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/relcache.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 : /* GUC settings */
83 : char *locale_messages;
84 : char *locale_monetary;
85 : char *locale_numeric;
86 : char *locale_time;
87 :
88 : int icu_validation_level = WARNING;
89 :
90 : /*
91 : * lc_time localization cache.
92 : *
93 : * We use only the first 7 or 12 entries of these arrays. The last array
94 : * element is left as NULL for the convenience of outside code that wants
95 : * to sequentially scan these arrays.
96 : */
97 : char *localized_abbrev_days[7 + 1];
98 : char *localized_full_days[7 + 1];
99 : char *localized_abbrev_months[12 + 1];
100 : char *localized_full_months[12 + 1];
101 :
102 : /* is the databases's LC_CTYPE the C locale? */
103 : bool database_ctype_is_c = false;
104 :
105 : static pg_locale_t default_locale = NULL;
106 :
107 : /* indicates whether locale information cache is valid */
108 : static bool CurrentLocaleConvValid = false;
109 : static bool CurrentLCTimeValid = false;
110 :
111 : /* Cache for collation-related knowledge */
112 :
113 : typedef struct
114 : {
115 : Oid collid; /* hash key: pg_collation OID */
116 : pg_locale_t locale; /* locale_t struct, or 0 if not valid */
117 :
118 : /* needed for simplehash */
119 : uint32 hash;
120 : char status;
121 : } collation_cache_entry;
122 :
123 : #define SH_PREFIX collation_cache
124 : #define SH_ELEMENT_TYPE collation_cache_entry
125 : #define SH_KEY_TYPE Oid
126 : #define SH_KEY collid
127 : #define SH_HASH_KEY(tb, key) murmurhash32((uint32) key)
128 : #define SH_EQUAL(tb, a, b) (a == b)
129 : #define SH_GET_HASH(tb, a) a->hash
130 : #define SH_SCOPE static inline
131 : #define SH_STORE_HASH
132 : #define SH_DECLARE
133 : #define SH_DEFINE
134 : #include "lib/simplehash.h"
135 :
136 : static MemoryContext CollationCacheContext = NULL;
137 : static collation_cache_hash *CollationCache = NULL;
138 :
139 : /*
140 : * The collation cache is often accessed repeatedly for the same collation, so
141 : * remember the last one used.
142 : */
143 : static Oid last_collation_cache_oid = InvalidOid;
144 : static pg_locale_t last_collation_cache_locale = NULL;
145 :
146 : #if defined(WIN32) && defined(LC_MESSAGES)
147 : static char *IsoLocaleName(const char *);
148 : #endif
149 :
150 : /*
151 : * pg_perm_setlocale
152 : *
153 : * This wraps the libc function setlocale(), with two additions. First, when
154 : * changing LC_CTYPE, update gettext's encoding for the current message
155 : * domain. GNU gettext automatically tracks LC_CTYPE on most platforms, but
156 : * not on Windows. Second, if the operation is successful, the corresponding
157 : * LC_XXX environment variable is set to match. By setting the environment
158 : * variable, we ensure that any subsequent use of setlocale(..., "") will
159 : * preserve the settings made through this routine. Of course, LC_ALL must
160 : * also be unset to fully ensure that, but that has to be done elsewhere after
161 : * all the individual LC_XXX variables have been set correctly. (Thank you
162 : * Perl for making this kluge necessary.)
163 : */
164 : char *
165 71848 : pg_perm_setlocale(int category, const char *locale)
166 : {
167 : char *result;
168 : const char *envvar;
169 :
170 : #ifndef WIN32
171 71848 : result = setlocale(category, locale);
172 : #else
173 :
174 : /*
175 : * On Windows, setlocale(LC_MESSAGES) does not work, so just assume that
176 : * the given value is good and set it in the environment variables. We
177 : * must ignore attempts to set to "", which means "keep using the old
178 : * environment value".
179 : */
180 : #ifdef LC_MESSAGES
181 : if (category == LC_MESSAGES)
182 : {
183 : result = (char *) locale;
184 : if (locale == NULL || locale[0] == '\0')
185 : return result;
186 : }
187 : else
188 : #endif
189 : result = setlocale(category, locale);
190 : #endif /* WIN32 */
191 :
192 71848 : if (result == NULL)
193 0 : return result; /* fall out immediately on failure */
194 :
195 : /*
196 : * Use the right encoding in translated messages. Under ENABLE_NLS, let
197 : * pg_bind_textdomain_codeset() figure it out. Under !ENABLE_NLS, message
198 : * format strings are ASCII, but database-encoding strings may enter the
199 : * message via %s. This makes the overall message encoding equal to the
200 : * database encoding.
201 : */
202 71848 : if (category == LC_CTYPE)
203 : {
204 : static char save_lc_ctype[LOCALE_NAME_BUFLEN];
205 :
206 : /* copy setlocale() return value before callee invokes it again */
207 35328 : strlcpy(save_lc_ctype, result, sizeof(save_lc_ctype));
208 35328 : result = save_lc_ctype;
209 :
210 : #ifdef ENABLE_NLS
211 35328 : SetMessageEncoding(pg_bind_textdomain_codeset(textdomain(NULL)));
212 : #else
213 : SetMessageEncoding(GetDatabaseEncoding());
214 : #endif
215 : }
216 :
217 71848 : switch (category)
218 : {
219 3738 : case LC_COLLATE:
220 3738 : envvar = "LC_COLLATE";
221 3738 : break;
222 35328 : case LC_CTYPE:
223 35328 : envvar = "LC_CTYPE";
224 35328 : break;
225 : #ifdef LC_MESSAGES
226 21568 : case LC_MESSAGES:
227 21568 : envvar = "LC_MESSAGES";
228 : #ifdef WIN32
229 : result = IsoLocaleName(locale);
230 : if (result == NULL)
231 : result = (char *) locale;
232 : elog(DEBUG3, "IsoLocaleName() executed; locale: \"%s\"", result);
233 : #endif /* WIN32 */
234 21568 : break;
235 : #endif /* LC_MESSAGES */
236 3738 : case LC_MONETARY:
237 3738 : envvar = "LC_MONETARY";
238 3738 : break;
239 3738 : case LC_NUMERIC:
240 3738 : envvar = "LC_NUMERIC";
241 3738 : break;
242 3738 : case LC_TIME:
243 3738 : envvar = "LC_TIME";
244 3738 : break;
245 0 : default:
246 0 : elog(FATAL, "unrecognized LC category: %d", category);
247 : return NULL; /* keep compiler quiet */
248 : }
249 :
250 71848 : if (setenv(envvar, result, 1) != 0)
251 0 : return NULL;
252 :
253 71848 : return result;
254 : }
255 :
256 :
257 : /*
258 : * Is the locale name valid for the locale category?
259 : *
260 : * If successful, and canonname isn't NULL, a palloc'd copy of the locale's
261 : * canonical name is stored there. This is especially useful for figuring out
262 : * what locale name "" means (ie, the server environment value). (Actually,
263 : * it seems that on most implementations that's the only thing it's good for;
264 : * we could wish that setlocale gave back a canonically spelled version of
265 : * the locale name, but typically it doesn't.)
266 : */
267 : bool
268 68676 : check_locale(int category, const char *locale, char **canonname)
269 : {
270 : char *save;
271 : char *res;
272 :
273 : /* Don't let Windows' non-ASCII locale names in. */
274 68676 : if (!pg_is_ascii(locale))
275 : {
276 0 : ereport(WARNING,
277 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
278 : errmsg("locale name \"%s\" contains non-ASCII characters",
279 : locale)));
280 0 : return false;
281 : }
282 :
283 68676 : if (canonname)
284 1574 : *canonname = NULL; /* in case of failure */
285 :
286 68676 : save = setlocale(category, NULL);
287 68676 : if (!save)
288 0 : return false; /* won't happen, we hope */
289 :
290 : /* save may be pointing at a modifiable scratch variable, see above. */
291 68676 : save = pstrdup(save);
292 :
293 : /* set the locale with setlocale, to see if it accepts it. */
294 68676 : res = setlocale(category, locale);
295 :
296 : /* save canonical name if requested. */
297 68676 : if (res && canonname)
298 1570 : *canonname = pstrdup(res);
299 :
300 : /* restore old value. */
301 68676 : if (!setlocale(category, save))
302 0 : elog(WARNING, "failed to restore old locale \"%s\"", save);
303 68676 : pfree(save);
304 :
305 : /* Don't let Windows' non-ASCII locale names out. */
306 68676 : if (canonname && *canonname && !pg_is_ascii(*canonname))
307 : {
308 0 : ereport(WARNING,
309 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
310 : errmsg("locale name \"%s\" contains non-ASCII characters",
311 : *canonname)));
312 0 : pfree(*canonname);
313 0 : *canonname = NULL;
314 0 : return false;
315 : }
316 :
317 68676 : return (res != NULL);
318 : }
319 :
320 :
321 : /*
322 : * GUC check/assign hooks
323 : *
324 : * For most locale categories, the assign hook doesn't actually set the locale
325 : * permanently, just reset flags so that the next use will cache the
326 : * appropriate values. (See explanation at the top of this file.)
327 : *
328 : * Note: we accept value = "" as selecting the postmaster's environment
329 : * value, whatever it was (so long as the environment setting is legal).
330 : * This will have been locked down by an earlier call to pg_perm_setlocale.
331 : */
332 : bool
333 18004 : check_locale_monetary(char **newval, void **extra, GucSource source)
334 : {
335 18004 : return check_locale(LC_MONETARY, *newval, NULL);
336 : }
337 :
338 : void
339 17798 : assign_locale_monetary(const char *newval, void *extra)
340 : {
341 17798 : CurrentLocaleConvValid = false;
342 17798 : }
343 :
344 : bool
345 18010 : check_locale_numeric(char **newval, void **extra, GucSource source)
346 : {
347 18010 : return check_locale(LC_NUMERIC, *newval, NULL);
348 : }
349 :
350 : void
351 17810 : assign_locale_numeric(const char *newval, void *extra)
352 : {
353 17810 : CurrentLocaleConvValid = false;
354 17810 : }
355 :
356 : bool
357 18004 : check_locale_time(char **newval, void **extra, GucSource source)
358 : {
359 18004 : return check_locale(LC_TIME, *newval, NULL);
360 : }
361 :
362 : void
363 17798 : assign_locale_time(const char *newval, void *extra)
364 : {
365 17798 : CurrentLCTimeValid = false;
366 17798 : }
367 :
368 : /*
369 : * We allow LC_MESSAGES to actually be set globally.
370 : *
371 : * Note: we normally disallow value = "" because it wouldn't have consistent
372 : * semantics (it'd effectively just use the previous value). However, this
373 : * is the value passed for PGC_S_DEFAULT, so don't complain in that case,
374 : * not even if the attempted setting fails due to invalid environment value.
375 : * The idea there is just to accept the environment setting *if possible*
376 : * during startup, until we can read the proper value from postgresql.conf.
377 : */
378 : bool
379 18038 : check_locale_messages(char **newval, void **extra, GucSource source)
380 : {
381 18038 : if (**newval == '\0')
382 : {
383 4954 : if (source == PGC_S_DEFAULT)
384 4954 : return true;
385 : else
386 0 : return false;
387 : }
388 :
389 : /*
390 : * LC_MESSAGES category does not exist everywhere, but accept it anyway
391 : *
392 : * On Windows, we can't even check the value, so accept blindly
393 : */
394 : #if defined(LC_MESSAGES) && !defined(WIN32)
395 13084 : return check_locale(LC_MESSAGES, *newval, NULL);
396 : #else
397 : return true;
398 : #endif
399 : }
400 :
401 : void
402 17830 : assign_locale_messages(const char *newval, void *extra)
403 : {
404 : /*
405 : * LC_MESSAGES category does not exist everywhere, but accept it anyway.
406 : * We ignore failure, as per comment above.
407 : */
408 : #ifdef LC_MESSAGES
409 17830 : (void) pg_perm_setlocale(LC_MESSAGES, newval);
410 : #endif
411 17830 : }
412 :
413 :
414 : /*
415 : * Frees the malloced content of a struct lconv. (But not the struct
416 : * itself.) It's important that this not throw elog(ERROR).
417 : */
418 : static void
419 6 : free_struct_lconv(struct lconv *s)
420 : {
421 6 : free(s->decimal_point);
422 6 : free(s->thousands_sep);
423 6 : free(s->grouping);
424 6 : free(s->int_curr_symbol);
425 6 : free(s->currency_symbol);
426 6 : free(s->mon_decimal_point);
427 6 : free(s->mon_thousands_sep);
428 6 : free(s->mon_grouping);
429 6 : free(s->positive_sign);
430 6 : free(s->negative_sign);
431 6 : }
432 :
433 : /*
434 : * Check that all fields of a struct lconv (or at least, the ones we care
435 : * about) are non-NULL. The field list must match free_struct_lconv().
436 : */
437 : static bool
438 68 : struct_lconv_is_valid(struct lconv *s)
439 : {
440 68 : if (s->decimal_point == NULL)
441 0 : return false;
442 68 : if (s->thousands_sep == NULL)
443 0 : return false;
444 68 : if (s->grouping == NULL)
445 0 : return false;
446 68 : if (s->int_curr_symbol == NULL)
447 0 : return false;
448 68 : if (s->currency_symbol == NULL)
449 0 : return false;
450 68 : if (s->mon_decimal_point == NULL)
451 0 : return false;
452 68 : if (s->mon_thousands_sep == NULL)
453 0 : return false;
454 68 : if (s->mon_grouping == NULL)
455 0 : return false;
456 68 : if (s->positive_sign == NULL)
457 0 : return false;
458 68 : if (s->negative_sign == NULL)
459 0 : return false;
460 68 : return true;
461 : }
462 :
463 :
464 : /*
465 : * Convert the strdup'd string at *str from the specified encoding to the
466 : * database encoding.
467 : */
468 : static void
469 544 : db_encoding_convert(int encoding, char **str)
470 : {
471 : char *pstr;
472 : char *mstr;
473 :
474 : /* convert the string to the database encoding */
475 544 : pstr = pg_any_to_server(*str, strlen(*str), encoding);
476 544 : if (pstr == *str)
477 544 : return; /* no conversion happened */
478 :
479 : /* need it malloc'd not palloc'd */
480 0 : mstr = strdup(pstr);
481 0 : if (mstr == NULL)
482 0 : ereport(ERROR,
483 : (errcode(ERRCODE_OUT_OF_MEMORY),
484 : errmsg("out of memory")));
485 :
486 : /* replace old string */
487 0 : free(*str);
488 0 : *str = mstr;
489 :
490 0 : pfree(pstr);
491 : }
492 :
493 :
494 : /*
495 : * Return the POSIX lconv struct (contains number/money formatting
496 : * information) with locale information for all categories.
497 : */
498 : struct lconv *
499 2982 : PGLC_localeconv(void)
500 : {
501 : static struct lconv CurrentLocaleConv;
502 : static bool CurrentLocaleConvAllocated = false;
503 : struct lconv *extlconv;
504 : struct lconv tmp;
505 2982 : struct lconv worklconv = {0};
506 :
507 : /* Did we do it already? */
508 2982 : if (CurrentLocaleConvValid)
509 2914 : return &CurrentLocaleConv;
510 :
511 : /* Free any already-allocated storage */
512 68 : if (CurrentLocaleConvAllocated)
513 : {
514 6 : free_struct_lconv(&CurrentLocaleConv);
515 6 : CurrentLocaleConvAllocated = false;
516 : }
517 :
518 : /*
519 : * Use thread-safe method of obtaining a copy of lconv from the operating
520 : * system.
521 : */
522 68 : if (pg_localeconv_r(locale_monetary,
523 : locale_numeric,
524 : &tmp) != 0)
525 0 : elog(ERROR,
526 : "could not get lconv for LC_MONETARY = \"%s\", LC_NUMERIC = \"%s\": %m",
527 : locale_monetary, locale_numeric);
528 :
529 : /* Must copy data now so we can re-encode it. */
530 68 : extlconv = &tmp;
531 68 : worklconv.decimal_point = strdup(extlconv->decimal_point);
532 68 : worklconv.thousands_sep = strdup(extlconv->thousands_sep);
533 68 : worklconv.grouping = strdup(extlconv->grouping);
534 68 : worklconv.int_curr_symbol = strdup(extlconv->int_curr_symbol);
535 68 : worklconv.currency_symbol = strdup(extlconv->currency_symbol);
536 68 : worklconv.mon_decimal_point = strdup(extlconv->mon_decimal_point);
537 68 : worklconv.mon_thousands_sep = strdup(extlconv->mon_thousands_sep);
538 68 : worklconv.mon_grouping = strdup(extlconv->mon_grouping);
539 68 : worklconv.positive_sign = strdup(extlconv->positive_sign);
540 68 : worklconv.negative_sign = strdup(extlconv->negative_sign);
541 : /* Copy scalar fields as well */
542 68 : worklconv.int_frac_digits = extlconv->int_frac_digits;
543 68 : worklconv.frac_digits = extlconv->frac_digits;
544 68 : worklconv.p_cs_precedes = extlconv->p_cs_precedes;
545 68 : worklconv.p_sep_by_space = extlconv->p_sep_by_space;
546 68 : worklconv.n_cs_precedes = extlconv->n_cs_precedes;
547 68 : worklconv.n_sep_by_space = extlconv->n_sep_by_space;
548 68 : worklconv.p_sign_posn = extlconv->p_sign_posn;
549 68 : worklconv.n_sign_posn = extlconv->n_sign_posn;
550 :
551 : /* Free the contents of the object populated by pg_localeconv_r(). */
552 68 : pg_localeconv_free(&tmp);
553 :
554 : /* If any of the preceding strdup calls failed, complain now. */
555 68 : if (!struct_lconv_is_valid(&worklconv))
556 0 : ereport(ERROR,
557 : (errcode(ERRCODE_OUT_OF_MEMORY),
558 : errmsg("out of memory")));
559 :
560 68 : PG_TRY();
561 : {
562 : int encoding;
563 :
564 : /*
565 : * Now we must perform encoding conversion from whatever's associated
566 : * with the locales into the database encoding. If we can't identify
567 : * the encoding implied by LC_NUMERIC or LC_MONETARY (ie we get -1),
568 : * use PG_SQL_ASCII, which will result in just validating that the
569 : * strings are OK in the database encoding.
570 : */
571 68 : encoding = pg_get_encoding_from_locale(locale_numeric, true);
572 68 : if (encoding < 0)
573 0 : encoding = PG_SQL_ASCII;
574 :
575 68 : db_encoding_convert(encoding, &worklconv.decimal_point);
576 68 : db_encoding_convert(encoding, &worklconv.thousands_sep);
577 : /* grouping is not text and does not require conversion */
578 :
579 68 : encoding = pg_get_encoding_from_locale(locale_monetary, true);
580 68 : if (encoding < 0)
581 0 : encoding = PG_SQL_ASCII;
582 :
583 68 : db_encoding_convert(encoding, &worklconv.int_curr_symbol);
584 68 : db_encoding_convert(encoding, &worklconv.currency_symbol);
585 68 : db_encoding_convert(encoding, &worklconv.mon_decimal_point);
586 68 : db_encoding_convert(encoding, &worklconv.mon_thousands_sep);
587 : /* mon_grouping is not text and does not require conversion */
588 68 : db_encoding_convert(encoding, &worklconv.positive_sign);
589 68 : db_encoding_convert(encoding, &worklconv.negative_sign);
590 : }
591 0 : PG_CATCH();
592 : {
593 0 : free_struct_lconv(&worklconv);
594 0 : PG_RE_THROW();
595 : }
596 68 : PG_END_TRY();
597 :
598 : /*
599 : * Everything is good, so save the results.
600 : */
601 68 : CurrentLocaleConv = worklconv;
602 68 : CurrentLocaleConvAllocated = true;
603 68 : CurrentLocaleConvValid = true;
604 68 : return &CurrentLocaleConv;
605 : }
606 :
607 : #ifdef WIN32
608 : /*
609 : * On Windows, strftime() returns its output in encoding CP_ACP (the default
610 : * operating system codepage for the computer), which is likely different
611 : * from SERVER_ENCODING. This is especially important in Japanese versions
612 : * of Windows which will use SJIS encoding, which we don't support as a
613 : * server encoding.
614 : *
615 : * So, instead of using strftime(), use wcsftime() to return the value in
616 : * wide characters (internally UTF16) and then convert to UTF8, which we
617 : * know how to handle directly.
618 : *
619 : * Note that this only affects the calls to strftime() in this file, which are
620 : * used to get the locale-aware strings. Other parts of the backend use
621 : * pg_strftime(), which isn't locale-aware and does not need to be replaced.
622 : */
623 : static size_t
624 : strftime_l_win32(char *dst, size_t dstlen,
625 : const char *format, const struct tm *tm, locale_t locale)
626 : {
627 : size_t len;
628 : wchar_t wformat[8]; /* formats used below need 3 chars */
629 : wchar_t wbuf[MAX_L10N_DATA];
630 :
631 : /*
632 : * Get a wchar_t version of the format string. We only actually use
633 : * plain-ASCII formats in this file, so we can say that they're UTF8.
634 : */
635 : len = MultiByteToWideChar(CP_UTF8, 0, format, -1,
636 : wformat, lengthof(wformat));
637 : if (len == 0)
638 : elog(ERROR, "could not convert format string from UTF-8: error code %lu",
639 : GetLastError());
640 :
641 : len = _wcsftime_l(wbuf, MAX_L10N_DATA, wformat, tm, locale);
642 : if (len == 0)
643 : {
644 : /*
645 : * wcsftime failed, possibly because the result would not fit in
646 : * MAX_L10N_DATA. Return 0 with the contents of dst unspecified.
647 : */
648 : return 0;
649 : }
650 :
651 : len = WideCharToMultiByte(CP_UTF8, 0, wbuf, len, dst, dstlen - 1,
652 : NULL, NULL);
653 : if (len == 0)
654 : elog(ERROR, "could not convert string to UTF-8: error code %lu",
655 : GetLastError());
656 :
657 : dst[len] = '\0';
658 :
659 : return len;
660 : }
661 :
662 : /* redefine strftime_l() */
663 : #define strftime_l(a,b,c,d,e) strftime_l_win32(a,b,c,d,e)
664 : #endif /* WIN32 */
665 :
666 : /*
667 : * Subroutine for cache_locale_time().
668 : * Convert the given string from encoding "encoding" to the database
669 : * encoding, and store the result at *dst, replacing any previous value.
670 : */
671 : static void
672 1748 : cache_single_string(char **dst, const char *src, int encoding)
673 : {
674 : char *ptr;
675 : char *olddst;
676 :
677 : /* Convert the string to the database encoding, or validate it's OK */
678 1748 : ptr = pg_any_to_server(src, strlen(src), encoding);
679 :
680 : /* Store the string in long-lived storage, replacing any previous value */
681 1748 : olddst = *dst;
682 1748 : *dst = MemoryContextStrdup(TopMemoryContext, ptr);
683 1748 : if (olddst)
684 0 : pfree(olddst);
685 :
686 : /* Might as well clean up any palloc'd conversion result, too */
687 1748 : if (ptr != src)
688 0 : pfree(ptr);
689 1748 : }
690 :
691 : /*
692 : * Update the lc_time localization cache variables if needed.
693 : */
694 : void
695 49546 : cache_locale_time(void)
696 : {
697 : char buf[(2 * 7 + 2 * 12) * MAX_L10N_DATA];
698 : char *bufptr;
699 : time_t timenow;
700 : struct tm *timeinfo;
701 : struct tm timeinfobuf;
702 49546 : bool strftimefail = false;
703 : int encoding;
704 : int i;
705 : locale_t locale;
706 :
707 : /* did we do this already? */
708 49546 : if (CurrentLCTimeValid)
709 49500 : return;
710 :
711 46 : elog(DEBUG3, "cache_locale_time() executed; locale: \"%s\"", locale_time);
712 :
713 46 : errno = ENOENT;
714 : #ifdef WIN32
715 : locale = _create_locale(LC_ALL, locale_time);
716 : if (locale == (locale_t) 0)
717 : _dosmaperr(GetLastError());
718 : #else
719 46 : locale = newlocale(LC_ALL_MASK, locale_time, (locale_t) 0);
720 : #endif
721 46 : if (!locale)
722 0 : report_newlocale_failure(locale_time);
723 :
724 : /* We use times close to current time as data for strftime(). */
725 46 : timenow = time(NULL);
726 46 : timeinfo = gmtime_r(&timenow, &timeinfobuf);
727 :
728 : /* Store the strftime results in MAX_L10N_DATA-sized portions of buf[] */
729 46 : bufptr = buf;
730 :
731 : /*
732 : * MAX_L10N_DATA is sufficient buffer space for every known locale, and
733 : * POSIX defines no strftime() errors. (Buffer space exhaustion is not an
734 : * error.) An implementation might report errors (e.g. ENOMEM) by
735 : * returning 0 (or, less plausibly, a negative value) and setting errno.
736 : * Report errno just in case the implementation did that, but clear it in
737 : * advance of the calls so we don't emit a stale, unrelated errno.
738 : */
739 46 : errno = 0;
740 :
741 : /* localized days */
742 368 : for (i = 0; i < 7; i++)
743 : {
744 322 : timeinfo->tm_wday = i;
745 322 : if (strftime_l(bufptr, MAX_L10N_DATA, "%a", timeinfo, locale) <= 0)
746 0 : strftimefail = true;
747 322 : bufptr += MAX_L10N_DATA;
748 322 : if (strftime_l(bufptr, MAX_L10N_DATA, "%A", timeinfo, locale) <= 0)
749 0 : strftimefail = true;
750 322 : bufptr += MAX_L10N_DATA;
751 : }
752 :
753 : /* localized months */
754 598 : for (i = 0; i < 12; i++)
755 : {
756 552 : timeinfo->tm_mon = i;
757 552 : timeinfo->tm_mday = 1; /* make sure we don't have invalid date */
758 552 : if (strftime_l(bufptr, MAX_L10N_DATA, "%b", timeinfo, locale) <= 0)
759 0 : strftimefail = true;
760 552 : bufptr += MAX_L10N_DATA;
761 552 : if (strftime_l(bufptr, MAX_L10N_DATA, "%B", timeinfo, locale) <= 0)
762 0 : strftimefail = true;
763 552 : bufptr += MAX_L10N_DATA;
764 : }
765 :
766 : #ifdef WIN32
767 : _free_locale(locale);
768 : #else
769 46 : freelocale(locale);
770 : #endif
771 :
772 : /*
773 : * At this point we've done our best to clean up, and can throw errors, or
774 : * call functions that might throw errors, with a clean conscience.
775 : */
776 46 : if (strftimefail)
777 0 : elog(ERROR, "strftime_l() failed");
778 :
779 : #ifndef WIN32
780 :
781 : /*
782 : * As in PGLC_localeconv(), we must convert strftime()'s output from the
783 : * encoding implied by LC_TIME to the database encoding. If we can't
784 : * identify the LC_TIME encoding, just perform encoding validation.
785 : */
786 46 : encoding = pg_get_encoding_from_locale(locale_time, true);
787 46 : if (encoding < 0)
788 0 : encoding = PG_SQL_ASCII;
789 :
790 : #else
791 :
792 : /*
793 : * On Windows, strftime_win32() always returns UTF8 data, so convert from
794 : * that if necessary.
795 : */
796 : encoding = PG_UTF8;
797 :
798 : #endif /* WIN32 */
799 :
800 46 : bufptr = buf;
801 :
802 : /* localized days */
803 368 : for (i = 0; i < 7; i++)
804 : {
805 322 : cache_single_string(&localized_abbrev_days[i], bufptr, encoding);
806 322 : bufptr += MAX_L10N_DATA;
807 322 : cache_single_string(&localized_full_days[i], bufptr, encoding);
808 322 : bufptr += MAX_L10N_DATA;
809 : }
810 46 : localized_abbrev_days[7] = NULL;
811 46 : localized_full_days[7] = NULL;
812 :
813 : /* localized months */
814 598 : for (i = 0; i < 12; i++)
815 : {
816 552 : cache_single_string(&localized_abbrev_months[i], bufptr, encoding);
817 552 : bufptr += MAX_L10N_DATA;
818 552 : cache_single_string(&localized_full_months[i], bufptr, encoding);
819 552 : bufptr += MAX_L10N_DATA;
820 : }
821 46 : localized_abbrev_months[12] = NULL;
822 46 : localized_full_months[12] = NULL;
823 :
824 46 : CurrentLCTimeValid = true;
825 : }
826 :
827 :
828 : #if defined(WIN32) && defined(LC_MESSAGES)
829 : /*
830 : * Convert a Windows setlocale() argument to a Unix-style one.
831 : *
832 : * Regardless of platform, we install message catalogs under a Unix-style
833 : * LL[_CC][.ENCODING][@VARIANT] naming convention. Only LC_MESSAGES settings
834 : * following that style will elicit localized interface strings.
835 : *
836 : * Before Visual Studio 2012 (msvcr110.dll), Windows setlocale() accepted "C"
837 : * (but not "c") and strings of the form <Language>[_<Country>][.<CodePage>],
838 : * case-insensitive. setlocale() returns the fully-qualified form; for
839 : * example, setlocale("thaI") returns "Thai_Thailand.874". Internally,
840 : * setlocale() and _create_locale() select a "locale identifier"[1] and store
841 : * it in an undocumented _locale_t field. From that LCID, we can retrieve the
842 : * ISO 639 language and the ISO 3166 country. Character encoding does not
843 : * matter, because the server and client encodings govern that.
844 : *
845 : * Windows Vista introduced the "locale name" concept[2], closely following
846 : * RFC 4646. Locale identifiers are now deprecated. Starting with Visual
847 : * Studio 2012, setlocale() accepts locale names in addition to the strings it
848 : * accepted historically. It does not standardize them; setlocale("Th-tH")
849 : * returns "Th-tH". setlocale(category, "") still returns a traditional
850 : * string. Furthermore, msvcr110.dll changed the undocumented _locale_t
851 : * content to carry locale names instead of locale identifiers.
852 : *
853 : * Visual Studio 2015 should still be able to do the same as Visual Studio
854 : * 2012, but the declaration of locale_name is missing in _locale_t, causing
855 : * this code compilation to fail, hence this falls back instead on to
856 : * enumerating all system locales by using EnumSystemLocalesEx to find the
857 : * required locale name. If the input argument is in Unix-style then we can
858 : * get ISO Locale name directly by using GetLocaleInfoEx() with LCType as
859 : * LOCALE_SNAME.
860 : *
861 : * This function returns a pointer to a static buffer bearing the converted
862 : * name or NULL if conversion fails.
863 : *
864 : * [1] https://docs.microsoft.com/en-us/windows/win32/intl/locale-identifiers
865 : * [2] https://docs.microsoft.com/en-us/windows/win32/intl/locale-names
866 : */
867 :
868 : /*
869 : * Callback function for EnumSystemLocalesEx() in get_iso_localename().
870 : *
871 : * This function enumerates all system locales, searching for one that matches
872 : * an input with the format: <Language>[_<Country>], e.g.
873 : * English[_United States]
874 : *
875 : * The input is a three wchar_t array as an LPARAM. The first element is the
876 : * locale_name we want to match, the second element is an allocated buffer
877 : * where the Unix-style locale is copied if a match is found, and the third
878 : * element is the search status, 1 if a match was found, 0 otherwise.
879 : */
880 : static BOOL CALLBACK
881 : search_locale_enum(LPWSTR pStr, DWORD dwFlags, LPARAM lparam)
882 : {
883 : wchar_t test_locale[LOCALE_NAME_MAX_LENGTH];
884 : wchar_t **argv;
885 :
886 : (void) (dwFlags);
887 :
888 : argv = (wchar_t **) lparam;
889 : *argv[2] = (wchar_t) 0;
890 :
891 : memset(test_locale, 0, sizeof(test_locale));
892 :
893 : /* Get the name of the <Language> in English */
894 : if (GetLocaleInfoEx(pStr, LOCALE_SENGLISHLANGUAGENAME,
895 : test_locale, LOCALE_NAME_MAX_LENGTH))
896 : {
897 : /*
898 : * If the enumerated locale does not have a hyphen ("en") OR the
899 : * locale_name input does not have an underscore ("English"), we only
900 : * need to compare the <Language> tags.
901 : */
902 : if (wcsrchr(pStr, '-') == NULL || wcsrchr(argv[0], '_') == NULL)
903 : {
904 : if (_wcsicmp(argv[0], test_locale) == 0)
905 : {
906 : wcscpy(argv[1], pStr);
907 : *argv[2] = (wchar_t) 1;
908 : return FALSE;
909 : }
910 : }
911 :
912 : /*
913 : * We have to compare a full <Language>_<Country> tag, so we append
914 : * the underscore and name of the country/region in English, e.g.
915 : * "English_United States".
916 : */
917 : else
918 : {
919 : size_t len;
920 :
921 : wcscat(test_locale, L"_");
922 : len = wcslen(test_locale);
923 : if (GetLocaleInfoEx(pStr, LOCALE_SENGLISHCOUNTRYNAME,
924 : test_locale + len,
925 : LOCALE_NAME_MAX_LENGTH - len))
926 : {
927 : if (_wcsicmp(argv[0], test_locale) == 0)
928 : {
929 : wcscpy(argv[1], pStr);
930 : *argv[2] = (wchar_t) 1;
931 : return FALSE;
932 : }
933 : }
934 : }
935 : }
936 :
937 : return TRUE;
938 : }
939 :
940 : /*
941 : * This function converts a Windows locale name to an ISO formatted version
942 : * for Visual Studio 2015 or greater.
943 : *
944 : * Returns NULL, if no valid conversion was found.
945 : */
946 : static char *
947 : get_iso_localename(const char *winlocname)
948 : {
949 : wchar_t wc_locale_name[LOCALE_NAME_MAX_LENGTH];
950 : wchar_t buffer[LOCALE_NAME_MAX_LENGTH];
951 : static char iso_lc_messages[LOCALE_NAME_MAX_LENGTH];
952 : char *period;
953 : int len;
954 : int ret_val;
955 :
956 : /*
957 : * Valid locales have the following syntax:
958 : * <Language>[_<Country>[.<CodePage>]]
959 : *
960 : * GetLocaleInfoEx can only take locale name without code-page and for the
961 : * purpose of this API the code-page doesn't matter.
962 : */
963 : period = strchr(winlocname, '.');
964 : if (period != NULL)
965 : len = period - winlocname;
966 : else
967 : len = pg_mbstrlen(winlocname);
968 :
969 : memset(wc_locale_name, 0, sizeof(wc_locale_name));
970 : memset(buffer, 0, sizeof(buffer));
971 : MultiByteToWideChar(CP_ACP, 0, winlocname, len, wc_locale_name,
972 : LOCALE_NAME_MAX_LENGTH);
973 :
974 : /*
975 : * If the lc_messages is already a Unix-style string, we have a direct
976 : * match with LOCALE_SNAME, e.g. en-US, en_US.
977 : */
978 : ret_val = GetLocaleInfoEx(wc_locale_name, LOCALE_SNAME, (LPWSTR) &buffer,
979 : LOCALE_NAME_MAX_LENGTH);
980 : if (!ret_val)
981 : {
982 : /*
983 : * Search for a locale in the system that matches language and country
984 : * name.
985 : */
986 : wchar_t *argv[3];
987 :
988 : argv[0] = wc_locale_name;
989 : argv[1] = buffer;
990 : argv[2] = (wchar_t *) &ret_val;
991 : EnumSystemLocalesEx(search_locale_enum, LOCALE_WINDOWS, (LPARAM) argv,
992 : NULL);
993 : }
994 :
995 : if (ret_val)
996 : {
997 : size_t rc;
998 : char *hyphen;
999 :
1000 : /* Locale names use only ASCII, any conversion locale suffices. */
1001 : rc = wchar2char(iso_lc_messages, buffer, sizeof(iso_lc_messages), NULL);
1002 : if (rc == -1 || rc == sizeof(iso_lc_messages))
1003 : return NULL;
1004 :
1005 : /*
1006 : * Since the message catalogs sit on a case-insensitive filesystem, we
1007 : * need not standardize letter case here. So long as we do not ship
1008 : * message catalogs for which it would matter, we also need not
1009 : * translate the script/variant portion, e.g. uz-Cyrl-UZ to
1010 : * uz_UZ@cyrillic. Simply replace the hyphen with an underscore.
1011 : */
1012 : hyphen = strchr(iso_lc_messages, '-');
1013 : if (hyphen)
1014 : *hyphen = '_';
1015 : return iso_lc_messages;
1016 : }
1017 :
1018 : return NULL;
1019 : }
1020 :
1021 : static char *
1022 : IsoLocaleName(const char *winlocname)
1023 : {
1024 : static char iso_lc_messages[LOCALE_NAME_MAX_LENGTH];
1025 :
1026 : if (pg_strcasecmp("c", winlocname) == 0 ||
1027 : pg_strcasecmp("posix", winlocname) == 0)
1028 : {
1029 : strcpy(iso_lc_messages, "C");
1030 : return iso_lc_messages;
1031 : }
1032 : else
1033 : return get_iso_localename(winlocname);
1034 : }
1035 :
1036 : #endif /* WIN32 && LC_MESSAGES */
1037 :
1038 : /*
1039 : * Create a new pg_locale_t struct for the given collation oid.
1040 : */
1041 : static pg_locale_t
1042 4242 : create_pg_locale(Oid collid, MemoryContext context)
1043 : {
1044 : HeapTuple tp;
1045 : Form_pg_collation collform;
1046 : pg_locale_t result;
1047 : Datum datum;
1048 : bool isnull;
1049 :
1050 4242 : tp = SearchSysCache1(COLLOID, ObjectIdGetDatum(collid));
1051 4242 : if (!HeapTupleIsValid(tp))
1052 0 : elog(ERROR, "cache lookup failed for collation %u", collid);
1053 4242 : collform = (Form_pg_collation) GETSTRUCT(tp);
1054 :
1055 4242 : if (collform->collprovider == COLLPROVIDER_BUILTIN)
1056 56 : result = create_pg_locale_builtin(collid, context);
1057 4186 : else if (collform->collprovider == COLLPROVIDER_ICU)
1058 186 : result = create_pg_locale_icu(collid, context);
1059 4000 : else if (collform->collprovider == COLLPROVIDER_LIBC)
1060 4000 : result = create_pg_locale_libc(collid, context);
1061 : else
1062 : /* shouldn't happen */
1063 0 : PGLOCALE_SUPPORT_ERROR(collform->collprovider);
1064 :
1065 4236 : result->is_default = false;
1066 :
1067 : Assert((result->collate_is_c && result->collate == NULL) ||
1068 : (!result->collate_is_c && result->collate != NULL));
1069 :
1070 : Assert((result->ctype_is_c && result->ctype == NULL) ||
1071 : (!result->ctype_is_c && result->ctype != NULL));
1072 :
1073 4236 : datum = SysCacheGetAttr(COLLOID, tp, Anum_pg_collation_collversion,
1074 : &isnull);
1075 4236 : if (!isnull)
1076 : {
1077 : char *actual_versionstr;
1078 : char *collversionstr;
1079 :
1080 236 : collversionstr = TextDatumGetCString(datum);
1081 :
1082 236 : if (collform->collprovider == COLLPROVIDER_LIBC)
1083 0 : datum = SysCacheGetAttrNotNull(COLLOID, tp, Anum_pg_collation_collcollate);
1084 : else
1085 236 : datum = SysCacheGetAttrNotNull(COLLOID, tp, Anum_pg_collation_colllocale);
1086 :
1087 236 : actual_versionstr = get_collation_actual_version(collform->collprovider,
1088 236 : TextDatumGetCString(datum));
1089 236 : if (!actual_versionstr)
1090 : {
1091 : /*
1092 : * This could happen when specifying a version in CREATE COLLATION
1093 : * but the provider does not support versioning, or manually
1094 : * creating a mess in the catalogs.
1095 : */
1096 0 : ereport(ERROR,
1097 : (errmsg("collation \"%s\" has no actual version, but a version was recorded",
1098 : NameStr(collform->collname))));
1099 : }
1100 :
1101 236 : if (strcmp(actual_versionstr, collversionstr) != 0)
1102 0 : ereport(WARNING,
1103 : (errmsg("collation \"%s\" has version mismatch",
1104 : NameStr(collform->collname)),
1105 : errdetail("The collation in the database was created using version %s, "
1106 : "but the operating system provides version %s.",
1107 : collversionstr, actual_versionstr),
1108 : errhint("Rebuild all objects affected by this collation and run "
1109 : "ALTER COLLATION %s REFRESH VERSION, "
1110 : "or build PostgreSQL with the right library version.",
1111 : quote_qualified_identifier(get_namespace_name(collform->collnamespace),
1112 : NameStr(collform->collname)))));
1113 : }
1114 :
1115 4236 : ReleaseSysCache(tp);
1116 :
1117 4236 : return result;
1118 : }
1119 :
1120 : /*
1121 : * Initialize default_locale with database locale settings.
1122 : */
1123 : void
1124 31590 : init_database_collation(void)
1125 : {
1126 : HeapTuple tup;
1127 : Form_pg_database dbform;
1128 : pg_locale_t result;
1129 :
1130 : Assert(default_locale == NULL);
1131 :
1132 : /* Fetch our pg_database row normally, via syscache */
1133 31590 : tup = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(MyDatabaseId));
1134 31590 : if (!HeapTupleIsValid(tup))
1135 0 : elog(ERROR, "cache lookup failed for database %u", MyDatabaseId);
1136 31590 : dbform = (Form_pg_database) GETSTRUCT(tup);
1137 :
1138 31590 : if (dbform->datlocprovider == COLLPROVIDER_BUILTIN)
1139 1742 : result = create_pg_locale_builtin(DEFAULT_COLLATION_OID,
1140 : TopMemoryContext);
1141 29848 : else if (dbform->datlocprovider == COLLPROVIDER_ICU)
1142 26 : result = create_pg_locale_icu(DEFAULT_COLLATION_OID,
1143 : TopMemoryContext);
1144 29822 : else if (dbform->datlocprovider == COLLPROVIDER_LIBC)
1145 29822 : result = create_pg_locale_libc(DEFAULT_COLLATION_OID,
1146 : TopMemoryContext);
1147 : else
1148 : /* shouldn't happen */
1149 0 : PGLOCALE_SUPPORT_ERROR(dbform->datlocprovider);
1150 :
1151 31586 : result->is_default = true;
1152 31586 : ReleaseSysCache(tup);
1153 :
1154 31586 : default_locale = result;
1155 31586 : }
1156 :
1157 : /*
1158 : * Create a pg_locale_t from a collation OID. Results are cached for the
1159 : * lifetime of the backend. Thus, do not free the result with freelocale().
1160 : *
1161 : * For simplicity, we always generate COLLATE + CTYPE even though we
1162 : * might only need one of them. Since this is called only once per session,
1163 : * it shouldn't cost much.
1164 : */
1165 : pg_locale_t
1166 28681100 : pg_newlocale_from_collation(Oid collid)
1167 : {
1168 : collation_cache_entry *cache_entry;
1169 : bool found;
1170 :
1171 28681100 : if (collid == DEFAULT_COLLATION_OID)
1172 24854666 : return default_locale;
1173 :
1174 3826434 : if (!OidIsValid(collid))
1175 0 : elog(ERROR, "cache lookup failed for collation %u", collid);
1176 :
1177 3826434 : AssertCouldGetRelation();
1178 :
1179 3826434 : if (last_collation_cache_oid == collid)
1180 3820252 : return last_collation_cache_locale;
1181 :
1182 6182 : if (CollationCache == NULL)
1183 : {
1184 3930 : CollationCacheContext = AllocSetContextCreate(TopMemoryContext,
1185 : "collation cache",
1186 : ALLOCSET_DEFAULT_SIZES);
1187 3930 : CollationCache = collation_cache_create(CollationCacheContext,
1188 : 16, NULL);
1189 : }
1190 :
1191 6182 : cache_entry = collation_cache_insert(CollationCache, collid, &found);
1192 6182 : if (!found)
1193 : {
1194 : /*
1195 : * Make sure cache entry is marked invalid, in case we fail before
1196 : * setting things.
1197 : */
1198 4242 : cache_entry->locale = 0;
1199 : }
1200 :
1201 6182 : if (cache_entry->locale == 0)
1202 : {
1203 4242 : cache_entry->locale = create_pg_locale(collid, CollationCacheContext);
1204 : }
1205 :
1206 6176 : last_collation_cache_oid = collid;
1207 6176 : last_collation_cache_locale = cache_entry->locale;
1208 :
1209 6176 : return cache_entry->locale;
1210 : }
1211 :
1212 : /*
1213 : * Get provider-specific collation version string for the given collation from
1214 : * the operating system/library.
1215 : */
1216 : char *
1217 110600 : get_collation_actual_version(char collprovider, const char *collcollate)
1218 : {
1219 110600 : char *collversion = NULL;
1220 :
1221 110600 : if (collprovider == COLLPROVIDER_BUILTIN)
1222 1872 : collversion = get_collation_actual_version_builtin(collcollate);
1223 : #ifdef USE_ICU
1224 108728 : else if (collprovider == COLLPROVIDER_ICU)
1225 79458 : collversion = get_collation_actual_version_icu(collcollate);
1226 : #endif
1227 29270 : else if (collprovider == COLLPROVIDER_LIBC)
1228 29270 : collversion = get_collation_actual_version_libc(collcollate);
1229 :
1230 110600 : return collversion;
1231 : }
1232 :
1233 : size_t
1234 436442 : pg_strlower(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1235 : pg_locale_t locale)
1236 : {
1237 436442 : return locale->ctype->strlower(dst, dstsize, src, srclen, locale);
1238 : }
1239 :
1240 : size_t
1241 232 : pg_strtitle(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1242 : pg_locale_t locale)
1243 : {
1244 232 : return locale->ctype->strtitle(dst, dstsize, src, srclen, locale);
1245 : }
1246 :
1247 : size_t
1248 1035090 : pg_strupper(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1249 : pg_locale_t locale)
1250 : {
1251 1035090 : return locale->ctype->strupper(dst, dstsize, src, srclen, locale);
1252 : }
1253 :
1254 : size_t
1255 24 : pg_strfold(char *dst, size_t dstsize, const char *src, ssize_t srclen,
1256 : pg_locale_t locale)
1257 : {
1258 24 : if (locale->ctype->strfold)
1259 24 : return locale->ctype->strfold(dst, dstsize, src, srclen, locale);
1260 : else
1261 0 : return locale->ctype->strlower(dst, dstsize, src, srclen, locale);
1262 : }
1263 :
1264 : /*
1265 : * pg_strcoll
1266 : *
1267 : * Like pg_strncoll for NUL-terminated input strings.
1268 : */
1269 : int
1270 24824564 : pg_strcoll(const char *arg1, const char *arg2, pg_locale_t locale)
1271 : {
1272 24824564 : return locale->collate->strncoll(arg1, -1, arg2, -1, locale);
1273 : }
1274 :
1275 : /*
1276 : * pg_strncoll
1277 : *
1278 : * Call ucol_strcollUTF8(), ucol_strcoll(), strcoll_l() or wcscoll_l() as
1279 : * appropriate for the given locale, platform, and database encoding. If the
1280 : * locale is not specified, use the database collation.
1281 : *
1282 : * The input strings must be encoded in the database encoding. If an input
1283 : * string is NUL-terminated, its length may be specified as -1.
1284 : *
1285 : * The caller is responsible for breaking ties if the collation is
1286 : * deterministic; this maintains consistency with pg_strnxfrm(), which cannot
1287 : * easily account for deterministic collations.
1288 : */
1289 : int
1290 4456454 : pg_strncoll(const char *arg1, ssize_t len1, const char *arg2, ssize_t len2,
1291 : pg_locale_t locale)
1292 : {
1293 4456454 : return locale->collate->strncoll(arg1, len1, arg2, len2, locale);
1294 : }
1295 :
1296 : /*
1297 : * Return true if the collation provider supports pg_strxfrm() and
1298 : * pg_strnxfrm(); otherwise false.
1299 : *
1300 : *
1301 : * No similar problem is known for the ICU provider.
1302 : */
1303 : bool
1304 46170 : pg_strxfrm_enabled(pg_locale_t locale)
1305 : {
1306 : /*
1307 : * locale->collate->strnxfrm is still a required method, even if it may
1308 : * have the wrong behavior, because the planner uses it for estimates in
1309 : * some cases.
1310 : */
1311 46170 : return locale->collate->strxfrm_is_safe;
1312 : }
1313 :
1314 : /*
1315 : * pg_strxfrm
1316 : *
1317 : * Like pg_strnxfrm for a NUL-terminated input string.
1318 : */
1319 : size_t
1320 144 : pg_strxfrm(char *dest, const char *src, size_t destsize, pg_locale_t locale)
1321 : {
1322 144 : return locale->collate->strnxfrm(dest, destsize, src, -1, locale);
1323 : }
1324 :
1325 : /*
1326 : * pg_strnxfrm
1327 : *
1328 : * Transforms 'src' to a nul-terminated string stored in 'dest' such that
1329 : * ordinary strcmp() on transformed strings is equivalent to pg_strcoll() on
1330 : * untransformed strings.
1331 : *
1332 : * The input string must be encoded in the database encoding. If the input
1333 : * string is NUL-terminated, its length may be specified as -1. If 'destsize'
1334 : * is zero, 'dest' may be NULL.
1335 : *
1336 : * Not all providers support pg_strnxfrm() safely. The caller should check
1337 : * pg_strxfrm_enabled() first, otherwise this function may return wrong
1338 : * results or an error.
1339 : *
1340 : * Returns the number of bytes needed (or more) to store the transformed
1341 : * string, excluding the terminating nul byte. If the value returned is
1342 : * 'destsize' or greater, the resulting contents of 'dest' are undefined.
1343 : */
1344 : size_t
1345 10020 : pg_strnxfrm(char *dest, size_t destsize, const char *src, ssize_t srclen,
1346 : pg_locale_t locale)
1347 : {
1348 10020 : return locale->collate->strnxfrm(dest, destsize, src, srclen, locale);
1349 : }
1350 :
1351 : /*
1352 : * Return true if the collation provider supports pg_strxfrm_prefix() and
1353 : * pg_strnxfrm_prefix(); otherwise false.
1354 : */
1355 : bool
1356 1668 : pg_strxfrm_prefix_enabled(pg_locale_t locale)
1357 : {
1358 1668 : return (locale->collate->strnxfrm_prefix != NULL);
1359 : }
1360 :
1361 : /*
1362 : * pg_strxfrm_prefix
1363 : *
1364 : * Like pg_strnxfrm_prefix for a NUL-terminated input string.
1365 : */
1366 : size_t
1367 1668 : pg_strxfrm_prefix(char *dest, const char *src, size_t destsize,
1368 : pg_locale_t locale)
1369 : {
1370 1668 : return locale->collate->strnxfrm_prefix(dest, destsize, src, -1, locale);
1371 : }
1372 :
1373 : /*
1374 : * pg_strnxfrm_prefix
1375 : *
1376 : * Transforms 'src' to a byte sequence stored in 'dest' such that ordinary
1377 : * memcmp() on the byte sequence is equivalent to pg_strncoll() on
1378 : * untransformed strings. The result is not nul-terminated.
1379 : *
1380 : * The input string must be encoded in the database encoding. If the input
1381 : * string is NUL-terminated, its length may be specified as -1.
1382 : *
1383 : * Not all providers support pg_strnxfrm_prefix() safely. The caller should
1384 : * check pg_strxfrm_prefix_enabled() first, otherwise this function may return
1385 : * wrong results or an error.
1386 : *
1387 : * If destsize is not large enough to hold the resulting byte sequence, stores
1388 : * only the first destsize bytes in 'dest'. Returns the number of bytes
1389 : * actually copied to 'dest'.
1390 : */
1391 : size_t
1392 0 : pg_strnxfrm_prefix(char *dest, size_t destsize, const char *src,
1393 : ssize_t srclen, pg_locale_t locale)
1394 : {
1395 0 : return locale->collate->strnxfrm_prefix(dest, destsize, src, srclen, locale);
1396 : }
1397 :
1398 : /*
1399 : * char_is_cased()
1400 : *
1401 : * Fuzzy test of whether the given char is case-varying or not. The argument
1402 : * is a single byte, so in a multibyte encoding, just assume any non-ASCII
1403 : * char is case-varying.
1404 : */
1405 : bool
1406 126 : char_is_cased(char ch, pg_locale_t locale)
1407 : {
1408 126 : return locale->ctype->char_is_cased(ch, locale);
1409 : }
1410 :
1411 : /*
1412 : * char_tolower_enabled()
1413 : *
1414 : * Does the provider support char_tolower()?
1415 : */
1416 : bool
1417 68498 : char_tolower_enabled(pg_locale_t locale)
1418 : {
1419 68498 : return (locale->ctype->char_tolower != NULL);
1420 : }
1421 :
1422 : /*
1423 : * char_tolower()
1424 : *
1425 : * Convert char (single-byte encoding) to lowercase.
1426 : */
1427 : char
1428 0 : char_tolower(unsigned char ch, pg_locale_t locale)
1429 : {
1430 0 : return locale->ctype->char_tolower(ch, locale);
1431 : }
1432 :
1433 : /*
1434 : * Return required encoding ID for the given locale, or -1 if any encoding is
1435 : * valid for the locale.
1436 : */
1437 : int
1438 1932 : builtin_locale_encoding(const char *locale)
1439 : {
1440 1932 : if (strcmp(locale, "C") == 0)
1441 62 : return -1;
1442 1870 : else if (strcmp(locale, "C.UTF-8") == 0)
1443 1834 : return PG_UTF8;
1444 36 : else if (strcmp(locale, "PG_UNICODE_FAST") == 0)
1445 36 : return PG_UTF8;
1446 :
1447 :
1448 0 : ereport(ERROR,
1449 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1450 : errmsg("invalid locale name \"%s\" for builtin provider",
1451 : locale)));
1452 :
1453 : return 0; /* keep compiler quiet */
1454 : }
1455 :
1456 :
1457 : /*
1458 : * Validate the locale and encoding combination, and return the canonical form
1459 : * of the locale name.
1460 : */
1461 : const char *
1462 1912 : builtin_validate_locale(int encoding, const char *locale)
1463 : {
1464 1912 : const char *canonical_name = NULL;
1465 : int required_encoding;
1466 :
1467 1912 : if (strcmp(locale, "C") == 0)
1468 50 : canonical_name = "C";
1469 1862 : else if (strcmp(locale, "C.UTF-8") == 0 || strcmp(locale, "C.UTF8") == 0)
1470 1818 : canonical_name = "C.UTF-8";
1471 44 : else if (strcmp(locale, "PG_UNICODE_FAST") == 0)
1472 26 : canonical_name = "PG_UNICODE_FAST";
1473 :
1474 1912 : if (!canonical_name)
1475 18 : ereport(ERROR,
1476 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1477 : errmsg("invalid locale name \"%s\" for builtin provider",
1478 : locale)));
1479 :
1480 1894 : required_encoding = builtin_locale_encoding(canonical_name);
1481 1894 : if (required_encoding >= 0 && encoding != required_encoding)
1482 2 : ereport(ERROR,
1483 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1484 : errmsg("encoding \"%s\" does not match locale \"%s\"",
1485 : pg_encoding_to_char(encoding), locale)));
1486 :
1487 1892 : return canonical_name;
1488 : }
1489 :
1490 :
1491 :
1492 : /*
1493 : * Return the BCP47 language tag representation of the requested locale.
1494 : *
1495 : * This function should be called before passing the string to ucol_open(),
1496 : * because conversion to a language tag also performs "level 2
1497 : * canonicalization". In addition to producing a consistent format, level 2
1498 : * canonicalization is able to more accurately interpret different input
1499 : * locale string formats, such as POSIX and .NET IDs.
1500 : */
1501 : char *
1502 79148 : icu_language_tag(const char *loc_str, int elevel)
1503 : {
1504 : #ifdef USE_ICU
1505 : UErrorCode status;
1506 : char *langtag;
1507 79148 : size_t buflen = 32; /* arbitrary starting buffer size */
1508 79148 : const bool strict = true;
1509 :
1510 : /*
1511 : * A BCP47 language tag doesn't have a clearly-defined upper limit (cf.
1512 : * RFC5646 section 4.4). Additionally, in older ICU versions,
1513 : * uloc_toLanguageTag() doesn't always return the ultimate length on the
1514 : * first call, necessitating a loop.
1515 : */
1516 79148 : langtag = palloc(buflen);
1517 : while (true)
1518 : {
1519 79148 : status = U_ZERO_ERROR;
1520 79148 : uloc_toLanguageTag(loc_str, langtag, buflen, strict, &status);
1521 :
1522 : /* try again if the buffer is not large enough */
1523 79148 : if ((status == U_BUFFER_OVERFLOW_ERROR ||
1524 79148 : status == U_STRING_NOT_TERMINATED_WARNING) &&
1525 : buflen < MaxAllocSize)
1526 : {
1527 0 : buflen = Min(buflen * 2, MaxAllocSize);
1528 0 : langtag = repalloc(langtag, buflen);
1529 0 : continue;
1530 : }
1531 :
1532 79148 : break;
1533 : }
1534 :
1535 79148 : if (U_FAILURE(status))
1536 : {
1537 18 : pfree(langtag);
1538 :
1539 18 : if (elevel > 0)
1540 14 : ereport(elevel,
1541 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1542 : errmsg("could not convert locale name \"%s\" to language tag: %s",
1543 : loc_str, u_errorName(status))));
1544 12 : return NULL;
1545 : }
1546 :
1547 79130 : return langtag;
1548 : #else /* not USE_ICU */
1549 : ereport(ERROR,
1550 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1551 : errmsg("ICU is not supported in this build")));
1552 : return NULL; /* keep compiler quiet */
1553 : #endif /* not USE_ICU */
1554 : }
1555 :
1556 : /*
1557 : * Perform best-effort check that the locale is a valid one.
1558 : */
1559 : void
1560 168 : icu_validate_locale(const char *loc_str)
1561 : {
1562 : #ifdef USE_ICU
1563 : UCollator *collator;
1564 : UErrorCode status;
1565 : char lang[ULOC_LANG_CAPACITY];
1566 168 : bool found = false;
1567 168 : int elevel = icu_validation_level;
1568 :
1569 : /* no validation */
1570 168 : if (elevel < 0)
1571 12 : return;
1572 :
1573 : /* downgrade to WARNING during pg_upgrade */
1574 156 : if (IsBinaryUpgrade && elevel > WARNING)
1575 0 : elevel = WARNING;
1576 :
1577 : /* validate that we can extract the language */
1578 156 : status = U_ZERO_ERROR;
1579 156 : uloc_getLanguage(loc_str, lang, ULOC_LANG_CAPACITY, &status);
1580 156 : if (U_FAILURE(status) || status == U_STRING_NOT_TERMINATED_WARNING)
1581 : {
1582 0 : ereport(elevel,
1583 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1584 : errmsg("could not get language from ICU locale \"%s\": %s",
1585 : loc_str, u_errorName(status)),
1586 : errhint("To disable ICU locale validation, set the parameter \"%s\" to \"%s\".",
1587 : "icu_validation_level", "disabled")));
1588 0 : return;
1589 : }
1590 :
1591 : /* check for special language name */
1592 156 : if (strcmp(lang, "") == 0 ||
1593 46 : strcmp(lang, "root") == 0 || strcmp(lang, "und") == 0)
1594 110 : found = true;
1595 :
1596 : /* search for matching language within ICU */
1597 15706 : for (int32_t i = 0; !found && i < uloc_countAvailable(); i++)
1598 : {
1599 15550 : const char *otherloc = uloc_getAvailable(i);
1600 : char otherlang[ULOC_LANG_CAPACITY];
1601 :
1602 15550 : status = U_ZERO_ERROR;
1603 15550 : uloc_getLanguage(otherloc, otherlang, ULOC_LANG_CAPACITY, &status);
1604 15550 : if (U_FAILURE(status) || status == U_STRING_NOT_TERMINATED_WARNING)
1605 0 : continue;
1606 :
1607 15550 : if (strcmp(lang, otherlang) == 0)
1608 32 : found = true;
1609 : }
1610 :
1611 156 : if (!found)
1612 14 : ereport(elevel,
1613 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1614 : errmsg("ICU locale \"%s\" has unknown language \"%s\"",
1615 : loc_str, lang),
1616 : errhint("To disable ICU locale validation, set the parameter \"%s\" to \"%s\".",
1617 : "icu_validation_level", "disabled")));
1618 :
1619 : /* check that it can be opened */
1620 150 : collator = pg_ucol_open(loc_str);
1621 142 : ucol_close(collator);
1622 : #else /* not USE_ICU */
1623 : /* could get here if a collation was created by a build with ICU */
1624 : ereport(ERROR,
1625 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1626 : errmsg("ICU is not supported in this build")));
1627 : #endif /* not USE_ICU */
1628 : }
|
