1283 : {
1284 0 : char c = src[i];
1285 :
1286 0 : if (wasalnum)
1287 0 : dst[i] = pg_ascii_tolower(c);
1288 : else
1289 0 : dst[i] = pg_ascii_toupper(c);
1290 :
1291 0 : wasalnum = ((c >= '0' && c <= '9') ||
1292 0 : (c >= 'A' && c <= 'Z') ||
1293 0 : (c >= 'a' && c <= 'z'));
1294 : }
1295 0 : if (i < dstsize)
1296 0 : dst[i] = '\0';
1297 0 : return srclen;
1298 : }
1299 :
1300 : /* uppercasing in C locale */
1301 : static size_t
1302 0 : strupper_c(char *dst, size_t dstsize, const char *src, ssize_t srclen)
1303 : {
1304 : int i;
1305 :
1306 0 : srclen = (srclen >= 0) ? srclen : strlen(src);
1307 0 : for (i = 0; i < srclen && i < dstsize; i++)
1308 0 : dst[i] = pg_ascii_toupper(src[i]);
1309 0 : if (i < dstsize)
1310 0 : dst[i] = '\0';
1311 0 : return srclen;
1312 : }
1313 :
1314 : size_t
1315 437576 : pg_strlower(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1316 : pg_locale_t locale)
1317 : {
1318 437576 : if (locale->ctype == NULL)
1319 0 : return strlower_c(dst, dstsize, src, srclen);
1320 : else
1321 437576 : return locale->ctype->strlower(dst, dstsize, src, srclen, locale);
1322 : }
1323 :
1324 : size_t
1325 232 : pg_strtitle(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1326 : pg_locale_t locale)
1327 : {
1328 232 : if (locale->ctype == NULL)
1329 0 : return strtitle_c(dst, dstsize, src, srclen);
1330 : else
1331 232 : return locale->ctype->strtitle(dst, dstsize, src, srclen, locale);
1332 : }
1333 :
1334 : size_t
1335 1036952 : pg_strupper(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1336 : pg_locale_t locale)
1337 : {
1338 1036952 : if (locale->ctype == NULL)
1339 0 : return strupper_c(dst, dstsize, src, srclen);
1340 : else
1341 1036952 : return locale->ctype->strupper(dst, dstsize, src, srclen, locale);
1342 : }
1343 :
1344 : size_t
1345 24 : pg_strfold(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1346 : pg_locale_t locale)
1347 : {
1348 : /* in the C locale, casefolding is the same as lowercasing */
1349 24 : if (locale->ctype == NULL)
1350 0 : return strlower_c(dst, dstsize, src, srclen);
1351 : else
1352 24 : return locale->ctype->strfold(dst, dstsize, src, srclen, locale);
1353 : }
1354 :
1355 : /*
1356 : * pg_strcoll
1357 : *
1358 : * Like pg_strncoll for NUL-terminated input strings.
1359 : */
1360 : int
1361 25755976 : pg_strcoll(const char *arg1, const char *arg2, pg_locale_t locale)
1362 : {
1363 25755976 : return locale->collate->strncoll(arg1, -1, arg2, -1, locale);
1364 : }
1365 :
1366 : /*
1367 : * pg_strncoll
1368 : *
1369 : * Call ucol_strcollUTF8(), ucol_strcoll(), strcoll_l() or wcscoll_l() as
1370 : * appropriate for the given locale, platform, and database encoding. If the
1371 : * locale is not specified, use the database collation.
1372 : *
1373 : * The input strings must be encoded in the database encoding. If an input
1374 : * string is NUL-terminated, its length may be specified as -1.
1375 : *
1376 : * The caller is responsible for breaking ties if the collation is
1377 : * deterministic; this maintains consistency with pg_strnxfrm(), which cannot
1378 : * easily account for deterministic collations.
1379 : */
1380 : int
1381 4448204 : pg_strncoll(const char *arg1, ssize_t len1, const char *arg2, ssize_t len2,
1382 : pg_locale_t locale)
1383 : {
1384 4448204 : return locale->collate->strncoll(arg1, len1, arg2, len2, locale);
1385 : }
1386 :
1387 : /*
1388 : * Return true if the collation provider supports pg_strxfrm() and
1389 : * pg_strnxfrm(); otherwise false.
1390 : *
1391 : *
1392 : * No similar problem is known for the ICU provider.
1393 : */
1394 : bool
1395 45658 : pg_strxfrm_enabled(pg_locale_t locale)
1396 : {
1397 : /*
1398 : * locale->collate->strnxfrm is still a required method, even if it may
1399 : * have the wrong behavior, because the planner uses it for estimates in
1400 : * some cases.
1401 : */
1402 45658 : return locale->collate->strxfrm_is_safe;
1403 : }
1404 :
1405 : /*
1406 : * pg_strxfrm
1407 : *
1408 : * Like pg_strnxfrm for a NUL-terminated input string.
1409 : */
1410 : size_t
1411 144 : pg_strxfrm(char *dest, const char *src, size_t destsize, pg_locale_t locale)
1412 : {
1413 144 : return locale->collate->strnxfrm(dest, destsize, src, -1, locale);
1414 : }
1415 :
1416 : /*
1417 : * pg_strnxfrm
1418 : *
1419 : * Transforms 'src' to a nul-terminated string stored in 'dest' such that
1420 : * ordinary strcmp() on transformed strings is equivalent to pg_strcoll() on
1421 : * untransformed strings.
1422 : *
1423 : * The input string must be encoded in the database encoding. If the input
1424 : * string is NUL-terminated, its length may be specified as -1. If 'destsize'
1425 : * is zero, 'dest' may be NULL.
1426 : *
1427 : * Not all providers support pg_strnxfrm() safely. The caller should check
1428 : * pg_strxfrm_enabled() first, otherwise this function may return wrong
1429 : * results or an error.
1430 : *
1431 : * Returns the number of bytes needed (or more) to store the transformed
1432 : * string, excluding the terminating nul byte. If the value returned is
1433 : * 'destsize' or greater, the resulting contents of 'dest' are undefined.
1434 : */
1435 : size_t
1436 5748 : pg_strnxfrm(char *dest, size_t destsize, const char *src, ssize_t srclen,
1437 : pg_locale_t locale)
1438 : {
1439 5748 : return locale->collate->strnxfrm(dest, destsize, src, srclen, locale);
1440 : }
1441 :
1442 : /*
1443 : * Return true if the collation provider supports pg_strxfrm_prefix() and
1444 : * pg_strnxfrm_prefix(); otherwise false.
1445 : */
1446 : bool
1447 1668 : pg_strxfrm_prefix_enabled(pg_locale_t locale)
1448 : {
1449 1668 : return (locale->collate->strnxfrm_prefix != NULL);
1450 : }
1451 :
1452 : /*
1453 : * pg_strxfrm_prefix
1454 : *
1455 : * Like pg_strnxfrm_prefix for a NUL-terminated input string.
1456 : */
1457 : size_t
1458 1668 : pg_strxfrm_prefix(char *dest, const char *src, size_t destsize,
1459 : pg_locale_t locale)
1460 : {
1461 1668 : return locale->collate->strnxfrm_prefix(dest, destsize, src, -1, locale);
1462 : }
1463 :
1464 : /*
1465 : * pg_strnxfrm_prefix
1466 : *
1467 : * Transforms 'src' to a byte sequence stored in 'dest' such that ordinary
1468 : * memcmp() on the byte sequence is equivalent to pg_strncoll() on
1469 : * untransformed strings. The result is not nul-terminated.
1470 : *
1471 : * The input string must be encoded in the database encoding. If the input
1472 : * string is NUL-terminated, its length may be specified as -1.
1473 : *
1474 : * Not all providers support pg_strnxfrm_prefix() safely. The caller should
1475 : * check pg_strxfrm_prefix_enabled() first, otherwise this function may return
1476 : * wrong results or an error.
1477 : *
1478 : * If destsize is not large enough to hold the resulting byte sequence, stores
1479 : * only the first destsize bytes in 'dest'. Returns the number of bytes
1480 : * actually copied to 'dest'.
1481 : */
1482 : size_t
1483 0 : pg_strnxfrm_prefix(char *dest, size_t destsize, const char *src,
1484 : ssize_t srclen, pg_locale_t locale)
1485 : {
1486 0 : return locale->collate->strnxfrm_prefix(dest, destsize, src, srclen, locale);
1487 : }
1488 :
1489 : bool
1490 37132 : pg_iswdigit(pg_wchar wc, pg_locale_t locale)
1491 : {
1492 37132 : if (locale->ctype == NULL)
1493 0 : return (wc <= (pg_wchar) 127 &&
1494 0 : (pg_char_properties[wc] & PG_ISDIGIT));
1495 : else
1496 37132 : return locale->ctype->wc_isdigit(wc, locale);
1497 : }
1498 :
1499 : bool
1500 104140 : pg_iswalpha(pg_wchar wc, pg_locale_t locale)
1501 : {
1502 104140 : if (locale->ctype == NULL)
1503 0 : return (wc <= (pg_wchar) 127 &&
1504 0 : (pg_char_properties[wc] & PG_ISALPHA));
1505 : else
1506 104140 : return locale->ctype->wc_isalpha(wc, locale);
1507 : }
1508 :
1509 : bool
1510 2804958 : pg_iswalnum(pg_wchar wc, pg_locale_t locale)
1511 : {
1512 2804958 : if (locale->ctype == NULL)
1513 0 : return (wc <= (pg_wchar) 127 &&
1514 0 : (pg_char_properties[wc] & PG_ISALNUM));
1515 : else
1516 2804958 : return locale->ctype->wc_isalnum(wc, locale);
1517 : }
1518 :
1519 : bool
1520 0 : pg_iswupper(pg_wchar wc, pg_locale_t locale)
1521 : {
1522 0 : if (locale->ctype == NULL)
1523 0 : return (wc <= (pg_wchar) 127 &&
1524 0 : (pg_char_properties[wc] & PG_ISUPPER));
1525 : else
1526 0 : return locale->ctype->wc_isupper(wc, locale);
1527 : }
1528 :
1529 : bool
1530 0 : pg_iswlower(pg_wchar wc, pg_locale_t locale)
1531 : {
1532 0 : if (locale->ctype == NULL)
1533 0 : return (wc <= (pg_wchar) 127 &&
1534 0 : (pg_char_properties[wc] & PG_ISLOWER));
1535 : else
1536 0 : return locale->ctype->wc_islower(wc, locale);
1537 : }
1538 :
1539 : bool
1540 0 : pg_iswgraph(pg_wchar wc, pg_locale_t locale)
1541 : {
1542 0 : if (locale->ctype == NULL)
1543 0 : return (wc <= (pg_wchar) 127 &&
1544 0 : (pg_char_properties[wc] & PG_ISGRAPH));
1545 : else
1546 0 : return locale->ctype->wc_isgraph(wc, locale);
1547 : }
1548 :
1549 : bool
1550 0 : pg_iswprint(pg_wchar wc, pg_locale_t locale)
1551 : {
1552 0 : if (locale->ctype == NULL)
1553 0 : return (wc <= (pg_wchar) 127 &&
1554 0 : (pg_char_properties[wc] & PG_ISPRINT));
1555 : else
1556 0 : return locale->ctype->wc_isprint(wc, locale);
1557 : }
1558 :
1559 : bool
1560 0 : pg_iswpunct(pg_wchar wc, pg_locale_t locale)
1561 : {
1562 0 : if (locale->ctype == NULL)
1563 0 : return (wc <= (pg_wchar) 127 &&
1564 0 : (pg_char_properties[wc] & PG_ISPUNCT));
1565 : else
1566 0 : return locale->ctype->wc_ispunct(wc, locale);
1567 : }
1568 :
1569 : bool
1570 678 : pg_iswspace(pg_wchar wc, pg_locale_t locale)
1571 : {
1572 678 : if (locale->ctype == NULL)
1573 0 : return (wc <= (pg_wchar) 127 &&
1574 0 : (pg_char_properties[wc] & PG_ISSPACE));
1575 : else
1576 678 : return locale->ctype->wc_isspace(wc, locale);
1577 : }
1578 :
1579 : bool
1580 18 : pg_iswxdigit(pg_wchar wc, pg_locale_t locale)
1581 : {
1582 18 : if (locale->ctype == NULL)
1583 0 : return (wc <= (pg_wchar) 127 &&
1584 0 : ((pg_char_properties[wc] & PG_ISDIGIT) ||
1585 0 : ((wc >= 'A' && wc <= 'F') ||
1586 0 : (wc >= 'a' && wc <= 'f'))));
1587 : else
1588 18 : return locale->ctype->wc_isxdigit(wc, locale);
1589 : }
1590 :
1591 : pg_wchar
1592 0 : pg_towupper(pg_wchar wc, pg_locale_t locale)
1593 : {
1594 0 : if (locale->ctype == NULL)
1595 : {
1596 0 : if (wc <= (pg_wchar) 127)
1597 0 : return pg_ascii_toupper((unsigned char) wc);
1598 0 : return wc;
1599 : }
1600 : else
1601 0 : return locale->ctype->wc_toupper(wc, locale);
1602 : }
1603 :
1604 : pg_wchar
1605 0 : pg_towlower(pg_wchar wc, pg_locale_t locale)
1606 : {
1607 0 : if (locale->ctype == NULL)
1608 : {
1609 0 : if (wc <= (pg_wchar) 127)
1610 0 : return pg_ascii_tolower((unsigned char) wc);
1611 0 : return wc;
1612 : }
1613 : else
1614 0 : return locale->ctype->wc_tolower(wc, locale);
1615 : }
1616 :
1617 : /*
1618 : * char_is_cased()
1619 : *
1620 : * Fuzzy test of whether the given char is case-varying or not. The argument
1621 : * is a single byte, so in a multibyte encoding, just assume any non-ASCII
1622 : * char is case-varying.
1623 : */
1624 : bool
1625 126 : char_is_cased(char ch, pg_locale_t locale)
1626 : {
1627 126 : if (locale->ctype == NULL)
1628 0 : return (ch >= 'A' && ch <= 'Z') || (ch >= 'a' && ch <= 'z');
1629 126 : return locale->ctype->char_is_cased(ch, locale);
1630 : }
1631 :
1632 : /*
1633 : * char_tolower_enabled()
1634 : *
1635 : * Does the provider support char_tolower()?
1636 : */
1637 : bool
1638 68504 : char_tolower_enabled(pg_locale_t locale)
1639 : {
1640 68504 : if (locale->ctype == NULL)
1641 0 : return true;
1642 68504 : return (locale->ctype->char_tolower != NULL);
1643 : }
1644 :
1645 : /*
1646 : * char_tolower()
1647 : *
1648 : * Convert char (single-byte encoding) to lowercase.
1649 : */
1650 : char
1651 0 : char_tolower(unsigned char ch, pg_locale_t locale)
1652 : {
1653 0 : if (locale->ctype == NULL)
1654 0 : return pg_ascii_tolower(ch);
1655 0 : return locale->ctype->char_tolower(ch, locale);
1656 : }
1657 :
1658 : /*
1659 : * Return required encoding ID for the given locale, or -1 if any encoding is
1660 : * valid for the locale.
1661 : */
1662 : int
1663 1960 : builtin_locale_encoding(const char *locale)
1664 : {
1665 1960 : if (strcmp(locale, "C") == 0)
1666 76 : return -1;
1667 1884 : else if (strcmp(locale, "C.UTF-8") == 0)
1668 1854 : return PG_UTF8;
1669 30 : else if (strcmp(locale, "PG_UNICODE_FAST") == 0)
1670 30 : return PG_UTF8;
1671 :
1672 :
1673 0 : ereport(ERROR,
1674 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1675 : errmsg("invalid locale name \"%s\" for builtin provider",
1676 : locale)));
1677 :
1678 : return 0; /* keep compiler quiet */
1679 : }
1680 :
1681 :
1682 : /*
1683 : * Validate the locale and encoding combination, and return the canonical form
1684 : * of the locale name.
1685 : */
1686 : const char *
1687 1944 : builtin_validate_locale(int encoding, const char *locale)
1688 : {
1689 1944 : const char *canonical_name = NULL;
1690 : int required_encoding;
1691 :
1692 1944 : if (strcmp(locale, "C") == 0)
1693 64 : canonical_name = "C";
1694 1880 : else if (strcmp(locale, "C.UTF-8") == 0 || strcmp(locale, "C.UTF8") == 0)
1695 1840 : canonical_name = "C.UTF-8";
1696 40 : else if (strcmp(locale, "PG_UNICODE_FAST") == 0)
1697 22 : canonical_name = "PG_UNICODE_FAST";
1698 :
1699 1944 : if (!canonical_name)
1700 18 : ereport(ERROR,
1701 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1702 : errmsg("invalid locale name \"%s\" for builtin provider",
1703 : locale)));
1704 :
1705 1926 : required_encoding = builtin_locale_encoding(canonical_name);
1706 1926 : if (required_encoding >= 0 && encoding != required_encoding)
1707 2 : ereport(ERROR,
1708 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1709 : errmsg("encoding \"%s\" does not match locale \"%s\"",
1710 : pg_encoding_to_char(encoding), locale)));
1711 :
1712 1924 : return canonical_name;
1713 : }
1714 :
1715 :
1716 :
1717 : /*
1718 : * Return the BCP47 language tag representation of the requested locale.
1719 : *
1720 : * This function should be called before passing the string to ucol_open(),
1721 : * because conversion to a language tag also performs "level 2
1722 : * canonicalization". In addition to producing a consistent format, level 2
1723 : * canonicalization is able to more accurately interpret different input
1724 : * locale string formats, such as POSIX and .NET IDs.
1725 : */
1726 : char *
1727 77534 : icu_language_tag(const char *loc_str, int elevel)
1728 : {
1729 : #ifdef USE_ICU
1730 : UErrorCode status;
1731 : char *langtag;
1732 77534 : size_t buflen = 32; /* arbitrary starting buffer size */
1733 77534 : const bool strict = true;
1734 :
1735 : /*
1736 : * A BCP47 language tag doesn't have a clearly-defined upper limit (cf.
1737 : * RFC5646 section 4.4). Additionally, in older ICU versions,
1738 : * uloc_toLanguageTag() doesn't always return the ultimate length on the
1739 : * first call, necessitating a loop.
1740 : */
1741 77534 : langtag = palloc(buflen);
1742 : while (true)
1743 : {
1744 77534 : status = U_ZERO_ERROR;
1745 77534 : uloc_toLanguageTag(loc_str, langtag, buflen, strict, &status);
1746 :
1747 : /* try again if the buffer is not large enough */
1748 77534 : if ((status == U_BUFFER_OVERFLOW_ERROR ||
1749 77534 : status == U_STRING_NOT_TERMINATED_WARNING) &&
1750 : buflen < MaxAllocSize)
1751 : {
1752 0 : buflen = Min(buflen * 2, MaxAllocSize);
1753 0 : langtag = repalloc(langtag, buflen);
1754 0 : continue;
1755 : }
1756 :
1757 77534 : break;
1758 : }
1759 :
1760 77534 : if (U_FAILURE(status))
1761 : {
1762 18 : pfree(langtag);
1763 :
1764 18 : if (elevel > 0)
1765 14 : ereport(elevel,
1766 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1767 : errmsg("could not convert locale name \"%s\" to language tag: %s",
1768 : loc_str, u_errorName(status))));
1769 12 : return NULL;
1770 : }
1771 :
1772 77516 : return langtag;
1773 : #else /* not USE_ICU */
1774 : ereport(ERROR,
1775 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1776 : errmsg("ICU is not supported in this build")));
1777 : return NULL; /* keep compiler quiet */
1778 : #endif /* not USE_ICU */
1779 : }
1780 :
1781 : /*
1782 : * Perform best-effort check that the locale is a valid one.
1783 : */
1784 : void
1785 166 : icu_validate_locale(const char *loc_str)
1786 : {
1787 : #ifdef USE_ICU
1788 : UCollator *collator;
1789 : UErrorCode status;
1790 : char lang[ULOC_LANG_CAPACITY];
1791 166 : bool found = false;
1792 166 : int elevel = icu_validation_level;
1793 :
1794 : /* no validation */
1795 166 : if (elevel < 0)
1796 12 : return;
1797 :
1798 : /* downgrade to WARNING during pg_upgrade */
1799 154 : if (IsBinaryUpgrade && elevel > WARNING)
1800 0 : elevel = WARNING;
1801 :
1802 : /* validate that we can extract the language */
1803 154 : status = U_ZERO_ERROR;
1804 154 : uloc_getLanguage(loc_str, lang, ULOC_LANG_CAPACITY, &status);
1805 154 : if (U_FAILURE(status) || status == U_STRING_NOT_TERMINATED_WARNING)
1806 : {
1807 0 : ereport(elevel,
1808 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1809 : errmsg("could not get language from ICU locale \"%s\": %s",
1810 : loc_str, u_errorName(status)),
1811 : errhint("To disable ICU locale validation, set the parameter \"%s\" to \"%s\".",
1812 : "icu_validation_level", "disabled")));
1813 0 : return;
1814 : }
1815 :
1816 : /* check for special language name */
1817 154 : if (strcmp(lang, "") == 0 ||
1818 46 : strcmp(lang, "root") == 0 || strcmp(lang, "und") == 0)
1819 108 : found = true;
1820 :
1821 : /* search for matching language within ICU */
1822 15704 : for (int32_t i = 0; !found && i < uloc_countAvailable(); i++)
1823 : {
1824 15550 : const char *otherloc = uloc_getAvailable(i);
1825 : char otherlang[ULOC_LANG_CAPACITY];
1826 :
1827 15550 : status = U_ZERO_ERROR;
1828 15550 : uloc_getLanguage(otherloc, otherlang, ULOC_LANG_CAPACITY, &status);
1829 15550 : if (U_FAILURE(status) || status == U_STRING_NOT_TERMINATED_WARNING)
1830 0 : continue;
1831 :
1832 15550 : if (strcmp(lang, otherlang) == 0)
1833 32 : found = true;
1834 : }
1835 :
1836 154 : if (!found)
1837 14 : ereport(elevel,
1838 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1839 : errmsg("ICU locale \"%s\" has unknown language \"%s\"",
1840 : loc_str, lang),
1841 : errhint("To disable ICU locale validation, set the parameter \"%s\" to \"%s\".",
1842 : "icu_validation_level", "disabled")));
1843 :
1844 : /* check that it can be opened */
1845 148 : collator = pg_ucol_open(loc_str);
1846 140 : ucol_close(collator);
1847 : #else /* not USE_ICU */
1848 : /* could get here if a collation was created by a build with ICU */
1849 : ereport(ERROR,
1850 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1851 : errmsg("ICU is not supported in this build")));
1852 : #endif /* not USE_ICU */
1853 : }
|
