Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * varlena.c
4 : * Functions for the variable-length built-in types.
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/adt/varlena.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <ctype.h>
18 : #include <limits.h>
19 :
20 : #include "access/detoast.h"
21 : #include "access/toast_compression.h"
22 : #include "catalog/pg_collation.h"
23 : #include "catalog/pg_type.h"
24 : #include "common/hashfn.h"
25 : #include "common/int.h"
26 : #include "common/unicode_category.h"
27 : #include "common/unicode_norm.h"
28 : #include "common/unicode_version.h"
29 : #include "funcapi.h"
30 : #include "lib/hyperloglog.h"
31 : #include "libpq/pqformat.h"
32 : #include "miscadmin.h"
33 : #include "nodes/execnodes.h"
34 : #include "parser/scansup.h"
35 : #include "port/pg_bswap.h"
36 : #include "regex/regex.h"
37 : #include "utils/builtins.h"
38 : #include "utils/guc.h"
39 : #include "utils/lsyscache.h"
40 : #include "utils/memutils.h"
41 : #include "utils/pg_locale.h"
42 : #include "utils/sortsupport.h"
43 : #include "utils/varlena.h"
44 :
45 : typedef varlena VarString;
46 :
47 : /*
48 : * State for text_position_* functions.
49 : */
50 : typedef struct
51 : {
52 : pg_locale_t locale; /* collation used for substring matching */
53 : bool is_multibyte_char_in_char; /* need to check char boundaries? */
54 : bool greedy; /* find longest possible substring? */
55 :
56 : char *str1; /* haystack string */
57 : char *str2; /* needle string */
58 : int len1; /* string lengths in bytes */
59 : int len2;
60 :
61 : /* Skip table for Boyer-Moore-Horspool search algorithm: */
62 : int skiptablemask; /* mask for ANDing with skiptable subscripts */
63 : int skiptable[256]; /* skip distance for given mismatched char */
64 :
65 : /*
66 : * Note that with nondeterministic collations, the length of the last
67 : * match is not necessarily equal to the length of the "needle" passed in.
68 : */
69 : char *last_match; /* pointer to last match in 'str1' */
70 : int last_match_len; /* length of last match */
71 : int last_match_len_tmp; /* same but for internal use */
72 :
73 : /*
74 : * Sometimes we need to convert the byte position of a match to a
75 : * character position. These store the last position that was converted,
76 : * so that on the next call, we can continue from that point, rather than
77 : * count characters from the very beginning.
78 : */
79 : char *refpoint; /* pointer within original haystack string */
80 : int refpos; /* 0-based character offset of the same point */
81 : } TextPositionState;
82 :
83 : typedef struct
84 : {
85 : char *buf1; /* 1st string, or abbreviation original string
86 : * buf */
87 : char *buf2; /* 2nd string, or abbreviation strxfrm() buf */
88 : int buflen1; /* Allocated length of buf1 */
89 : int buflen2; /* Allocated length of buf2 */
90 : int last_len1; /* Length of last buf1 string/strxfrm() input */
91 : int last_len2; /* Length of last buf2 string/strxfrm() blob */
92 : int last_returned; /* Last comparison result (cache) */
93 : bool cache_blob; /* Does buf2 contain strxfrm() blob, etc? */
94 : bool collate_c;
95 : Oid typid; /* Actual datatype (text/bpchar/name) */
96 : hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
97 : hyperLogLogState full_card; /* Full key cardinality state */
98 : double prop_card; /* Required cardinality proportion */
99 : pg_locale_t locale;
100 : } VarStringSortSupport;
101 :
102 : /*
103 : * Output data for split_text(): we output either to an array or a table.
104 : * tupstore and tupdesc must be set up in advance to output to a table.
105 : */
106 : typedef struct
107 : {
108 : ArrayBuildState *astate;
109 : Tuplestorestate *tupstore;
110 : TupleDesc tupdesc;
111 : } SplitTextOutputData;
112 :
113 : /*
114 : * This should be large enough that most strings will fit, but small enough
115 : * that we feel comfortable putting it on the stack
116 : */
117 : #define TEXTBUFLEN 1024
118 :
119 : #define DatumGetVarStringP(X) ((VarString *) PG_DETOAST_DATUM(X))
120 : #define DatumGetVarStringPP(X) ((VarString *) PG_DETOAST_DATUM_PACKED(X))
121 :
122 : static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup);
123 : static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup);
124 : static int namefastcmp_c(Datum x, Datum y, SortSupport ssup);
125 : static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup);
126 : static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup);
127 : static int varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup);
128 : static Datum varstr_abbrev_convert(Datum original, SortSupport ssup);
129 : static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup);
130 : static int32 text_length(Datum str);
131 : static text *text_catenate(text *t1, text *t2);
132 : static text *text_substring(Datum str,
133 : int32 start,
134 : int32 length,
135 : bool length_not_specified);
136 : static int pg_mbcharcliplen_chars(const char *mbstr, int len, int limit);
137 : static text *text_overlay(text *t1, text *t2, int sp, int sl);
138 : static int text_position(text *t1, text *t2, Oid collid);
139 : static void text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state);
140 : static bool text_position_next(TextPositionState *state);
141 : static char *text_position_next_internal(char *start_ptr, TextPositionState *state);
142 : static char *text_position_get_match_ptr(TextPositionState *state);
143 : static int text_position_get_match_pos(TextPositionState *state);
144 : static void text_position_cleanup(TextPositionState *state);
145 : static void check_collation_set(Oid collid);
146 : static int text_cmp(text *arg1, text *arg2, Oid collid);
147 : static void appendStringInfoText(StringInfo str, const text *t);
148 : static bool split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate);
149 : static void split_text_accum_result(SplitTextOutputData *tstate,
150 : text *field_value,
151 : text *null_string,
152 : Oid collation);
153 : static text *array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
154 : const char *fldsep, const char *null_string);
155 : static StringInfo makeStringAggState(FunctionCallInfo fcinfo);
156 : static bool text_format_parse_digits(const char **ptr, const char *end_ptr,
157 : int *value);
158 : static const char *text_format_parse_format(const char *start_ptr,
159 : const char *end_ptr,
160 : int *argpos, int *widthpos,
161 : int *flags, int *width);
162 : static void text_format_string_conversion(StringInfo buf, char conversion,
163 : FmgrInfo *typOutputInfo,
164 : Datum value, bool isNull,
165 : int flags, int width);
166 : static void text_format_append_string(StringInfo buf, const char *str,
167 : int flags, int width);
168 :
169 :
170 : /*****************************************************************************
171 : * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
172 : *****************************************************************************/
173 :
174 : /*
175 : * cstring_to_text
176 : *
177 : * Create a text value from a null-terminated C string.
178 : *
179 : * The new text value is freshly palloc'd with a full-size VARHDR.
180 : */
181 : text *
182 12871951 : cstring_to_text(const char *s)
183 : {
184 12871951 : return cstring_to_text_with_len(s, strlen(s));
185 : }
186 :
187 : /*
188 : * cstring_to_text_with_len
189 : *
190 : * Same as cstring_to_text except the caller specifies the string length;
191 : * the string need not be null_terminated.
192 : */
193 : text *
194 14237867 : cstring_to_text_with_len(const char *s, int len)
195 : {
196 14237867 : text *result = (text *) palloc(len + VARHDRSZ);
197 :
198 14237867 : SET_VARSIZE(result, len + VARHDRSZ);
199 14237867 : memcpy(VARDATA(result), s, len);
200 :
201 14237867 : return result;
202 : }
203 :
204 : /*
205 : * text_to_cstring
206 : *
207 : * Create a palloc'd, null-terminated C string from a text value.
208 : *
209 : * We support being passed a compressed or toasted text value.
210 : * This is a bit bogus since such values shouldn't really be referred to as
211 : * "text *", but it seems useful for robustness. If we didn't handle that
212 : * case here, we'd need another routine that did, anyway.
213 : */
214 : char *
215 9202727 : text_to_cstring(const text *t)
216 : {
217 : /* must cast away the const, unfortunately */
218 9202727 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
219 9202727 : int len = VARSIZE_ANY_EXHDR(tunpacked);
220 : char *result;
221 :
222 9202727 : result = (char *) palloc(len + 1);
223 9202727 : memcpy(result, VARDATA_ANY(tunpacked), len);
224 9202727 : result[len] = '\0';
225 :
226 9202727 : if (tunpacked != t)
227 22973 : pfree(tunpacked);
228 :
229 9202727 : return result;
230 : }
231 :
232 : /*
233 : * text_to_cstring_buffer
234 : *
235 : * Copy a text value into a caller-supplied buffer of size dst_len.
236 : *
237 : * The text string is truncated if necessary to fit. The result is
238 : * guaranteed null-terminated (unless dst_len == 0).
239 : *
240 : * We support being passed a compressed or toasted text value.
241 : * This is a bit bogus since such values shouldn't really be referred to as
242 : * "text *", but it seems useful for robustness. If we didn't handle that
243 : * case here, we'd need another routine that did, anyway.
244 : */
245 : void
246 503 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
247 : {
248 : /* must cast away the const, unfortunately */
249 503 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
250 503 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
251 :
252 503 : if (dst_len > 0)
253 : {
254 503 : dst_len--;
255 503 : if (dst_len >= src_len)
256 503 : dst_len = src_len;
257 : else /* ensure truncation is encoding-safe */
258 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
259 503 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
260 503 : dst[dst_len] = '\0';
261 : }
262 :
263 503 : if (srcunpacked != src)
264 0 : pfree(srcunpacked);
265 503 : }
266 :
267 :
268 : /*****************************************************************************
269 : * USER I/O ROUTINES *
270 : *****************************************************************************/
271 :
272 : /*
273 : * textin - converts cstring to internal representation
274 : */
275 : Datum
276 11199098 : textin(PG_FUNCTION_ARGS)
277 : {
278 11199098 : char *inputText = PG_GETARG_CSTRING(0);
279 :
280 11199098 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
281 : }
282 :
283 : /*
284 : * textout - converts internal representation to cstring
285 : */
286 : Datum
287 4150689 : textout(PG_FUNCTION_ARGS)
288 : {
289 4150689 : Datum txt = PG_GETARG_DATUM(0);
290 :
291 4150689 : PG_RETURN_CSTRING(TextDatumGetCString(txt));
292 : }
293 :
294 : /*
295 : * textrecv - converts external binary format to text
296 : */
297 : Datum
298 24 : textrecv(PG_FUNCTION_ARGS)
299 : {
300 24 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
301 : text *result;
302 : char *str;
303 : int nbytes;
304 :
305 24 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
306 :
307 24 : result = cstring_to_text_with_len(str, nbytes);
308 24 : pfree(str);
309 24 : PG_RETURN_TEXT_P(result);
310 : }
311 :
312 : /*
313 : * textsend - converts text to binary format
314 : */
315 : Datum
316 2359 : textsend(PG_FUNCTION_ARGS)
317 : {
318 2359 : text *t = PG_GETARG_TEXT_PP(0);
319 : StringInfoData buf;
320 :
321 2359 : pq_begintypsend(&buf);
322 2359 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
323 2359 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
324 : }
325 :
326 :
327 : /*
328 : * unknownin - converts cstring to internal representation
329 : */
330 : Datum
331 0 : unknownin(PG_FUNCTION_ARGS)
332 : {
333 0 : char *str = PG_GETARG_CSTRING(0);
334 :
335 : /* representation is same as cstring */
336 0 : PG_RETURN_CSTRING(pstrdup(str));
337 : }
338 :
339 : /*
340 : * unknownout - converts internal representation to cstring
341 : */
342 : Datum
343 470 : unknownout(PG_FUNCTION_ARGS)
344 : {
345 : /* representation is same as cstring */
346 470 : char *str = PG_GETARG_CSTRING(0);
347 :
348 470 : PG_RETURN_CSTRING(pstrdup(str));
349 : }
350 :
351 : /*
352 : * unknownrecv - converts external binary format to unknown
353 : */
354 : Datum
355 0 : unknownrecv(PG_FUNCTION_ARGS)
356 : {
357 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
358 : char *str;
359 : int nbytes;
360 :
361 0 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
362 : /* representation is same as cstring */
363 0 : PG_RETURN_CSTRING(str);
364 : }
365 :
366 : /*
367 : * unknownsend - converts unknown to binary format
368 : */
369 : Datum
370 0 : unknownsend(PG_FUNCTION_ARGS)
371 : {
372 : /* representation is same as cstring */
373 0 : char *str = PG_GETARG_CSTRING(0);
374 : StringInfoData buf;
375 :
376 0 : pq_begintypsend(&buf);
377 0 : pq_sendtext(&buf, str, strlen(str));
378 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
379 : }
380 :
381 :
382 : /* ========== PUBLIC ROUTINES ========== */
383 :
384 : /*
385 : * textlen -
386 : * returns the logical length of a text*
387 : * (which is less than the VARSIZE of the text*)
388 : */
389 : Datum
390 215448 : textlen(PG_FUNCTION_ARGS)
391 : {
392 215448 : Datum str = PG_GETARG_DATUM(0);
393 :
394 : /* try to avoid decompressing argument */
395 215448 : PG_RETURN_INT32(text_length(str));
396 : }
397 :
398 : /*
399 : * text_length -
400 : * Does the real work for textlen()
401 : *
402 : * This is broken out so it can be called directly by other string processing
403 : * functions. Note that the argument is passed as a Datum, to indicate that
404 : * it may still be in compressed form. We can avoid decompressing it at all
405 : * in some cases.
406 : */
407 : static int32
408 215454 : text_length(Datum str)
409 : {
410 : /* fastpath when max encoding length is one */
411 215454 : if (pg_database_encoding_max_length() == 1)
412 10 : return (toast_raw_datum_size(str) - VARHDRSZ);
413 : else
414 : {
415 215444 : text *t = DatumGetTextPP(str);
416 :
417 215444 : return (pg_mbstrlen_with_len(VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t)));
418 : }
419 : }
420 :
421 : /*
422 : * textoctetlen -
423 : * returns the physical length of a text*
424 : * (which is less than the VARSIZE of the text*)
425 : */
426 : Datum
427 35 : textoctetlen(PG_FUNCTION_ARGS)
428 : {
429 35 : Datum str = PG_GETARG_DATUM(0);
430 :
431 : /* We need not detoast the input at all */
432 35 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
433 : }
434 :
435 : /*
436 : * textcat -
437 : * takes two text* and returns a text* that is the concatenation of
438 : * the two.
439 : *
440 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
441 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
442 : * Allocate space for output in all cases.
443 : * XXX - thomas 1997-07-10
444 : */
445 : Datum
446 981602 : textcat(PG_FUNCTION_ARGS)
447 : {
448 981602 : text *t1 = PG_GETARG_TEXT_PP(0);
449 981602 : text *t2 = PG_GETARG_TEXT_PP(1);
450 :
451 981602 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
452 : }
453 :
454 : /*
455 : * text_catenate
456 : * Guts of textcat(), broken out so it can be used by other functions
457 : *
458 : * Arguments can be in short-header form, but not compressed or out-of-line
459 : */
460 : static text *
461 981642 : text_catenate(text *t1, text *t2)
462 : {
463 : text *result;
464 : int len1,
465 : len2,
466 : len;
467 : char *ptr;
468 :
469 981642 : len1 = VARSIZE_ANY_EXHDR(t1);
470 981642 : len2 = VARSIZE_ANY_EXHDR(t2);
471 :
472 : /* paranoia ... probably should throw error instead? */
473 981642 : if (len1 < 0)
474 0 : len1 = 0;
475 981642 : if (len2 < 0)
476 0 : len2 = 0;
477 :
478 981642 : len = len1 + len2 + VARHDRSZ;
479 981642 : result = (text *) palloc(len);
480 :
481 : /* Set size of result string... */
482 981642 : SET_VARSIZE(result, len);
483 :
484 : /* Fill data field of result string... */
485 981642 : ptr = VARDATA(result);
486 981642 : if (len1 > 0)
487 981230 : memcpy(ptr, VARDATA_ANY(t1), len1);
488 981642 : if (len2 > 0)
489 981537 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
490 :
491 981642 : return result;
492 : }
493 :
494 : /*
495 : * charlen_to_bytelen()
496 : * Compute the number of bytes occupied by n characters starting at *p
497 : *
498 : * The caller shall ensure there are n complete characters. Callers achieve
499 : * this by deriving "n" from regmatch_t findings from searching a wchar array.
500 : * pg_mb2wchar_with_len() skips any trailing incomplete character, so regex
501 : * matches will end no later than the last complete character. (The string
502 : * need not be null-terminated.)
503 : */
504 : static int
505 8656 : charlen_to_bytelen(const char *p, int n)
506 : {
507 8656 : if (pg_database_encoding_max_length() == 1)
508 : {
509 : /* Optimization for single-byte encodings */
510 90 : return n;
511 : }
512 : else
513 : {
514 : const char *s;
515 :
516 3031409 : for (s = p; n > 0; n--)
517 3022843 : s += pg_mblen_unbounded(s); /* caller verified encoding */
518 :
519 8566 : return s - p;
520 : }
521 : }
522 :
523 : /*
524 : * text_substr()
525 : * Return a substring starting at the specified position.
526 : * - thomas 1997-12-31
527 : *
528 : * Input:
529 : * - string
530 : * - starting position (is one-based)
531 : * - string length
532 : *
533 : * If the starting position is zero or less, then return from the start of the string
534 : * adjusting the length to be consistent with the "negative start" per SQL.
535 : * If the length is less than zero, return the remaining string.
536 : *
537 : * Added multibyte support.
538 : * - Tatsuo Ishii 1998-4-21
539 : * Changed behavior if starting position is less than one to conform to SQL behavior.
540 : * Formerly returned the entire string; now returns a portion.
541 : * - Thomas Lockhart 1998-12-10
542 : * Now uses faster TOAST-slicing interface
543 : * - John Gray 2002-02-22
544 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
545 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
546 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
547 : * S > LC and < LC + 4 sometimes garbage characters are returned.
548 : * - Joe Conway 2002-08-10
549 : */
550 : Datum
551 331104 : text_substr(PG_FUNCTION_ARGS)
552 : {
553 331104 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
554 : PG_GETARG_INT32(1),
555 : PG_GETARG_INT32(2),
556 : false));
557 : }
558 :
559 : /*
560 : * text_substr_no_len -
561 : * Wrapper to avoid opr_sanity failure due to
562 : * one function accepting a different number of args.
563 : */
564 : Datum
565 18 : text_substr_no_len(PG_FUNCTION_ARGS)
566 : {
567 18 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
568 : PG_GETARG_INT32(1),
569 : -1, true));
570 : }
571 :
572 : /*
573 : * text_substring -
574 : * Does the real work for text_substr() and text_substr_no_len()
575 : *
576 : * This is broken out so it can be called directly by other string processing
577 : * functions. Note that the argument is passed as a Datum, to indicate that
578 : * it may still be in compressed/toasted form. We can avoid detoasting all
579 : * of it in some cases.
580 : *
581 : * The result is always a freshly palloc'd datum.
582 : */
583 : static text *
584 351178 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
585 : {
586 351178 : int32 eml = pg_database_encoding_max_length();
587 351178 : int32 S = start; /* start position */
588 : int32 S1; /* adjusted start position */
589 : int32 L1; /* adjusted substring length */
590 : int32 E; /* end position, exclusive */
591 :
592 : /*
593 : * SQL99 says S can be zero or negative (which we don't document), but we
594 : * still must fetch from the start of the string.
595 : * https://www.postgresql.org/message-id/170905442373.643.11536838320909376197%40wrigleys.postgresql.org
596 : */
597 351178 : S1 = Max(S, 1);
598 :
599 : /* life is easy if the encoding max length is 1 */
600 351178 : if (eml == 1)
601 : {
602 11 : if (length_not_specified) /* special case - get length to end of
603 : * string */
604 0 : L1 = -1;
605 11 : else if (length < 0)
606 : {
607 : /* SQL99 says to throw an error for E < S, i.e., negative length */
608 0 : ereport(ERROR,
609 : (errcode(ERRCODE_SUBSTRING_ERROR),
610 : errmsg("negative substring length not allowed")));
611 : L1 = -1; /* silence stupider compilers */
612 : }
613 11 : else if (pg_add_s32_overflow(S, length, &E))
614 : {
615 : /*
616 : * L could be large enough for S + L to overflow, in which case
617 : * the substring must run to end of string.
618 : */
619 0 : L1 = -1;
620 : }
621 : else
622 : {
623 : /*
624 : * A zero or negative value for the end position can happen if the
625 : * start was negative or one. SQL99 says to return a zero-length
626 : * string.
627 : */
628 11 : if (E < 1)
629 0 : return cstring_to_text("");
630 :
631 11 : L1 = E - S1;
632 : }
633 :
634 : /*
635 : * If the start position is past the end of the string, SQL99 says to
636 : * return a zero-length string -- DatumGetTextPSlice() will do that
637 : * for us. We need only convert S1 to zero-based starting position.
638 : */
639 11 : return DatumGetTextPSlice(str, S1 - 1, L1);
640 : }
641 351167 : else if (eml > 1)
642 : {
643 : /*
644 : * When encoding max length is > 1, we can't get LC without
645 : * detoasting, so we'll grab a conservatively large slice now and go
646 : * back later to do the right thing
647 : */
648 : int32 slice_start;
649 : int32 slice_size;
650 : int32 slice_strlen;
651 : int32 slice_len;
652 : text *slice;
653 : int32 E1;
654 : int32 i;
655 : char *p;
656 : char *s;
657 : text *ret;
658 :
659 : /*
660 : * We need to start at position zero because there is no way to know
661 : * in advance which byte offset corresponds to the supplied start
662 : * position.
663 : */
664 351167 : slice_start = 0;
665 :
666 351167 : if (length_not_specified) /* special case - get length to end of
667 : * string */
668 38 : E = slice_size = L1 = -1;
669 351129 : else if (length < 0)
670 : {
671 : /* SQL99 says to throw an error for E < S, i.e., negative length */
672 6 : ereport(ERROR,
673 : (errcode(ERRCODE_SUBSTRING_ERROR),
674 : errmsg("negative substring length not allowed")));
675 : E = slice_size = L1 = -1; /* silence stupider compilers */
676 : }
677 351123 : else if (pg_add_s32_overflow(S, length, &E))
678 : {
679 : /*
680 : * L could be large enough for S + L to overflow, in which case
681 : * the substring must run to end of string.
682 : */
683 3 : slice_size = L1 = -1;
684 : }
685 : else
686 : {
687 : /*
688 : * Ending at position 1, exclusive, obviously yields an empty
689 : * string. A zero or negative value can happen if the start was
690 : * negative or one. SQL99 says to return a zero-length string.
691 : */
692 351120 : if (E <= 1)
693 6 : return cstring_to_text("");
694 :
695 : /*
696 : * if E is past the end of the string, the tuple toaster will
697 : * truncate the length for us
698 : */
699 351114 : L1 = E - S1;
700 :
701 : /*
702 : * Total slice size in bytes can't be any longer than the
703 : * inclusive end position times the encoding max length. If that
704 : * overflows, we can just use -1.
705 : */
706 351114 : if (pg_mul_s32_overflow(E - 1, eml, &slice_size))
707 3 : slice_size = -1;
708 : }
709 :
710 : /*
711 : * If we're working with an untoasted source, no need to do an extra
712 : * copying step.
713 : */
714 702262 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
715 351107 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
716 201 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
717 : else
718 350954 : slice = (text *) DatumGetPointer(str);
719 :
720 : /* see if we got back an empty string */
721 351155 : slice_len = VARSIZE_ANY_EXHDR(slice);
722 351155 : if (slice_len == 0)
723 : {
724 0 : if (slice != (text *) DatumGetPointer(str))
725 0 : pfree(slice);
726 0 : return cstring_to_text("");
727 : }
728 :
729 : /*
730 : * Now we can get the actual length of the slice in MB characters,
731 : * stopping at the end of the substring. Continuing beyond the
732 : * substring end could find an incomplete character attributable
733 : * solely to DatumGetTextPSlice() chopping in the middle of a
734 : * character, and it would be superfluous work at best.
735 : */
736 351149 : slice_strlen =
737 351155 : (slice_size == -1 ?
738 351155 : pg_mbstrlen_with_len(VARDATA_ANY(slice), slice_len) :
739 351111 : pg_mbcharcliplen_chars(VARDATA_ANY(slice), slice_len, E - 1));
740 :
741 : /*
742 : * Check that the start position wasn't > slice_strlen. If so, SQL99
743 : * says to return a zero-length string.
744 : */
745 351149 : if (S1 > slice_strlen)
746 : {
747 20 : if (slice != (text *) DatumGetPointer(str))
748 3 : pfree(slice);
749 20 : return cstring_to_text("");
750 : }
751 :
752 : /*
753 : * Adjust L1 and E1 now that we know the slice string length. Again
754 : * remember that S1 is one based, and slice_start is zero based.
755 : */
756 351129 : if (L1 > -1)
757 351099 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
758 : else
759 30 : E1 = slice_start + 1 + slice_strlen;
760 :
761 : /*
762 : * Find the start position in the slice; remember S1 is not zero based
763 : */
764 351129 : p = VARDATA_ANY(slice);
765 3379462 : for (i = 0; i < S1 - 1; i++)
766 3028333 : p += pg_mblen_unbounded(p);
767 :
768 : /* hang onto a pointer to our start position */
769 351129 : s = p;
770 :
771 : /*
772 : * Count the actual bytes used by the substring of the requested
773 : * length.
774 : */
775 4982588 : for (i = S1; i < E1; i++)
776 4631459 : p += pg_mblen_unbounded(p);
777 :
778 351129 : ret = (text *) palloc(VARHDRSZ + (p - s));
779 351129 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
780 351129 : memcpy(VARDATA(ret), s, (p - s));
781 :
782 351129 : if (slice != (text *) DatumGetPointer(str))
783 195 : pfree(slice);
784 :
785 351129 : return ret;
786 : }
787 : else
788 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
789 :
790 : /* not reached: suppress compiler warning */
791 : return NULL;
792 : }
793 :
794 : /*
795 : * pg_mbcharcliplen_chars -
796 : * Mirror pg_mbcharcliplen(), except return value unit is chars, not bytes.
797 : *
798 : * This mirrors all the dubious historical behavior, so it's static to
799 : * discourage proliferation. The assertions are specific to the one caller.
800 : */
801 : static int
802 351111 : pg_mbcharcliplen_chars(const char *mbstr, int len, int limit)
803 : {
804 351111 : int nch = 0;
805 : int l;
806 :
807 : Assert(len > 0);
808 : Assert(limit > 0);
809 : Assert(pg_database_encoding_max_length() > 1);
810 :
811 6484001 : while (len > 0 && *mbstr)
812 : {
813 6483686 : l = pg_mblen_with_len(mbstr, len);
814 6483680 : nch++;
815 6483680 : if (nch == limit)
816 350790 : break;
817 6132890 : len -= l;
818 6132890 : mbstr += l;
819 : }
820 351105 : return nch;
821 : }
822 :
823 : /*
824 : * textoverlay
825 : * Replace specified substring of first string with second
826 : *
827 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
828 : * This code is a direct implementation of what the standard says.
829 : */
830 : Datum
831 14 : textoverlay(PG_FUNCTION_ARGS)
832 : {
833 14 : text *t1 = PG_GETARG_TEXT_PP(0);
834 14 : text *t2 = PG_GETARG_TEXT_PP(1);
835 14 : int sp = PG_GETARG_INT32(2); /* substring start position */
836 14 : int sl = PG_GETARG_INT32(3); /* substring length */
837 :
838 14 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
839 : }
840 :
841 : Datum
842 6 : textoverlay_no_len(PG_FUNCTION_ARGS)
843 : {
844 6 : text *t1 = PG_GETARG_TEXT_PP(0);
845 6 : text *t2 = PG_GETARG_TEXT_PP(1);
846 6 : int sp = PG_GETARG_INT32(2); /* substring start position */
847 : int sl;
848 :
849 6 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
850 6 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
851 : }
852 :
853 : static text *
854 20 : text_overlay(text *t1, text *t2, int sp, int sl)
855 : {
856 : text *result;
857 : text *s1;
858 : text *s2;
859 : int sp_pl_sl;
860 :
861 : /*
862 : * Check for possible integer-overflow cases. For negative sp, throw a
863 : * "substring length" error because that's what should be expected
864 : * according to the spec's definition of OVERLAY().
865 : */
866 20 : if (sp <= 0)
867 0 : ereport(ERROR,
868 : (errcode(ERRCODE_SUBSTRING_ERROR),
869 : errmsg("negative substring length not allowed")));
870 20 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
871 0 : ereport(ERROR,
872 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
873 : errmsg("integer out of range")));
874 :
875 20 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
876 20 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
877 20 : result = text_catenate(s1, t2);
878 20 : result = text_catenate(result, s2);
879 :
880 20 : return result;
881 : }
882 :
883 : /*
884 : * textpos -
885 : * Return the position of the specified substring.
886 : * Implements the SQL POSITION() function.
887 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
888 : * - thomas 1997-07-27
889 : */
890 : Datum
891 68 : textpos(PG_FUNCTION_ARGS)
892 : {
893 68 : text *str = PG_GETARG_TEXT_PP(0);
894 68 : text *search_str = PG_GETARG_TEXT_PP(1);
895 :
896 68 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
897 : }
898 :
899 : /*
900 : * text_position -
901 : * Does the real work for textpos()
902 : *
903 : * Inputs:
904 : * t1 - string to be searched
905 : * t2 - pattern to match within t1
906 : * Result:
907 : * Character index of the first matched char, starting from 1,
908 : * or 0 if no match.
909 : *
910 : * This is broken out so it can be called directly by other string processing
911 : * functions.
912 : */
913 : static int
914 68 : text_position(text *t1, text *t2, Oid collid)
915 : {
916 : TextPositionState state;
917 : int result;
918 :
919 68 : check_collation_set(collid);
920 :
921 : /* Empty needle always matches at position 1 */
922 68 : if (VARSIZE_ANY_EXHDR(t2) < 1)
923 6 : return 1;
924 :
925 : /* Otherwise, can't match if haystack is shorter than needle */
926 62 : if (VARSIZE_ANY_EXHDR(t1) < VARSIZE_ANY_EXHDR(t2) &&
927 11 : pg_newlocale_from_collation(collid)->deterministic)
928 11 : return 0;
929 :
930 51 : text_position_setup(t1, t2, collid, &state);
931 : /* don't need greedy mode here */
932 51 : state.greedy = false;
933 :
934 51 : if (!text_position_next(&state))
935 12 : result = 0;
936 : else
937 39 : result = text_position_get_match_pos(&state);
938 51 : text_position_cleanup(&state);
939 51 : return result;
940 : }
941 :
942 :
943 : /*
944 : * text_position_setup, text_position_next, text_position_cleanup -
945 : * Component steps of text_position()
946 : *
947 : * These are broken out so that a string can be efficiently searched for
948 : * multiple occurrences of the same pattern. text_position_next may be
949 : * called multiple times, and it advances to the next match on each call.
950 : * text_position_get_match_ptr() and text_position_get_match_pos() return
951 : * a pointer or 1-based character position of the last match, respectively.
952 : *
953 : * The "state" variable is normally just a local variable in the caller.
954 : *
955 : * NOTE: text_position_next skips over the matched portion. For example,
956 : * searching for "xx" in "xxx" returns only one match, not two.
957 : */
958 :
959 : static void
960 965 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
961 : {
962 965 : int len1 = VARSIZE_ANY_EXHDR(t1);
963 965 : int len2 = VARSIZE_ANY_EXHDR(t2);
964 :
965 965 : check_collation_set(collid);
966 :
967 965 : state->locale = pg_newlocale_from_collation(collid);
968 :
969 : /*
970 : * Most callers need greedy mode, but some might want to unset this to
971 : * optimize.
972 : */
973 965 : state->greedy = true;
974 :
975 : Assert(len2 > 0);
976 :
977 : /*
978 : * Even with a multi-byte encoding, we perform the search using the raw
979 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
980 : * because in UTF-8 the byte sequence of one character cannot contain
981 : * another character. For other multi-byte encodings, we do the search
982 : * initially as a simple byte search, ignoring multibyte issues, but
983 : * verify afterwards that the match we found is at a character boundary,
984 : * and continue the search if it was a false match.
985 : */
986 965 : if (pg_database_encoding_max_length() == 1)
987 54 : state->is_multibyte_char_in_char = false;
988 911 : else if (GetDatabaseEncoding() == PG_UTF8)
989 911 : state->is_multibyte_char_in_char = false;
990 : else
991 0 : state->is_multibyte_char_in_char = true;
992 :
993 965 : state->str1 = VARDATA_ANY(t1);
994 965 : state->str2 = VARDATA_ANY(t2);
995 965 : state->len1 = len1;
996 965 : state->len2 = len2;
997 965 : state->last_match = NULL;
998 965 : state->refpoint = state->str1;
999 965 : state->refpos = 0;
1000 :
1001 : /*
1002 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
1003 : * notes we use the terminology that the "haystack" is the string to be
1004 : * searched (t1) and the "needle" is the pattern being sought (t2).
1005 : *
1006 : * If the needle is empty or bigger than the haystack then there is no
1007 : * point in wasting cycles initializing the table. We also choose not to
1008 : * use B-M-H for needles of length 1, since the skip table can't possibly
1009 : * save anything in that case.
1010 : *
1011 : * (With nondeterministic collations, the search is already
1012 : * multibyte-aware, so we don't need this.)
1013 : */
1014 965 : if (len1 >= len2 && len2 > 1 && state->locale->deterministic)
1015 : {
1016 798 : int searchlength = len1 - len2;
1017 : int skiptablemask;
1018 : int last;
1019 : int i;
1020 798 : const char *str2 = state->str2;
1021 :
1022 : /*
1023 : * First we must determine how much of the skip table to use. The
1024 : * declaration of TextPositionState allows up to 256 elements, but for
1025 : * short search problems we don't really want to have to initialize so
1026 : * many elements --- it would take too long in comparison to the
1027 : * actual search time. So we choose a useful skip table size based on
1028 : * the haystack length minus the needle length. The closer the needle
1029 : * length is to the haystack length the less useful skipping becomes.
1030 : *
1031 : * Note: since we use bit-masking to select table elements, the skip
1032 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
1033 : */
1034 798 : if (searchlength < 16)
1035 57 : skiptablemask = 3;
1036 741 : else if (searchlength < 64)
1037 17 : skiptablemask = 7;
1038 724 : else if (searchlength < 128)
1039 13 : skiptablemask = 15;
1040 711 : else if (searchlength < 512)
1041 166 : skiptablemask = 31;
1042 545 : else if (searchlength < 2048)
1043 404 : skiptablemask = 63;
1044 141 : else if (searchlength < 4096)
1045 99 : skiptablemask = 127;
1046 : else
1047 42 : skiptablemask = 255;
1048 798 : state->skiptablemask = skiptablemask;
1049 :
1050 : /*
1051 : * Initialize the skip table. We set all elements to the needle
1052 : * length, since this is the correct skip distance for any character
1053 : * not found in the needle.
1054 : */
1055 55962 : for (i = 0; i <= skiptablemask; i++)
1056 55164 : state->skiptable[i] = len2;
1057 :
1058 : /*
1059 : * Now examine the needle. For each character except the last one,
1060 : * set the corresponding table element to the appropriate skip
1061 : * distance. Note that when two characters share the same skip table
1062 : * entry, the one later in the needle must determine the skip
1063 : * distance.
1064 : */
1065 798 : last = len2 - 1;
1066 :
1067 10181 : for (i = 0; i < last; i++)
1068 9383 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1069 : }
1070 965 : }
1071 :
1072 : /*
1073 : * Advance to the next match, starting from the end of the previous match
1074 : * (or the beginning of the string, on first call). Returns true if a match
1075 : * is found.
1076 : *
1077 : * Note that this refuses to match an empty-string needle. Most callers
1078 : * will have handled that case specially and we'll never see it here.
1079 : */
1080 : static bool
1081 4886 : text_position_next(TextPositionState *state)
1082 : {
1083 4886 : int needle_len = state->len2;
1084 : char *start_ptr;
1085 : char *matchptr;
1086 :
1087 4886 : if (needle_len <= 0)
1088 0 : return false; /* result for empty pattern */
1089 :
1090 : /* Start from the point right after the previous match. */
1091 4886 : if (state->last_match)
1092 3915 : start_ptr = state->last_match + state->last_match_len;
1093 : else
1094 971 : start_ptr = state->str1;
1095 :
1096 4886 : retry:
1097 4886 : matchptr = text_position_next_internal(start_ptr, state);
1098 :
1099 4886 : if (!matchptr)
1100 920 : return false;
1101 :
1102 : /*
1103 : * Found a match for the byte sequence. If this is a multibyte encoding,
1104 : * where one character's byte sequence can appear inside a longer
1105 : * multi-byte character, we need to verify that the match was at a
1106 : * character boundary, not in the middle of a multi-byte character.
1107 : */
1108 3966 : if (state->is_multibyte_char_in_char && state->locale->deterministic)
1109 : {
1110 0 : const char *haystack_end = state->str1 + state->len1;
1111 :
1112 : /* Walk one character at a time, until we reach the match. */
1113 :
1114 : /* the search should never move backwards. */
1115 : Assert(state->refpoint <= matchptr);
1116 :
1117 0 : while (state->refpoint < matchptr)
1118 : {
1119 : /* step to next character. */
1120 0 : state->refpoint += pg_mblen_range(state->refpoint, haystack_end);
1121 0 : state->refpos++;
1122 :
1123 : /*
1124 : * If we stepped over the match's start position, then it was a
1125 : * false positive, where the byte sequence appeared in the middle
1126 : * of a multi-byte character. Skip it, and continue the search at
1127 : * the next character boundary.
1128 : */
1129 0 : if (state->refpoint > matchptr)
1130 : {
1131 0 : start_ptr = state->refpoint;
1132 0 : goto retry;
1133 : }
1134 : }
1135 : }
1136 :
1137 3966 : state->last_match = matchptr;
1138 3966 : state->last_match_len = state->last_match_len_tmp;
1139 3966 : return true;
1140 : }
1141 :
1142 : /*
1143 : * Subroutine of text_position_next(). This searches for the raw byte
1144 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1145 : * match starting at 'start_ptr', or NULL if no match is found.
1146 : */
1147 : static char *
1148 4886 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1149 : {
1150 4886 : int haystack_len = state->len1;
1151 4886 : int needle_len = state->len2;
1152 4886 : int skiptablemask = state->skiptablemask;
1153 4886 : const char *haystack = state->str1;
1154 4886 : const char *needle = state->str2;
1155 4886 : const char *haystack_end = &haystack[haystack_len];
1156 : const char *hptr;
1157 :
1158 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1159 : Assert(needle_len > 0);
1160 :
1161 4886 : state->last_match_len_tmp = needle_len;
1162 :
1163 4886 : if (!state->locale->deterministic)
1164 : {
1165 : /*
1166 : * With a nondeterministic collation, we have to use an unoptimized
1167 : * route. We walk through the haystack and see if at each position
1168 : * there is a substring of the remaining string that is equal to the
1169 : * needle under the given collation.
1170 : *
1171 : * Note, the found substring could have a different length than the
1172 : * needle. Callers that want to skip over the found string need to
1173 : * read the length of the found substring from last_match_len rather
1174 : * than just using the length of their needle.
1175 : *
1176 : * Most callers will require "greedy" semantics, meaning that we need
1177 : * to find the longest such substring, not the shortest. For callers
1178 : * that don't need greedy semantics, we can finish on the first match.
1179 : *
1180 : * This loop depends on the assumption that the needle is nonempty and
1181 : * any matching substring must also be nonempty. (Even if the
1182 : * collation would accept an empty match, returning one would send
1183 : * callers that search for successive matches into an infinite loop.)
1184 : */
1185 126 : const char *result_hptr = NULL;
1186 :
1187 126 : hptr = start_ptr;
1188 339 : while (hptr < haystack_end)
1189 : {
1190 : const char *test_end;
1191 :
1192 : /*
1193 : * First check the common case that there is a match in the
1194 : * haystack of exactly the length of the needle.
1195 : */
1196 282 : if (!state->greedy &&
1197 54 : haystack_end - hptr >= needle_len &&
1198 27 : pg_strncoll(hptr, needle_len, needle, needle_len, state->locale) == 0)
1199 6 : return (char *) hptr;
1200 :
1201 : /*
1202 : * Else check if any of the non-empty substrings starting at hptr
1203 : * compare equal to the needle.
1204 : */
1205 276 : test_end = hptr;
1206 : do
1207 : {
1208 1077 : test_end += pg_mblen_range(test_end, haystack_end);
1209 1077 : if (pg_strncoll(hptr, (test_end - hptr), needle, needle_len, state->locale) == 0)
1210 : {
1211 69 : state->last_match_len_tmp = (test_end - hptr);
1212 69 : result_hptr = hptr;
1213 69 : if (!state->greedy)
1214 0 : break;
1215 : }
1216 1077 : } while (test_end < haystack_end);
1217 :
1218 276 : if (result_hptr)
1219 63 : break;
1220 :
1221 213 : hptr += pg_mblen_range(hptr, haystack_end);
1222 : }
1223 :
1224 120 : return (char *) result_hptr;
1225 : }
1226 4760 : else if (needle_len == 1)
1227 : {
1228 : /* No point in using B-M-H for a one-character needle */
1229 380 : char nchar = *needle;
1230 :
1231 380 : hptr = start_ptr;
1232 2939 : while (hptr < haystack_end)
1233 : {
1234 2856 : if (*hptr == nchar)
1235 297 : return (char *) hptr;
1236 2559 : hptr++;
1237 : }
1238 : }
1239 : else
1240 : {
1241 4380 : const char *needle_last = &needle[needle_len - 1];
1242 :
1243 : /* Start at startpos plus the length of the needle */
1244 4380 : hptr = start_ptr + needle_len - 1;
1245 108360 : while (hptr < haystack_end)
1246 : {
1247 : /* Match the needle scanning *backward* */
1248 : const char *nptr;
1249 : const char *p;
1250 :
1251 107580 : nptr = needle_last;
1252 107580 : p = hptr;
1253 161573 : while (*nptr == *p)
1254 : {
1255 : /* Matched it all? If so, return 1-based position */
1256 57593 : if (nptr == needle)
1257 3600 : return (char *) p;
1258 53993 : nptr--, p--;
1259 : }
1260 :
1261 : /*
1262 : * No match, so use the haystack char at hptr to decide how far to
1263 : * advance. If the needle had any occurrence of that character
1264 : * (or more precisely, one sharing the same skiptable entry)
1265 : * before its last character, then we advance far enough to align
1266 : * the last such needle character with that haystack position.
1267 : * Otherwise we can advance by the whole needle length.
1268 : */
1269 103980 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1270 : }
1271 : }
1272 :
1273 863 : return 0; /* not found */
1274 : }
1275 :
1276 : /*
1277 : * Return a pointer to the current match.
1278 : *
1279 : * The returned pointer points into the original haystack string.
1280 : */
1281 : static char *
1282 3912 : text_position_get_match_ptr(TextPositionState *state)
1283 : {
1284 3912 : return state->last_match;
1285 : }
1286 :
1287 : /*
1288 : * Return the offset of the current match.
1289 : *
1290 : * The offset is in characters, 1-based.
1291 : */
1292 : static int
1293 39 : text_position_get_match_pos(TextPositionState *state)
1294 : {
1295 : /* Convert the byte position to char position. */
1296 78 : state->refpos += pg_mbstrlen_with_len(state->refpoint,
1297 39 : state->last_match - state->refpoint);
1298 39 : state->refpoint = state->last_match;
1299 39 : return state->refpos + 1;
1300 : }
1301 :
1302 : /*
1303 : * Reset search state to the initial state installed by text_position_setup.
1304 : *
1305 : * The next call to text_position_next will search from the beginning
1306 : * of the string.
1307 : */
1308 : static void
1309 6 : text_position_reset(TextPositionState *state)
1310 : {
1311 6 : state->last_match = NULL;
1312 6 : state->refpoint = state->str1;
1313 6 : state->refpos = 0;
1314 6 : }
1315 :
1316 : static void
1317 965 : text_position_cleanup(TextPositionState *state)
1318 : {
1319 : /* no cleanup needed */
1320 965 : }
1321 :
1322 :
1323 : static void
1324 8697842 : check_collation_set(Oid collid)
1325 : {
1326 8697842 : if (!OidIsValid(collid))
1327 : {
1328 : /*
1329 : * This typically means that the parser could not resolve a conflict
1330 : * of implicit collations, so report it that way.
1331 : */
1332 15 : ereport(ERROR,
1333 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1334 : errmsg("could not determine which collation to use for string comparison"),
1335 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1336 : }
1337 8697827 : }
1338 :
1339 : /*
1340 : * varstr_cmp()
1341 : *
1342 : * Comparison function for text strings with given lengths, using the
1343 : * appropriate locale. Returns an integer less than, equal to, or greater than
1344 : * zero, indicating whether arg1 is less than, equal to, or greater than arg2.
1345 : *
1346 : * Note: many functions that depend on this are marked leakproof; therefore,
1347 : * avoid reporting the actual contents of the input when throwing errors.
1348 : * All errors herein should be things that can't happen except on corrupt
1349 : * data, anyway; otherwise we will have trouble with indexing strings that
1350 : * would cause them.
1351 : */
1352 : int
1353 4931775 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1354 : {
1355 : int result;
1356 : pg_locale_t mylocale;
1357 :
1358 4931775 : check_collation_set(collid);
1359 :
1360 4931766 : mylocale = pg_newlocale_from_collation(collid);
1361 :
1362 4931766 : if (mylocale->collate_is_c)
1363 : {
1364 1900777 : result = memcmp(arg1, arg2, Min(len1, len2));
1365 1900777 : if ((result == 0) && (len1 != len2))
1366 67130 : result = (len1 < len2) ? -1 : 1;
1367 : }
1368 : else
1369 : {
1370 : /*
1371 : * memcmp() can't tell us which of two unequal strings sorts first,
1372 : * but it's a cheap way to tell if they're equal. Testing shows that
1373 : * memcmp() followed by strcoll() is only trivially slower than
1374 : * strcoll() by itself, so we don't lose much if this doesn't work out
1375 : * very often, and if it does - for example, because there are many
1376 : * equal strings in the input - then we win big by avoiding expensive
1377 : * collation-aware comparisons.
1378 : */
1379 3030989 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1380 783435 : return 0;
1381 :
1382 2247554 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1383 :
1384 : /* Break tie if necessary. */
1385 2247554 : if (result == 0 && mylocale->deterministic)
1386 : {
1387 0 : result = memcmp(arg1, arg2, Min(len1, len2));
1388 0 : if ((result == 0) && (len1 != len2))
1389 0 : result = (len1 < len2) ? -1 : 1;
1390 : }
1391 : }
1392 :
1393 4148331 : return result;
1394 : }
1395 :
1396 : /* text_cmp()
1397 : * Internal comparison function for text strings.
1398 : * Returns -1, 0 or 1
1399 : */
1400 : static int
1401 3960253 : text_cmp(text *arg1, text *arg2, Oid collid)
1402 : {
1403 : char *a1p,
1404 : *a2p;
1405 : int len1,
1406 : len2;
1407 :
1408 3960253 : a1p = VARDATA_ANY(arg1);
1409 3960253 : a2p = VARDATA_ANY(arg2);
1410 :
1411 3960253 : len1 = VARSIZE_ANY_EXHDR(arg1);
1412 3960253 : len2 = VARSIZE_ANY_EXHDR(arg2);
1413 :
1414 3960253 : return varstr_cmp(a1p, len1, a2p, len2, collid);
1415 : }
1416 :
1417 : /*
1418 : * Comparison functions for text strings.
1419 : *
1420 : * Note: btree indexes need these routines not to leak memory; therefore,
1421 : * be careful to free working copies of toasted datums. Most places don't
1422 : * need to be so careful.
1423 : */
1424 :
1425 : Datum
1426 3350390 : texteq(PG_FUNCTION_ARGS)
1427 : {
1428 3350390 : Oid collid = PG_GET_COLLATION();
1429 3350390 : pg_locale_t mylocale = 0;
1430 : bool result;
1431 :
1432 3350390 : check_collation_set(collid);
1433 :
1434 3350390 : mylocale = pg_newlocale_from_collation(collid);
1435 :
1436 3350390 : if (mylocale->deterministic)
1437 : {
1438 3348210 : Datum arg1 = PG_GETARG_DATUM(0);
1439 3348210 : Datum arg2 = PG_GETARG_DATUM(1);
1440 : Size len1,
1441 : len2;
1442 :
1443 : /*
1444 : * Since we only care about equality or not-equality, we can avoid all
1445 : * the expense of strcoll() here, and just do bitwise comparison. In
1446 : * fact, we don't even have to do a bitwise comparison if we can show
1447 : * the lengths of the strings are unequal; which might save us from
1448 : * having to detoast one or both values.
1449 : */
1450 3348210 : len1 = toast_raw_datum_size(arg1);
1451 3348210 : len2 = toast_raw_datum_size(arg2);
1452 3348210 : if (len1 != len2)
1453 1592363 : result = false;
1454 : else
1455 : {
1456 1755847 : text *targ1 = DatumGetTextPP(arg1);
1457 1755847 : text *targ2 = DatumGetTextPP(arg2);
1458 :
1459 1755847 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1460 : len1 - VARHDRSZ) == 0);
1461 :
1462 1755847 : PG_FREE_IF_COPY(targ1, 0);
1463 1755847 : PG_FREE_IF_COPY(targ2, 1);
1464 : }
1465 : }
1466 : else
1467 : {
1468 2180 : text *arg1 = PG_GETARG_TEXT_PP(0);
1469 2180 : text *arg2 = PG_GETARG_TEXT_PP(1);
1470 :
1471 2180 : result = (text_cmp(arg1, arg2, collid) == 0);
1472 :
1473 2180 : PG_FREE_IF_COPY(arg1, 0);
1474 2180 : PG_FREE_IF_COPY(arg2, 1);
1475 : }
1476 :
1477 3350390 : PG_RETURN_BOOL(result);
1478 : }
1479 :
1480 : Datum
1481 204301 : textne(PG_FUNCTION_ARGS)
1482 : {
1483 204301 : Oid collid = PG_GET_COLLATION();
1484 : pg_locale_t mylocale;
1485 : bool result;
1486 :
1487 204301 : check_collation_set(collid);
1488 :
1489 204301 : mylocale = pg_newlocale_from_collation(collid);
1490 :
1491 204301 : if (mylocale->deterministic)
1492 : {
1493 204289 : Datum arg1 = PG_GETARG_DATUM(0);
1494 204289 : Datum arg2 = PG_GETARG_DATUM(1);
1495 : Size len1,
1496 : len2;
1497 :
1498 : /* See comment in texteq() */
1499 204289 : len1 = toast_raw_datum_size(arg1);
1500 204289 : len2 = toast_raw_datum_size(arg2);
1501 204289 : if (len1 != len2)
1502 11185 : result = true;
1503 : else
1504 : {
1505 193104 : text *targ1 = DatumGetTextPP(arg1);
1506 193104 : text *targ2 = DatumGetTextPP(arg2);
1507 :
1508 193104 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1509 : len1 - VARHDRSZ) != 0);
1510 :
1511 193104 : PG_FREE_IF_COPY(targ1, 0);
1512 193104 : PG_FREE_IF_COPY(targ2, 1);
1513 : }
1514 : }
1515 : else
1516 : {
1517 12 : text *arg1 = PG_GETARG_TEXT_PP(0);
1518 12 : text *arg2 = PG_GETARG_TEXT_PP(1);
1519 :
1520 12 : result = (text_cmp(arg1, arg2, collid) != 0);
1521 :
1522 12 : PG_FREE_IF_COPY(arg1, 0);
1523 12 : PG_FREE_IF_COPY(arg2, 1);
1524 : }
1525 :
1526 204301 : PG_RETURN_BOOL(result);
1527 : }
1528 :
1529 : Datum
1530 104647 : text_lt(PG_FUNCTION_ARGS)
1531 : {
1532 104647 : text *arg1 = PG_GETARG_TEXT_PP(0);
1533 104647 : text *arg2 = PG_GETARG_TEXT_PP(1);
1534 : bool result;
1535 :
1536 104647 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1537 :
1538 104638 : PG_FREE_IF_COPY(arg1, 0);
1539 104638 : PG_FREE_IF_COPY(arg2, 1);
1540 :
1541 104638 : PG_RETURN_BOOL(result);
1542 : }
1543 :
1544 : Datum
1545 158863 : text_le(PG_FUNCTION_ARGS)
1546 : {
1547 158863 : text *arg1 = PG_GETARG_TEXT_PP(0);
1548 158863 : text *arg2 = PG_GETARG_TEXT_PP(1);
1549 : bool result;
1550 :
1551 158863 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1552 :
1553 158863 : PG_FREE_IF_COPY(arg1, 0);
1554 158863 : PG_FREE_IF_COPY(arg2, 1);
1555 :
1556 158863 : PG_RETURN_BOOL(result);
1557 : }
1558 :
1559 : Datum
1560 98038 : text_gt(PG_FUNCTION_ARGS)
1561 : {
1562 98038 : text *arg1 = PG_GETARG_TEXT_PP(0);
1563 98038 : text *arg2 = PG_GETARG_TEXT_PP(1);
1564 : bool result;
1565 :
1566 98038 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1567 :
1568 98038 : PG_FREE_IF_COPY(arg1, 0);
1569 98038 : PG_FREE_IF_COPY(arg2, 1);
1570 :
1571 98038 : PG_RETURN_BOOL(result);
1572 : }
1573 :
1574 : Datum
1575 87860 : text_ge(PG_FUNCTION_ARGS)
1576 : {
1577 87860 : text *arg1 = PG_GETARG_TEXT_PP(0);
1578 87860 : text *arg2 = PG_GETARG_TEXT_PP(1);
1579 : bool result;
1580 :
1581 87860 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1582 :
1583 87860 : PG_FREE_IF_COPY(arg1, 0);
1584 87860 : PG_FREE_IF_COPY(arg2, 1);
1585 :
1586 87860 : PG_RETURN_BOOL(result);
1587 : }
1588 :
1589 : Datum
1590 18957 : text_starts_with(PG_FUNCTION_ARGS)
1591 : {
1592 18957 : Datum arg1 = PG_GETARG_DATUM(0);
1593 18957 : Datum arg2 = PG_GETARG_DATUM(1);
1594 18957 : Oid collid = PG_GET_COLLATION();
1595 : pg_locale_t mylocale;
1596 : bool result;
1597 : Size len1,
1598 : len2;
1599 :
1600 18957 : check_collation_set(collid);
1601 :
1602 18957 : mylocale = pg_newlocale_from_collation(collid);
1603 :
1604 18957 : if (!mylocale->deterministic)
1605 0 : ereport(ERROR,
1606 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1607 : errmsg("nondeterministic collations are not supported for substring searches")));
1608 :
1609 18957 : len1 = toast_raw_datum_size(arg1);
1610 18957 : len2 = toast_raw_datum_size(arg2);
1611 18957 : if (len2 > len1)
1612 0 : result = false;
1613 : else
1614 : {
1615 18957 : text *targ1 = text_substring(arg1, 1, len2, false);
1616 18957 : text *targ2 = DatumGetTextPP(arg2);
1617 :
1618 18957 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1619 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1620 :
1621 18957 : PG_FREE_IF_COPY(targ1, 0);
1622 18957 : PG_FREE_IF_COPY(targ2, 1);
1623 : }
1624 :
1625 18957 : PG_RETURN_BOOL(result);
1626 : }
1627 :
1628 : Datum
1629 3350835 : bttextcmp(PG_FUNCTION_ARGS)
1630 : {
1631 3350835 : text *arg1 = PG_GETARG_TEXT_PP(0);
1632 3350835 : text *arg2 = PG_GETARG_TEXT_PP(1);
1633 : int32 result;
1634 :
1635 3350835 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1636 :
1637 3350835 : PG_FREE_IF_COPY(arg1, 0);
1638 3350835 : PG_FREE_IF_COPY(arg2, 1);
1639 :
1640 3350835 : PG_RETURN_INT32(result);
1641 : }
1642 :
1643 : Datum
1644 44518 : bttextsortsupport(PG_FUNCTION_ARGS)
1645 : {
1646 44518 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1647 44518 : Oid collid = ssup->ssup_collation;
1648 : MemoryContext oldcontext;
1649 :
1650 44518 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1651 :
1652 : /* Use generic string SortSupport */
1653 44518 : varstr_sortsupport(ssup, TEXTOID, collid);
1654 :
1655 44512 : MemoryContextSwitchTo(oldcontext);
1656 :
1657 44512 : PG_RETURN_VOID();
1658 : }
1659 :
1660 : /*
1661 : * Generic sortsupport interface for character type's operator classes.
1662 : * Includes locale support, and support for BpChar semantics (i.e. removing
1663 : * trailing spaces before comparison).
1664 : *
1665 : * Relies on the assumption that text, VarChar, and BpChar all have the
1666 : * same representation.
1667 : */
1668 : void
1669 70282 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1670 : {
1671 70282 : bool abbreviate = ssup->abbreviate;
1672 70282 : bool collate_c = false;
1673 : VarStringSortSupport *sss;
1674 : pg_locale_t locale;
1675 :
1676 70282 : check_collation_set(collid);
1677 :
1678 70276 : locale = pg_newlocale_from_collation(collid);
1679 :
1680 : /*
1681 : * If possible, set ssup->comparator to a function which can be used to
1682 : * directly compare two datums. If we can do this, we'll avoid the
1683 : * overhead of a trip through the fmgr layer for every comparison, which
1684 : * can be substantial.
1685 : *
1686 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1687 : * which uses strcoll() to perform comparisons. We use that for the
1688 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1689 : * LC_COLLATE = C, we can make things quite a bit faster with
1690 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1691 : * memcmp() rather than strcoll().
1692 : */
1693 70276 : if (locale->collate_is_c)
1694 : {
1695 46927 : if (typid == BPCHAROID)
1696 154 : ssup->comparator = bpcharfastcmp_c;
1697 46773 : else if (typid == NAMEOID)
1698 : {
1699 25242 : ssup->comparator = namefastcmp_c;
1700 : /* Not supporting abbreviation with type NAME, for now */
1701 25242 : abbreviate = false;
1702 : }
1703 : else
1704 21531 : ssup->comparator = varstrfastcmp_c;
1705 :
1706 46927 : collate_c = true;
1707 : }
1708 : else
1709 : {
1710 : /*
1711 : * We use varlenafastcmp_locale except for type NAME.
1712 : */
1713 23349 : if (typid == NAMEOID)
1714 : {
1715 0 : ssup->comparator = namefastcmp_locale;
1716 : /* Not supporting abbreviation with type NAME, for now */
1717 0 : abbreviate = false;
1718 : }
1719 : else
1720 23349 : ssup->comparator = varlenafastcmp_locale;
1721 :
1722 : /*
1723 : * Unfortunately, it seems that abbreviation for non-C collations is
1724 : * broken on many common platforms; see pg_strxfrm_enabled().
1725 : *
1726 : * Even apart from the risk of broken locales, it's possible that
1727 : * there are platforms where the use of abbreviated keys should be
1728 : * disabled at compile time. For example, macOS's strxfrm()
1729 : * implementation is known to not effectively concentrate a
1730 : * significant amount of entropy from the original string in earlier
1731 : * transformed blobs. It's possible that other supported platforms
1732 : * are similarly encumbered. So, if we ever get past disabling this
1733 : * categorically, we may still want or need to disable it for
1734 : * particular platforms.
1735 : */
1736 23349 : if (!pg_strxfrm_enabled(locale))
1737 22951 : abbreviate = false;
1738 : }
1739 :
1740 : /*
1741 : * If we're using abbreviated keys, or if we're using a locale-aware
1742 : * comparison, we need to initialize a VarStringSortSupport object. Both
1743 : * cases will make use of the temporary buffers we initialize here for
1744 : * scratch space (and to detect requirement for BpChar semantics from
1745 : * caller), and the abbreviation case requires additional state.
1746 : */
1747 70276 : if (abbreviate || !collate_c)
1748 : {
1749 35378 : sss = palloc_object(VarStringSortSupport);
1750 35378 : sss->buf1 = palloc(TEXTBUFLEN);
1751 35378 : sss->buflen1 = TEXTBUFLEN;
1752 35378 : sss->buf2 = palloc(TEXTBUFLEN);
1753 35378 : sss->buflen2 = TEXTBUFLEN;
1754 : /* Start with invalid values */
1755 35378 : sss->last_len1 = -1;
1756 35378 : sss->last_len2 = -1;
1757 : /* Initialize */
1758 35378 : sss->last_returned = 0;
1759 35378 : if (collate_c)
1760 12029 : sss->locale = NULL;
1761 : else
1762 23349 : sss->locale = locale;
1763 :
1764 : /*
1765 : * To avoid somehow confusing a strxfrm() blob and an original string,
1766 : * constantly keep track of the variety of data that buf1 and buf2
1767 : * currently contain.
1768 : *
1769 : * Comparisons may be interleaved with conversion calls. Frequently,
1770 : * conversions and comparisons are batched into two distinct phases,
1771 : * but the correctness of caching cannot hinge upon this. For
1772 : * comparison caching, buffer state is only trusted if cache_blob is
1773 : * found set to false, whereas strxfrm() caching only trusts the state
1774 : * when cache_blob is found set to true.
1775 : *
1776 : * Arbitrarily initialize cache_blob to true.
1777 : */
1778 35378 : sss->cache_blob = true;
1779 35378 : sss->collate_c = collate_c;
1780 35378 : sss->typid = typid;
1781 35378 : ssup->ssup_extra = sss;
1782 :
1783 : /*
1784 : * If possible, plan to use the abbreviated keys optimization. The
1785 : * core code may switch back to authoritative comparator should
1786 : * abbreviation be aborted.
1787 : */
1788 35378 : if (abbreviate)
1789 : {
1790 12328 : sss->prop_card = 0.20;
1791 12328 : initHyperLogLog(&sss->abbr_card, 10);
1792 12328 : initHyperLogLog(&sss->full_card, 10);
1793 12328 : ssup->abbrev_full_comparator = ssup->comparator;
1794 12328 : ssup->comparator = ssup_datum_unsigned_cmp;
1795 12328 : ssup->abbrev_converter = varstr_abbrev_convert;
1796 12328 : ssup->abbrev_abort = varstr_abbrev_abort;
1797 : }
1798 : }
1799 70276 : }
1800 :
1801 : /*
1802 : * sortsupport comparison func (for C locale case)
1803 : */
1804 : static int
1805 23362237 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
1806 : {
1807 23362237 : VarString *arg1 = DatumGetVarStringPP(x);
1808 23362237 : VarString *arg2 = DatumGetVarStringPP(y);
1809 : char *a1p,
1810 : *a2p;
1811 : int len1,
1812 : len2,
1813 : result;
1814 :
1815 23362237 : a1p = VARDATA_ANY(arg1);
1816 23362237 : a2p = VARDATA_ANY(arg2);
1817 :
1818 23362237 : len1 = VARSIZE_ANY_EXHDR(arg1);
1819 23362237 : len2 = VARSIZE_ANY_EXHDR(arg2);
1820 :
1821 23362237 : result = memcmp(a1p, a2p, Min(len1, len2));
1822 23362237 : if ((result == 0) && (len1 != len2))
1823 656179 : result = (len1 < len2) ? -1 : 1;
1824 :
1825 : /* We can't afford to leak memory here. */
1826 23362237 : if (PointerGetDatum(arg1) != x)
1827 0 : pfree(arg1);
1828 23362237 : if (PointerGetDatum(arg2) != y)
1829 0 : pfree(arg2);
1830 :
1831 23362237 : return result;
1832 : }
1833 :
1834 : /*
1835 : * sortsupport comparison func (for BpChar C locale case)
1836 : *
1837 : * BpChar outsources its sortsupport to this module. Specialization for the
1838 : * varstr_sortsupport BpChar case, modeled on
1839 : * internal_bpchar_pattern_compare().
1840 : */
1841 : static int
1842 31161 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
1843 : {
1844 31161 : BpChar *arg1 = DatumGetBpCharPP(x);
1845 31161 : BpChar *arg2 = DatumGetBpCharPP(y);
1846 : char *a1p,
1847 : *a2p;
1848 : int len1,
1849 : len2,
1850 : result;
1851 :
1852 31161 : a1p = VARDATA_ANY(arg1);
1853 31161 : a2p = VARDATA_ANY(arg2);
1854 :
1855 31161 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
1856 31161 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
1857 :
1858 31161 : result = memcmp(a1p, a2p, Min(len1, len2));
1859 31161 : if ((result == 0) && (len1 != len2))
1860 2 : result = (len1 < len2) ? -1 : 1;
1861 :
1862 : /* We can't afford to leak memory here. */
1863 31161 : if (PointerGetDatum(arg1) != x)
1864 0 : pfree(arg1);
1865 31161 : if (PointerGetDatum(arg2) != y)
1866 0 : pfree(arg2);
1867 :
1868 31161 : return result;
1869 : }
1870 :
1871 : /*
1872 : * sortsupport comparison func (for NAME C locale case)
1873 : */
1874 : static int
1875 21648794 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
1876 : {
1877 21648794 : Name arg1 = DatumGetName(x);
1878 21648794 : Name arg2 = DatumGetName(y);
1879 :
1880 21648794 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
1881 : }
1882 :
1883 : /*
1884 : * sortsupport comparison func (for locale case with all varlena types)
1885 : */
1886 : static int
1887 18714290 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
1888 : {
1889 18714290 : VarString *arg1 = DatumGetVarStringPP(x);
1890 18714290 : VarString *arg2 = DatumGetVarStringPP(y);
1891 : char *a1p,
1892 : *a2p;
1893 : int len1,
1894 : len2,
1895 : result;
1896 :
1897 18714290 : a1p = VARDATA_ANY(arg1);
1898 18714290 : a2p = VARDATA_ANY(arg2);
1899 :
1900 18714290 : len1 = VARSIZE_ANY_EXHDR(arg1);
1901 18714290 : len2 = VARSIZE_ANY_EXHDR(arg2);
1902 :
1903 18714290 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
1904 :
1905 : /* We can't afford to leak memory here. */
1906 18714290 : if (PointerGetDatum(arg1) != x)
1907 0 : pfree(arg1);
1908 18714290 : if (PointerGetDatum(arg2) != y)
1909 0 : pfree(arg2);
1910 :
1911 18714290 : return result;
1912 : }
1913 :
1914 : /*
1915 : * sortsupport comparison func (for locale case with NAME type)
1916 : */
1917 : static int
1918 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
1919 : {
1920 0 : Name arg1 = DatumGetName(x);
1921 0 : Name arg2 = DatumGetName(y);
1922 :
1923 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
1924 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
1925 : ssup);
1926 : }
1927 :
1928 : /*
1929 : * sortsupport comparison func for locale cases
1930 : */
1931 : static int
1932 18714290 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
1933 : {
1934 18714290 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1935 : int result;
1936 : bool arg1_match;
1937 :
1938 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
1939 18714290 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
1940 : {
1941 : /*
1942 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
1943 : * last_len2. Existing contents of buffers might still be used by
1944 : * next call.
1945 : *
1946 : * It's fine to allow the comparison of BpChar padding bytes here,
1947 : * even though that implies that the memcmp() will usually be
1948 : * performed for BpChar callers (though multibyte characters could
1949 : * still prevent that from occurring). The memcmp() is still very
1950 : * cheap, and BpChar's funny semantics have us remove trailing spaces
1951 : * (not limited to padding), so we need make no distinction between
1952 : * padding space characters and "real" space characters.
1953 : */
1954 4687160 : return 0;
1955 : }
1956 :
1957 14027130 : if (sss->typid == BPCHAROID)
1958 : {
1959 : /* Get true number of bytes, ignoring trailing spaces */
1960 18543 : len1 = bpchartruelen(a1p, len1);
1961 18543 : len2 = bpchartruelen(a2p, len2);
1962 : }
1963 :
1964 14027130 : if (len1 >= sss->buflen1)
1965 : {
1966 5 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
1967 5 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
1968 : }
1969 14027130 : if (len2 >= sss->buflen2)
1970 : {
1971 3 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
1972 3 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
1973 : }
1974 :
1975 : /*
1976 : * We're likely to be asked to compare the same strings repeatedly, and
1977 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
1978 : * comparisons, even though in general there is no reason to think that
1979 : * that will work out (every string datum may be unique). Caching does
1980 : * not slow things down measurably when it doesn't work out, and can speed
1981 : * things up by rather a lot when it does. In part, this is because the
1982 : * memcmp() compares data from cachelines that are needed in L1 cache even
1983 : * when the last comparison's result cannot be reused.
1984 : */
1985 14027130 : arg1_match = true;
1986 14027130 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
1987 : {
1988 12962797 : arg1_match = false;
1989 12962797 : memcpy(sss->buf1, a1p, len1);
1990 12962797 : sss->buf1[len1] = '\0';
1991 12962797 : sss->last_len1 = len1;
1992 : }
1993 :
1994 : /*
1995 : * If we're comparing the same two strings as last time, we can return the
1996 : * same answer without calling strcoll() again. This is more likely than
1997 : * it seems (at least with moderate to low cardinality sets), because
1998 : * quicksort compares the same pivot against many values.
1999 : */
2000 14027130 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2001 : {
2002 2150654 : memcpy(sss->buf2, a2p, len2);
2003 2150654 : sss->buf2[len2] = '\0';
2004 2150654 : sss->last_len2 = len2;
2005 : }
2006 11876476 : else if (arg1_match && !sss->cache_blob)
2007 : {
2008 : /* Use result cached following last actual strcoll() call */
2009 839644 : return sss->last_returned;
2010 : }
2011 :
2012 13187486 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
2013 :
2014 : /* Break tie if necessary. */
2015 13187486 : if (result == 0 && sss->locale->deterministic)
2016 0 : result = strcmp(sss->buf1, sss->buf2);
2017 :
2018 : /* Cache result, perhaps saving an expensive strcoll() call next time */
2019 13187486 : sss->cache_blob = false;
2020 13187486 : sss->last_returned = result;
2021 13187486 : return result;
2022 : }
2023 :
2024 : /*
2025 : * Conversion routine for sortsupport. Converts original to abbreviated key
2026 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
2027 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2028 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
2029 : * locale is used just memcpy() from original instead.
2030 : */
2031 : static Datum
2032 422327 : varstr_abbrev_convert(Datum original, SortSupport ssup)
2033 : {
2034 422327 : const size_t max_prefix_bytes = sizeof(Datum);
2035 422327 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2036 422327 : VarString *authoritative = DatumGetVarStringPP(original);
2037 422327 : char *authoritative_data = VARDATA_ANY(authoritative);
2038 :
2039 : /* working state */
2040 : Datum res;
2041 : char *pres;
2042 : int len;
2043 : uint32 hash;
2044 :
2045 422327 : pres = (char *) &res;
2046 : /* memset(), so any non-overwritten bytes are NUL */
2047 422327 : memset(pres, 0, max_prefix_bytes);
2048 422327 : len = VARSIZE_ANY_EXHDR(authoritative);
2049 :
2050 : /* Get number of bytes, ignoring trailing spaces */
2051 422327 : if (sss->typid == BPCHAROID)
2052 505 : len = bpchartruelen(authoritative_data, len);
2053 :
2054 : /*
2055 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2056 : * abbreviate keys. The full comparator for the C locale is also
2057 : * memcmp(). This should be faster than strxfrm().
2058 : */
2059 422327 : if (sss->collate_c)
2060 421409 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
2061 : else
2062 : {
2063 : Size bsize;
2064 :
2065 : /*
2066 : * We're not using the C collation, so fall back on strxfrm or ICU
2067 : * analogs.
2068 : */
2069 :
2070 : /* By convention, we use buffer 1 to store and NUL-terminate */
2071 918 : if (len >= sss->buflen1)
2072 : {
2073 0 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2074 0 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2075 : }
2076 :
2077 : /* Might be able to reuse strxfrm() blob from last call */
2078 918 : if (sss->last_len1 == len && sss->cache_blob &&
2079 459 : memcmp(sss->buf1, authoritative_data, len) == 0)
2080 : {
2081 84 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, sss->last_len2));
2082 : /* No change affecting cardinality, so no hashing required */
2083 84 : goto done;
2084 : }
2085 :
2086 834 : memcpy(sss->buf1, authoritative_data, len);
2087 :
2088 : /*
2089 : * pg_strxfrm() and pg_strxfrm_prefix expect NUL-terminated strings.
2090 : */
2091 834 : sss->buf1[len] = '\0';
2092 834 : sss->last_len1 = len;
2093 :
2094 834 : if (pg_strxfrm_prefix_enabled(sss->locale))
2095 : {
2096 834 : if (sss->buflen2 < max_prefix_bytes)
2097 : {
2098 0 : sss->buflen2 = Max(max_prefix_bytes,
2099 : Min(sss->buflen2 * 2, MaxAllocSize));
2100 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2101 : }
2102 :
2103 834 : bsize = pg_strxfrm_prefix(sss->buf2, sss->buf1,
2104 : max_prefix_bytes, sss->locale);
2105 834 : sss->last_len2 = bsize;
2106 : }
2107 : else
2108 : {
2109 : /*
2110 : * Loop: Call pg_strxfrm(), possibly enlarge buffer, and try
2111 : * again. The pg_strxfrm() function leaves the result buffer
2112 : * content undefined if the result did not fit, so we need to
2113 : * retry until everything fits, even though we only need the first
2114 : * few bytes in the end.
2115 : */
2116 : for (;;)
2117 : {
2118 0 : bsize = pg_strxfrm(sss->buf2, sss->buf1, sss->buflen2,
2119 : sss->locale);
2120 :
2121 0 : sss->last_len2 = bsize;
2122 0 : if (bsize < sss->buflen2)
2123 0 : break;
2124 :
2125 : /*
2126 : * Grow buffer and retry.
2127 : */
2128 0 : sss->buflen2 = Max(bsize + 1,
2129 : Min(sss->buflen2 * 2, MaxAllocSize));
2130 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2131 : }
2132 : }
2133 :
2134 : /*
2135 : * Every Datum byte is always compared. This is safe because the
2136 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2137 : * misinterpreting any NUL bytes not intended to be interpreted as
2138 : * logically representing termination.
2139 : */
2140 834 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, bsize));
2141 : }
2142 :
2143 : /*
2144 : * Maintain approximate cardinality of both abbreviated keys and original,
2145 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2146 : * the worst case, where we do many string transformations for no saving
2147 : * in full strcoll()-based comparisons. These statistics are used by
2148 : * varstr_abbrev_abort().
2149 : *
2150 : * First, Hash key proper, or a significant fraction of it. Mix in length
2151 : * in order to compensate for cases where differences are past
2152 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2153 : */
2154 422243 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2155 : Min(len, PG_CACHE_LINE_SIZE)));
2156 :
2157 422243 : if (len > PG_CACHE_LINE_SIZE)
2158 96 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2159 :
2160 422243 : addHyperLogLog(&sss->full_card, hash);
2161 :
2162 : /* Hash abbreviated key */
2163 : {
2164 : uint32 tmp;
2165 :
2166 422243 : tmp = DatumGetUInt32(res) ^ (uint32) (DatumGetUInt64(res) >> 32);
2167 422243 : hash = DatumGetUInt32(hash_uint32(tmp));
2168 : }
2169 :
2170 422243 : addHyperLogLog(&sss->abbr_card, hash);
2171 :
2172 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2173 422243 : sss->cache_blob = true;
2174 422327 : done:
2175 :
2176 : /*
2177 : * Byteswap on little-endian machines.
2178 : *
2179 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2180 : * 3-way comparator) works correctly on all platforms. If we didn't do
2181 : * this, the comparator would have to call memcmp() with a pair of
2182 : * pointers to the first byte of each abbreviated key, which is slower.
2183 : */
2184 422327 : res = DatumBigEndianToNative(res);
2185 :
2186 : /* Don't leak memory here */
2187 422327 : if (PointerGetDatum(authoritative) != original)
2188 1 : pfree(authoritative);
2189 :
2190 422327 : return res;
2191 : }
2192 :
2193 : /*
2194 : * Callback for estimating effectiveness of abbreviated key optimization, using
2195 : * heuristic rules. Returns value indicating if the abbreviation optimization
2196 : * should be aborted, based on its projected effectiveness.
2197 : */
2198 : static bool
2199 1193 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2200 : {
2201 1193 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2202 : double abbrev_distinct,
2203 : key_distinct;
2204 :
2205 : Assert(ssup->abbreviate);
2206 :
2207 : /* Have a little patience */
2208 1193 : if (memtupcount < 100)
2209 694 : return false;
2210 :
2211 499 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2212 499 : key_distinct = estimateHyperLogLog(&sss->full_card);
2213 :
2214 : /*
2215 : * Clamp cardinality estimates to at least one distinct value. While
2216 : * NULLs are generally disregarded, if only NULL values were seen so far,
2217 : * that might misrepresent costs if we failed to clamp.
2218 : */
2219 499 : if (abbrev_distinct < 1.0)
2220 0 : abbrev_distinct = 1.0;
2221 :
2222 499 : if (key_distinct < 1.0)
2223 0 : key_distinct = 1.0;
2224 :
2225 : /*
2226 : * In the worst case all abbreviated keys are identical, while at the same
2227 : * time there are differences within full key strings not captured in
2228 : * abbreviations.
2229 : */
2230 499 : if (trace_sort)
2231 : {
2232 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2233 :
2234 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2235 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2236 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2237 : sss->prop_card);
2238 : }
2239 :
2240 : /*
2241 : * If the number of distinct abbreviated keys approximately matches the
2242 : * number of distinct authoritative original keys, that's reason enough to
2243 : * proceed. We can win even with a very low cardinality set if most
2244 : * tie-breakers only memcmp(). This is by far the most important
2245 : * consideration.
2246 : *
2247 : * While comparisons that are resolved at the abbreviated key level are
2248 : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2249 : * those two outcomes are so much cheaper than a full strcoll() once
2250 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2251 : * cardinality against the overall size of the set in order to more
2252 : * accurately model costs. Assume that an abbreviated comparison, and an
2253 : * abbreviated comparison with a cheap memcmp()-based authoritative
2254 : * resolution are equivalent.
2255 : */
2256 499 : if (abbrev_distinct > key_distinct * sss->prop_card)
2257 : {
2258 : /*
2259 : * When we have exceeded 10,000 tuples, decay required cardinality
2260 : * aggressively for next call.
2261 : *
2262 : * This is useful because the number of comparisons required on
2263 : * average increases at a linearithmic rate, and at roughly 10,000
2264 : * tuples that factor will start to dominate over the linear costs of
2265 : * string transformation (this is a conservative estimate). The decay
2266 : * rate is chosen to be a little less aggressive than halving -- which
2267 : * (since we're called at points at which memtupcount has doubled)
2268 : * would never see the cost model actually abort past the first call
2269 : * following a decay. This decay rate is mostly a precaution against
2270 : * a sudden, violent swing in how well abbreviated cardinality tracks
2271 : * full key cardinality. The decay also serves to prevent a marginal
2272 : * case from being aborted too late, when too much has already been
2273 : * invested in string transformation.
2274 : *
2275 : * It's possible for sets of several million distinct strings with
2276 : * mere tens of thousands of distinct abbreviated keys to still
2277 : * benefit very significantly. This will generally occur provided
2278 : * each abbreviated key is a proxy for a roughly uniform number of the
2279 : * set's full keys. If it isn't so, we hope to catch that early and
2280 : * abort. If it isn't caught early, by the time the problem is
2281 : * apparent it's probably not worth aborting.
2282 : */
2283 499 : if (memtupcount > 10000)
2284 2 : sss->prop_card *= 0.65;
2285 :
2286 499 : return false;
2287 : }
2288 :
2289 : /*
2290 : * Abort abbreviation strategy.
2291 : *
2292 : * The worst case, where all abbreviated keys are identical while all
2293 : * original strings differ will typically only see a regression of about
2294 : * 10% in execution time for small to medium sized lists of strings.
2295 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2296 : * often expect very large improvements, particularly with sets of strings
2297 : * of moderately high to high abbreviated cardinality. There is little to
2298 : * lose but much to gain, which our strategy reflects.
2299 : */
2300 0 : if (trace_sort)
2301 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2302 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2303 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2304 :
2305 0 : return true;
2306 : }
2307 :
2308 : /*
2309 : * Generic equalimage support function for character type's operator classes.
2310 : * Disables the use of deduplication with nondeterministic collations.
2311 : */
2312 : Datum
2313 4599 : btvarstrequalimage(PG_FUNCTION_ARGS)
2314 : {
2315 : #ifdef NOT_USED
2316 : Oid opcintype = PG_GETARG_OID(0);
2317 : #endif
2318 4599 : Oid collid = PG_GET_COLLATION();
2319 : pg_locale_t locale;
2320 :
2321 4599 : check_collation_set(collid);
2322 :
2323 4599 : locale = pg_newlocale_from_collation(collid);
2324 :
2325 4599 : PG_RETURN_BOOL(locale->deterministic);
2326 : }
2327 :
2328 : Datum
2329 114780 : text_larger(PG_FUNCTION_ARGS)
2330 : {
2331 114780 : text *arg1 = PG_GETARG_TEXT_PP(0);
2332 114780 : text *arg2 = PG_GETARG_TEXT_PP(1);
2333 : text *result;
2334 :
2335 114780 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2336 :
2337 114780 : PG_RETURN_TEXT_P(result);
2338 : }
2339 :
2340 : Datum
2341 43038 : text_smaller(PG_FUNCTION_ARGS)
2342 : {
2343 43038 : text *arg1 = PG_GETARG_TEXT_PP(0);
2344 43038 : text *arg2 = PG_GETARG_TEXT_PP(1);
2345 : text *result;
2346 :
2347 43038 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2348 :
2349 43038 : PG_RETURN_TEXT_P(result);
2350 : }
2351 :
2352 :
2353 : /*
2354 : * Cross-type comparison functions for types text and name.
2355 : */
2356 :
2357 : Datum
2358 112495 : nameeqtext(PG_FUNCTION_ARGS)
2359 : {
2360 112495 : Name arg1 = PG_GETARG_NAME(0);
2361 112495 : text *arg2 = PG_GETARG_TEXT_PP(1);
2362 112495 : size_t len1 = strlen(NameStr(*arg1));
2363 112495 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2364 112495 : Oid collid = PG_GET_COLLATION();
2365 : bool result;
2366 :
2367 112495 : check_collation_set(collid);
2368 :
2369 112495 : if (collid == C_COLLATION_OID)
2370 127904 : result = (len1 == len2 &&
2371 62077 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2372 : else
2373 46668 : result = (varstr_cmp(NameStr(*arg1), len1,
2374 46668 : VARDATA_ANY(arg2), len2,
2375 : collid) == 0);
2376 :
2377 112495 : PG_FREE_IF_COPY(arg2, 1);
2378 :
2379 112495 : PG_RETURN_BOOL(result);
2380 : }
2381 :
2382 : Datum
2383 3992 : texteqname(PG_FUNCTION_ARGS)
2384 : {
2385 3992 : text *arg1 = PG_GETARG_TEXT_PP(0);
2386 3992 : Name arg2 = PG_GETARG_NAME(1);
2387 3992 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2388 3992 : size_t len2 = strlen(NameStr(*arg2));
2389 3992 : Oid collid = PG_GET_COLLATION();
2390 : bool result;
2391 :
2392 3992 : check_collation_set(collid);
2393 :
2394 3992 : if (collid == C_COLLATION_OID)
2395 284 : result = (len1 == len2 &&
2396 91 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2397 : else
2398 3799 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2399 3799 : NameStr(*arg2), len2,
2400 : collid) == 0);
2401 :
2402 3992 : PG_FREE_IF_COPY(arg1, 0);
2403 :
2404 3992 : PG_RETURN_BOOL(result);
2405 : }
2406 :
2407 : Datum
2408 9 : namenetext(PG_FUNCTION_ARGS)
2409 : {
2410 9 : Name arg1 = PG_GETARG_NAME(0);
2411 9 : text *arg2 = PG_GETARG_TEXT_PP(1);
2412 9 : size_t len1 = strlen(NameStr(*arg1));
2413 9 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2414 9 : Oid collid = PG_GET_COLLATION();
2415 : bool result;
2416 :
2417 9 : check_collation_set(collid);
2418 :
2419 9 : if (collid == C_COLLATION_OID)
2420 0 : result = !(len1 == len2 &&
2421 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2422 : else
2423 9 : result = !(varstr_cmp(NameStr(*arg1), len1,
2424 9 : VARDATA_ANY(arg2), len2,
2425 : collid) == 0);
2426 :
2427 9 : PG_FREE_IF_COPY(arg2, 1);
2428 :
2429 9 : PG_RETURN_BOOL(result);
2430 : }
2431 :
2432 : Datum
2433 9 : textnename(PG_FUNCTION_ARGS)
2434 : {
2435 9 : text *arg1 = PG_GETARG_TEXT_PP(0);
2436 9 : Name arg2 = PG_GETARG_NAME(1);
2437 9 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2438 9 : size_t len2 = strlen(NameStr(*arg2));
2439 9 : Oid collid = PG_GET_COLLATION();
2440 : bool result;
2441 :
2442 9 : check_collation_set(collid);
2443 :
2444 9 : if (collid == C_COLLATION_OID)
2445 0 : result = !(len1 == len2 &&
2446 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2447 : else
2448 9 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2449 9 : NameStr(*arg2), len2,
2450 : collid) == 0);
2451 :
2452 9 : PG_FREE_IF_COPY(arg1, 0);
2453 :
2454 9 : PG_RETURN_BOOL(result);
2455 : }
2456 :
2457 : Datum
2458 69966 : btnametextcmp(PG_FUNCTION_ARGS)
2459 : {
2460 69966 : Name arg1 = PG_GETARG_NAME(0);
2461 69966 : text *arg2 = PG_GETARG_TEXT_PP(1);
2462 : int32 result;
2463 :
2464 69966 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2465 69966 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2466 : PG_GET_COLLATION());
2467 :
2468 69966 : PG_FREE_IF_COPY(arg2, 1);
2469 :
2470 69966 : PG_RETURN_INT32(result);
2471 : }
2472 :
2473 : Datum
2474 22 : bttextnamecmp(PG_FUNCTION_ARGS)
2475 : {
2476 22 : text *arg1 = PG_GETARG_TEXT_PP(0);
2477 22 : Name arg2 = PG_GETARG_NAME(1);
2478 : int32 result;
2479 :
2480 22 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2481 22 : NameStr(*arg2), strlen(NameStr(*arg2)),
2482 : PG_GET_COLLATION());
2483 :
2484 22 : PG_FREE_IF_COPY(arg1, 0);
2485 :
2486 22 : PG_RETURN_INT32(result);
2487 : }
2488 :
2489 : #define CmpCall(cmpfunc) \
2490 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2491 : PG_GET_COLLATION(), \
2492 : PG_GETARG_DATUM(0), \
2493 : PG_GETARG_DATUM(1)))
2494 :
2495 : Datum
2496 33968 : namelttext(PG_FUNCTION_ARGS)
2497 : {
2498 33968 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2499 : }
2500 :
2501 : Datum
2502 0 : nameletext(PG_FUNCTION_ARGS)
2503 : {
2504 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2505 : }
2506 :
2507 : Datum
2508 0 : namegttext(PG_FUNCTION_ARGS)
2509 : {
2510 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2511 : }
2512 :
2513 : Datum
2514 29632 : namegetext(PG_FUNCTION_ARGS)
2515 : {
2516 29632 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2517 : }
2518 :
2519 : Datum
2520 0 : textltname(PG_FUNCTION_ARGS)
2521 : {
2522 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
2523 : }
2524 :
2525 : Datum
2526 0 : textlename(PG_FUNCTION_ARGS)
2527 : {
2528 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2529 : }
2530 :
2531 : Datum
2532 0 : textgtname(PG_FUNCTION_ARGS)
2533 : {
2534 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2535 : }
2536 :
2537 : Datum
2538 0 : textgename(PG_FUNCTION_ARGS)
2539 : {
2540 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2541 : }
2542 :
2543 : #undef CmpCall
2544 :
2545 :
2546 : /*
2547 : * The following operators support character-by-character comparison
2548 : * of text datums, to allow building indexes suitable for LIKE clauses.
2549 : * Note that the regular texteq/textne comparison operators, and regular
2550 : * support functions 1 and 2 with "C" collation are assumed to be
2551 : * compatible with these!
2552 : */
2553 :
2554 : static int
2555 80222 : internal_text_pattern_compare(text *arg1, text *arg2)
2556 : {
2557 : int result;
2558 : int len1,
2559 : len2;
2560 :
2561 80222 : len1 = VARSIZE_ANY_EXHDR(arg1);
2562 80222 : len2 = VARSIZE_ANY_EXHDR(arg2);
2563 :
2564 80222 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2565 80222 : if (result != 0)
2566 80156 : return result;
2567 66 : else if (len1 < len2)
2568 0 : return -1;
2569 66 : else if (len1 > len2)
2570 42 : return 1;
2571 : else
2572 24 : return 0;
2573 : }
2574 :
2575 :
2576 : Datum
2577 23933 : text_pattern_lt(PG_FUNCTION_ARGS)
2578 : {
2579 23933 : text *arg1 = PG_GETARG_TEXT_PP(0);
2580 23933 : text *arg2 = PG_GETARG_TEXT_PP(1);
2581 : int result;
2582 :
2583 23933 : result = internal_text_pattern_compare(arg1, arg2);
2584 :
2585 23933 : PG_FREE_IF_COPY(arg1, 0);
2586 23933 : PG_FREE_IF_COPY(arg2, 1);
2587 :
2588 23933 : PG_RETURN_BOOL(result < 0);
2589 : }
2590 :
2591 :
2592 : Datum
2593 18755 : text_pattern_le(PG_FUNCTION_ARGS)
2594 : {
2595 18755 : text *arg1 = PG_GETARG_TEXT_PP(0);
2596 18755 : text *arg2 = PG_GETARG_TEXT_PP(1);
2597 : int result;
2598 :
2599 18755 : result = internal_text_pattern_compare(arg1, arg2);
2600 :
2601 18755 : PG_FREE_IF_COPY(arg1, 0);
2602 18755 : PG_FREE_IF_COPY(arg2, 1);
2603 :
2604 18755 : PG_RETURN_BOOL(result <= 0);
2605 : }
2606 :
2607 :
2608 : Datum
2609 18767 : text_pattern_ge(PG_FUNCTION_ARGS)
2610 : {
2611 18767 : text *arg1 = PG_GETARG_TEXT_PP(0);
2612 18767 : text *arg2 = PG_GETARG_TEXT_PP(1);
2613 : int result;
2614 :
2615 18767 : result = internal_text_pattern_compare(arg1, arg2);
2616 :
2617 18767 : PG_FREE_IF_COPY(arg1, 0);
2618 18767 : PG_FREE_IF_COPY(arg2, 1);
2619 :
2620 18767 : PG_RETURN_BOOL(result >= 0);
2621 : }
2622 :
2623 :
2624 : Datum
2625 18755 : text_pattern_gt(PG_FUNCTION_ARGS)
2626 : {
2627 18755 : text *arg1 = PG_GETARG_TEXT_PP(0);
2628 18755 : text *arg2 = PG_GETARG_TEXT_PP(1);
2629 : int result;
2630 :
2631 18755 : result = internal_text_pattern_compare(arg1, arg2);
2632 :
2633 18755 : PG_FREE_IF_COPY(arg1, 0);
2634 18755 : PG_FREE_IF_COPY(arg2, 1);
2635 :
2636 18755 : PG_RETURN_BOOL(result > 0);
2637 : }
2638 :
2639 :
2640 : Datum
2641 12 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
2642 : {
2643 12 : text *arg1 = PG_GETARG_TEXT_PP(0);
2644 12 : text *arg2 = PG_GETARG_TEXT_PP(1);
2645 : int result;
2646 :
2647 12 : result = internal_text_pattern_compare(arg1, arg2);
2648 :
2649 12 : PG_FREE_IF_COPY(arg1, 0);
2650 12 : PG_FREE_IF_COPY(arg2, 1);
2651 :
2652 12 : PG_RETURN_INT32(result);
2653 : }
2654 :
2655 :
2656 : Datum
2657 58 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
2658 : {
2659 58 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
2660 : MemoryContext oldcontext;
2661 :
2662 58 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2663 :
2664 : /* Use generic string SortSupport, forcing "C" collation */
2665 58 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
2666 :
2667 58 : MemoryContextSwitchTo(oldcontext);
2668 :
2669 58 : PG_RETURN_VOID();
2670 : }
2671 :
2672 :
2673 : /* text_name()
2674 : * Converts a text type to a Name type.
2675 : */
2676 : Datum
2677 15441 : text_name(PG_FUNCTION_ARGS)
2678 : {
2679 15441 : text *s = PG_GETARG_TEXT_PP(0);
2680 : Name result;
2681 : int len;
2682 :
2683 15441 : len = VARSIZE_ANY_EXHDR(s);
2684 :
2685 : /* Truncate oversize input */
2686 15441 : if (len >= NAMEDATALEN)
2687 3 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
2688 :
2689 : /* We use palloc0 here to ensure result is zero-padded */
2690 15441 : result = (Name) palloc0(NAMEDATALEN);
2691 15441 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
2692 :
2693 15441 : PG_RETURN_NAME(result);
2694 : }
2695 :
2696 : /* name_text()
2697 : * Converts a Name type to a text type.
2698 : */
2699 : Datum
2700 328162 : name_text(PG_FUNCTION_ARGS)
2701 : {
2702 328162 : Name s = PG_GETARG_NAME(0);
2703 :
2704 328162 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
2705 : }
2706 :
2707 :
2708 : /*
2709 : * textToQualifiedNameList - convert a text object to list of names
2710 : *
2711 : * This implements the input parsing needed by nextval() and other
2712 : * functions that take a text parameter representing a qualified name.
2713 : * We split the name at dots, downcase if not double-quoted, and
2714 : * truncate names if they're too long.
2715 : */
2716 : List *
2717 2716 : textToQualifiedNameList(text *textval)
2718 : {
2719 : char *rawname;
2720 2716 : List *result = NIL;
2721 : List *namelist;
2722 : ListCell *l;
2723 :
2724 : /* Convert to C string (handles possible detoasting). */
2725 : /* Note we rely on being able to modify rawname below. */
2726 2716 : rawname = text_to_cstring(textval);
2727 :
2728 2716 : if (!SplitIdentifierString(rawname, '.', &namelist))
2729 0 : ereport(ERROR,
2730 : (errcode(ERRCODE_INVALID_NAME),
2731 : errmsg("invalid name syntax")));
2732 :
2733 2716 : if (namelist == NIL)
2734 0 : ereport(ERROR,
2735 : (errcode(ERRCODE_INVALID_NAME),
2736 : errmsg("invalid name syntax")));
2737 :
2738 5490 : foreach(l, namelist)
2739 : {
2740 2774 : char *curname = (char *) lfirst(l);
2741 :
2742 2774 : result = lappend(result, makeString(pstrdup(curname)));
2743 : }
2744 :
2745 2716 : pfree(rawname);
2746 2716 : list_free(namelist);
2747 :
2748 2716 : return result;
2749 : }
2750 :
2751 : /*
2752 : * SplitIdentifierString --- parse a string containing identifiers
2753 : *
2754 : * This is the guts of textToQualifiedNameList, and is exported for use in
2755 : * other situations such as parsing GUC variables. In the GUC case, it's
2756 : * important to avoid memory leaks, so the API is designed to minimize the
2757 : * amount of stuff that needs to be allocated and freed.
2758 : *
2759 : * Inputs:
2760 : * rawstring: the input string; must be overwritable! On return, it's
2761 : * been modified to contain the separated identifiers.
2762 : * separator: the separator punctuation expected between identifiers
2763 : * (typically '.' or ','). Whitespace may also appear around
2764 : * identifiers.
2765 : * Outputs:
2766 : * namelist: filled with a palloc'd list of pointers to identifiers within
2767 : * rawstring. Caller should list_free() this even on error return.
2768 : *
2769 : * Returns true if okay, false if there is a syntax error in the string.
2770 : *
2771 : * Note that an empty string is considered okay here, though not in
2772 : * textToQualifiedNameList.
2773 : */
2774 : bool
2775 191418 : SplitIdentifierString(char *rawstring, char separator,
2776 : List **namelist)
2777 : {
2778 191418 : char *nextp = rawstring;
2779 191418 : bool done = false;
2780 :
2781 191418 : *namelist = NIL;
2782 :
2783 191421 : while (scanner_isspace(*nextp))
2784 3 : nextp++; /* skip leading whitespace */
2785 :
2786 191418 : if (*nextp == '\0')
2787 15870 : return true; /* empty string represents empty list */
2788 :
2789 : /* At the top of the loop, we are at start of a new identifier. */
2790 : do
2791 : {
2792 : char *curname;
2793 : char *endp;
2794 :
2795 328466 : if (*nextp == '"')
2796 : {
2797 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
2798 21332 : curname = nextp + 1;
2799 : for (;;)
2800 : {
2801 21334 : endp = strchr(nextp + 1, '"');
2802 21333 : if (endp == NULL)
2803 0 : return false; /* mismatched quotes */
2804 21333 : if (endp[1] != '"')
2805 21332 : break; /* found end of quoted name */
2806 : /* Collapse adjacent quotes into one quote, and look again */
2807 1 : memmove(endp, endp + 1, strlen(endp));
2808 1 : nextp = endp;
2809 : }
2810 : /* endp now points at the terminating quote */
2811 21332 : nextp = endp + 1;
2812 : }
2813 : else
2814 : {
2815 : /* Unquoted name --- extends to separator or whitespace */
2816 : char *downname;
2817 : int len;
2818 :
2819 307134 : curname = nextp;
2820 2813518 : while (*nextp && *nextp != separator &&
2821 2506385 : !scanner_isspace(*nextp))
2822 2506384 : nextp++;
2823 307134 : endp = nextp;
2824 307134 : if (curname == nextp)
2825 0 : return false; /* empty unquoted name not allowed */
2826 :
2827 : /*
2828 : * Downcase the identifier, using same code as main lexer does.
2829 : *
2830 : * XXX because we want to overwrite the input in-place, we cannot
2831 : * support a downcasing transformation that increases the string
2832 : * length. This is not a problem given the current implementation
2833 : * of downcase_truncate_identifier, but we'll probably have to do
2834 : * something about this someday.
2835 : */
2836 307134 : len = endp - curname;
2837 307134 : downname = downcase_truncate_identifier(curname, len, false);
2838 : Assert(strlen(downname) <= len);
2839 307134 : strncpy(curname, downname, len); /* strncpy is required here */
2840 307134 : pfree(downname);
2841 : }
2842 :
2843 328467 : while (scanner_isspace(*nextp))
2844 1 : nextp++; /* skip trailing whitespace */
2845 :
2846 328466 : if (*nextp == separator)
2847 : {
2848 152918 : nextp++;
2849 293122 : while (scanner_isspace(*nextp))
2850 140204 : nextp++; /* skip leading whitespace for next */
2851 : /* we expect another name, so done remains false */
2852 : }
2853 175548 : else if (*nextp == '\0')
2854 175547 : done = true;
2855 : else
2856 1 : return false; /* invalid syntax */
2857 :
2858 : /* Now safe to overwrite separator with a null */
2859 328465 : *endp = '\0';
2860 :
2861 : /* Truncate name if it's overlength */
2862 328465 : truncate_identifier(curname, strlen(curname), false);
2863 :
2864 : /*
2865 : * Finished isolating current name --- add it to list
2866 : */
2867 328465 : *namelist = lappend(*namelist, curname);
2868 :
2869 : /* Loop back if we didn't reach end of string */
2870 328465 : } while (!done);
2871 :
2872 175547 : return true;
2873 : }
2874 :
2875 :
2876 : /*
2877 : * SplitDirectoriesString --- parse a string containing file/directory names
2878 : *
2879 : * This works fine on file names too; the function name is historical.
2880 : *
2881 : * This is similar to SplitIdentifierString, except that the parsing
2882 : * rules are meant to handle pathnames instead of identifiers: there is
2883 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
2884 : * and we apply canonicalize_path() to each extracted string. Because of the
2885 : * last, the returned strings are separately palloc'd rather than being
2886 : * pointers into rawstring --- but we still scribble on rawstring.
2887 : *
2888 : * Inputs:
2889 : * rawstring: the input string; must be modifiable!
2890 : * separator: the separator punctuation expected between directories
2891 : * (typically ',' or ';'). Whitespace may also appear around
2892 : * directories.
2893 : * Outputs:
2894 : * namelist: filled with a palloc'd list of directory names.
2895 : * Caller should list_free_deep() this even on error return.
2896 : *
2897 : * Returns true if okay, false if there is a syntax error in the string.
2898 : *
2899 : * Note that an empty string is considered okay here.
2900 : */
2901 : bool
2902 954 : SplitDirectoriesString(char *rawstring, char separator,
2903 : List **namelist)
2904 : {
2905 954 : char *nextp = rawstring;
2906 954 : bool done = false;
2907 :
2908 954 : *namelist = NIL;
2909 :
2910 954 : while (scanner_isspace(*nextp))
2911 0 : nextp++; /* skip leading whitespace */
2912 :
2913 954 : if (*nextp == '\0')
2914 1 : return true; /* empty string represents empty list */
2915 :
2916 : /* At the top of the loop, we are at start of a new directory. */
2917 : do
2918 : {
2919 : char *curname;
2920 : char *endp;
2921 :
2922 958 : if (*nextp == '"')
2923 : {
2924 : /* Quoted name --- collapse quote-quote pairs */
2925 0 : curname = nextp + 1;
2926 : for (;;)
2927 : {
2928 0 : endp = strchr(nextp + 1, '"');
2929 0 : if (endp == NULL)
2930 0 : return false; /* mismatched quotes */
2931 0 : if (endp[1] != '"')
2932 0 : break; /* found end of quoted name */
2933 : /* Collapse adjacent quotes into one quote, and look again */
2934 0 : memmove(endp, endp + 1, strlen(endp));
2935 0 : nextp = endp;
2936 : }
2937 : /* endp now points at the terminating quote */
2938 0 : nextp = endp + 1;
2939 : }
2940 : else
2941 : {
2942 : /* Unquoted name --- extends to separator or end of string */
2943 958 : curname = endp = nextp;
2944 16038 : while (*nextp && *nextp != separator)
2945 : {
2946 : /* trailing whitespace should not be included in name */
2947 15080 : if (!scanner_isspace(*nextp))
2948 15080 : endp = nextp + 1;
2949 15080 : nextp++;
2950 : }
2951 958 : if (curname == endp)
2952 0 : return false; /* empty unquoted name not allowed */
2953 : }
2954 :
2955 958 : while (scanner_isspace(*nextp))
2956 0 : nextp++; /* skip trailing whitespace */
2957 :
2958 958 : if (*nextp == separator)
2959 : {
2960 5 : nextp++;
2961 8 : while (scanner_isspace(*nextp))
2962 3 : nextp++; /* skip leading whitespace for next */
2963 : /* we expect another name, so done remains false */
2964 : }
2965 953 : else if (*nextp == '\0')
2966 953 : done = true;
2967 : else
2968 0 : return false; /* invalid syntax */
2969 :
2970 : /* Now safe to overwrite separator with a null */
2971 958 : *endp = '\0';
2972 :
2973 : /* Truncate path if it's overlength */
2974 958 : if (strlen(curname) >= MAXPGPATH)
2975 0 : curname[MAXPGPATH - 1] = '\0';
2976 :
2977 : /*
2978 : * Finished isolating current name --- add it to list
2979 : */
2980 958 : curname = pstrdup(curname);
2981 958 : canonicalize_path(curname);
2982 958 : *namelist = lappend(*namelist, curname);
2983 :
2984 : /* Loop back if we didn't reach end of string */
2985 958 : } while (!done);
2986 :
2987 953 : return true;
2988 : }
2989 :
2990 :
2991 : /*
2992 : * SplitGUCList --- parse a string containing identifiers or file names
2993 : *
2994 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
2995 : * presuming whether the elements will be taken as identifiers or file names.
2996 : * We assume the input has already been through flatten_set_variable_args(),
2997 : * so that we need never downcase (if appropriate, that was done already).
2998 : * Nor do we ever truncate, since we don't know the correct max length.
2999 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
3000 : * because any embedded whitespace should have led to double-quoting).
3001 : * Otherwise the API is identical to SplitIdentifierString.
3002 : *
3003 : * XXX it's annoying to have so many copies of this string-splitting logic.
3004 : * However, it's not clear that having one function with a bunch of option
3005 : * flags would be much better.
3006 : *
3007 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
3008 : * Be sure to update that if you have to change this.
3009 : *
3010 : * Inputs:
3011 : * rawstring: the input string; must be overwritable! On return, it's
3012 : * been modified to contain the separated identifiers.
3013 : * separator: the separator punctuation expected between identifiers
3014 : * (typically '.' or ','). Whitespace may also appear around
3015 : * identifiers.
3016 : * Outputs:
3017 : * namelist: filled with a palloc'd list of pointers to identifiers within
3018 : * rawstring. Caller should list_free() this even on error return.
3019 : *
3020 : * Returns true if okay, false if there is a syntax error in the string.
3021 : */
3022 : bool
3023 3700 : SplitGUCList(char *rawstring, char separator,
3024 : List **namelist)
3025 : {
3026 3700 : char *nextp = rawstring;
3027 3700 : bool done = false;
3028 :
3029 3700 : *namelist = NIL;
3030 :
3031 3700 : while (scanner_isspace(*nextp))
3032 0 : nextp++; /* skip leading whitespace */
3033 :
3034 3700 : if (*nextp == '\0')
3035 2062 : return true; /* empty string represents empty list */
3036 :
3037 : /* At the top of the loop, we are at start of a new identifier. */
3038 : do
3039 : {
3040 : char *curname;
3041 : char *endp;
3042 :
3043 1699 : if (*nextp == '"')
3044 : {
3045 : /* Quoted name --- collapse quote-quote pairs */
3046 12 : curname = nextp + 1;
3047 : for (;;)
3048 : {
3049 18 : endp = strchr(nextp + 1, '"');
3050 15 : if (endp == NULL)
3051 0 : return false; /* mismatched quotes */
3052 15 : if (endp[1] != '"')
3053 12 : break; /* found end of quoted name */
3054 : /* Collapse adjacent quotes into one quote, and look again */
3055 3 : memmove(endp, endp + 1, strlen(endp));
3056 3 : nextp = endp;
3057 : }
3058 : /* endp now points at the terminating quote */
3059 12 : nextp = endp + 1;
3060 : }
3061 : else
3062 : {
3063 : /* Unquoted name --- extends to separator or whitespace */
3064 1687 : curname = nextp;
3065 13671 : while (*nextp && *nextp != separator &&
3066 11984 : !scanner_isspace(*nextp))
3067 11984 : nextp++;
3068 1687 : endp = nextp;
3069 1687 : if (curname == nextp)
3070 0 : return false; /* empty unquoted name not allowed */
3071 : }
3072 :
3073 1699 : while (scanner_isspace(*nextp))
3074 0 : nextp++; /* skip trailing whitespace */
3075 :
3076 1699 : if (*nextp == separator)
3077 : {
3078 61 : nextp++;
3079 118 : while (scanner_isspace(*nextp))
3080 57 : nextp++; /* skip leading whitespace for next */
3081 : /* we expect another name, so done remains false */
3082 : }
3083 1638 : else if (*nextp == '\0')
3084 1638 : done = true;
3085 : else
3086 0 : return false; /* invalid syntax */
3087 :
3088 : /* Now safe to overwrite separator with a null */
3089 1699 : *endp = '\0';
3090 :
3091 : /*
3092 : * Finished isolating current name --- add it to list
3093 : */
3094 1699 : *namelist = lappend(*namelist, curname);
3095 :
3096 : /* Loop back if we didn't reach end of string */
3097 1699 : } while (!done);
3098 :
3099 1638 : return true;
3100 : }
3101 :
3102 : /*
3103 : * appendStringInfoText
3104 : *
3105 : * Append a text to str.
3106 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3107 : */
3108 : static void
3109 1087718 : appendStringInfoText(StringInfo str, const text *t)
3110 : {
3111 1087718 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
3112 1087718 : }
3113 :
3114 : /*
3115 : * replace_text
3116 : * replace all occurrences of 'old_sub_str' in 'orig_str'
3117 : * with 'new_sub_str' to form 'new_str'
3118 : *
3119 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3120 : * otherwise returns 'new_str'
3121 : */
3122 : Datum
3123 784 : replace_text(PG_FUNCTION_ARGS)
3124 : {
3125 784 : text *src_text = PG_GETARG_TEXT_PP(0);
3126 784 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
3127 784 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
3128 : int src_text_len;
3129 : int from_sub_text_len;
3130 : TextPositionState state;
3131 : text *ret_text;
3132 : int chunk_len;
3133 : char *curr_ptr;
3134 : char *start_ptr;
3135 : StringInfoData str;
3136 : bool found;
3137 :
3138 784 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
3139 784 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
3140 :
3141 : /* Return unmodified source string if empty source or pattern */
3142 784 : if (src_text_len < 1 || from_sub_text_len < 1)
3143 : {
3144 0 : PG_RETURN_TEXT_P(src_text);
3145 : }
3146 :
3147 784 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
3148 :
3149 784 : found = text_position_next(&state);
3150 :
3151 : /* When the from_sub_text is not found, there is nothing to do. */
3152 784 : if (!found)
3153 : {
3154 167 : text_position_cleanup(&state);
3155 167 : PG_RETURN_TEXT_P(src_text);
3156 : }
3157 617 : curr_ptr = text_position_get_match_ptr(&state);
3158 617 : start_ptr = VARDATA_ANY(src_text);
3159 :
3160 617 : initStringInfo(&str);
3161 :
3162 : do
3163 : {
3164 3608 : CHECK_FOR_INTERRUPTS();
3165 :
3166 : /* copy the data skipped over by last text_position_next() */
3167 3608 : chunk_len = curr_ptr - start_ptr;
3168 3608 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3169 :
3170 3608 : appendStringInfoText(&str, to_sub_text);
3171 :
3172 3608 : start_ptr = curr_ptr + state.last_match_len;
3173 :
3174 3608 : found = text_position_next(&state);
3175 3608 : if (found)
3176 2991 : curr_ptr = text_position_get_match_ptr(&state);
3177 : }
3178 3608 : while (found);
3179 :
3180 : /* copy trailing data */
3181 617 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3182 617 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3183 :
3184 617 : text_position_cleanup(&state);
3185 :
3186 617 : ret_text = cstring_to_text_with_len(str.data, str.len);
3187 617 : pfree(str.data);
3188 :
3189 617 : PG_RETURN_TEXT_P(ret_text);
3190 : }
3191 :
3192 : /*
3193 : * check_replace_text_has_escape
3194 : *
3195 : * Returns 0 if text contains no backslashes that need processing.
3196 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
3197 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
3198 : */
3199 : static int
3200 9393 : check_replace_text_has_escape(const text *replace_text)
3201 : {
3202 9393 : int result = 0;
3203 9393 : const char *p = VARDATA_ANY(replace_text);
3204 9393 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3205 :
3206 18808 : while (p < p_end)
3207 : {
3208 : /* Find next escape char, if any. */
3209 8834 : p = memchr(p, '\\', p_end - p);
3210 8834 : if (p == NULL)
3211 8410 : break;
3212 424 : p++;
3213 : /* Note: a backslash at the end doesn't require extra processing. */
3214 424 : if (p < p_end)
3215 : {
3216 424 : if (*p >= '1' && *p <= '9')
3217 402 : return 2; /* Found a submatch specifier, so done */
3218 22 : result = 1; /* Found some other sequence, keep looking */
3219 22 : p++;
3220 : }
3221 : }
3222 8991 : return result;
3223 : }
3224 :
3225 : /*
3226 : * appendStringInfoRegexpSubstr
3227 : *
3228 : * Append replace_text to str, substituting regexp back references for
3229 : * \n escapes. start_ptr is the start of the match in the source string,
3230 : * at logical character position data_pos.
3231 : */
3232 : static void
3233 127 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
3234 : regmatch_t *pmatch,
3235 : char *start_ptr, int data_pos)
3236 : {
3237 127 : const char *p = VARDATA_ANY(replace_text);
3238 127 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3239 :
3240 305 : while (p < p_end)
3241 : {
3242 268 : const char *chunk_start = p;
3243 : int so;
3244 : int eo;
3245 :
3246 : /* Find next escape char, if any. */
3247 268 : p = memchr(p, '\\', p_end - p);
3248 268 : if (p == NULL)
3249 87 : p = p_end;
3250 :
3251 : /* Copy the text we just scanned over, if any. */
3252 268 : if (p > chunk_start)
3253 159 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3254 :
3255 : /* Done if at end of string, else advance over escape char. */
3256 268 : if (p >= p_end)
3257 87 : break;
3258 181 : p++;
3259 :
3260 181 : if (p >= p_end)
3261 : {
3262 : /* Escape at very end of input. Treat same as unexpected char */
3263 3 : appendStringInfoChar(str, '\\');
3264 3 : break;
3265 : }
3266 :
3267 178 : if (*p >= '1' && *p <= '9')
3268 148 : {
3269 : /* Use the back reference of regexp. */
3270 148 : int idx = *p - '0';
3271 :
3272 148 : so = pmatch[idx].rm_so;
3273 148 : eo = pmatch[idx].rm_eo;
3274 148 : p++;
3275 : }
3276 30 : else if (*p == '&')
3277 : {
3278 : /* Use the entire matched string. */
3279 9 : so = pmatch[0].rm_so;
3280 9 : eo = pmatch[0].rm_eo;
3281 9 : p++;
3282 : }
3283 21 : else if (*p == '\\')
3284 : {
3285 : /* \\ means transfer one \ to output. */
3286 18 : appendStringInfoChar(str, '\\');
3287 18 : p++;
3288 18 : continue;
3289 : }
3290 : else
3291 : {
3292 : /*
3293 : * If escape char is not followed by any expected char, just treat
3294 : * it as ordinary data to copy. (XXX would it be better to throw
3295 : * an error?)
3296 : */
3297 3 : appendStringInfoChar(str, '\\');
3298 3 : continue;
3299 : }
3300 :
3301 157 : if (so >= 0 && eo >= 0)
3302 : {
3303 : /*
3304 : * Copy the text that is back reference of regexp. Note so and eo
3305 : * are counted in characters not bytes.
3306 : */
3307 : char *chunk_start;
3308 : int chunk_len;
3309 :
3310 : Assert(so >= data_pos);
3311 157 : chunk_start = start_ptr;
3312 157 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3313 157 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3314 157 : appendBinaryStringInfo(str, chunk_start, chunk_len);
3315 : }
3316 : }
3317 127 : }
3318 :
3319 : /*
3320 : * replace_text_regexp
3321 : *
3322 : * replace substring(s) in src_text that match pattern with replace_text.
3323 : * The replace_text can contain backslash markers to substitute
3324 : * (parts of) the matched text.
3325 : *
3326 : * cflags: regexp compile flags.
3327 : * collation: collation to use.
3328 : * search_start: the character (not byte) offset in src_text at which to
3329 : * begin searching.
3330 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
3331 : */
3332 : text *
3333 9393 : replace_text_regexp(text *src_text, text *pattern_text,
3334 : text *replace_text,
3335 : int cflags, Oid collation,
3336 : int search_start, int n)
3337 : {
3338 : text *ret_text;
3339 : regex_t *re;
3340 9393 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3341 9393 : int nmatches = 0;
3342 : StringInfoData buf;
3343 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
3344 9393 : int nmatch = lengthof(pmatch);
3345 : pg_wchar *data;
3346 : size_t data_len;
3347 : int data_pos;
3348 : char *start_ptr;
3349 : int escape_status;
3350 :
3351 9393 : initStringInfo(&buf);
3352 :
3353 : /* Convert data string to wide characters. */
3354 9393 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3355 9393 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3356 :
3357 : /* Check whether replace_text has escapes, especially regexp submatches. */
3358 9393 : escape_status = check_replace_text_has_escape(replace_text);
3359 :
3360 : /* If no regexp submatches, we can use REG_NOSUB. */
3361 9393 : if (escape_status < 2)
3362 : {
3363 8991 : cflags |= REG_NOSUB;
3364 : /* Also tell pg_regexec we only want the whole-match location. */
3365 8991 : nmatch = 1;
3366 : }
3367 :
3368 : /* Prepare the regexp. */
3369 9393 : re = RE_compile_and_cache(pattern_text, cflags, collation);
3370 :
3371 : /* start_ptr points to the data_pos'th character of src_text */
3372 9393 : start_ptr = (char *) VARDATA_ANY(src_text);
3373 9393 : data_pos = 0;
3374 :
3375 12603 : while (search_start <= data_len)
3376 : {
3377 : int regexec_result;
3378 :
3379 12600 : CHECK_FOR_INTERRUPTS();
3380 :
3381 12600 : regexec_result = pg_regexec(re,
3382 : data,
3383 : data_len,
3384 : search_start,
3385 : NULL, /* no details */
3386 : nmatch,
3387 : pmatch,
3388 : 0);
3389 :
3390 12600 : if (regexec_result == REG_NOMATCH)
3391 8355 : break;
3392 :
3393 4245 : if (regexec_result != REG_OKAY)
3394 : {
3395 : char errMsg[100];
3396 :
3397 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3398 0 : ereport(ERROR,
3399 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3400 : errmsg("regular expression failed: %s", errMsg)));
3401 : }
3402 :
3403 : /*
3404 : * Count matches, and decide whether to replace this match.
3405 : */
3406 4245 : nmatches++;
3407 4245 : if (n > 0 && nmatches != n)
3408 : {
3409 : /*
3410 : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
3411 : * we treat the matched text as if it weren't matched, and copy it
3412 : * to the output later.)
3413 : */
3414 30 : search_start = pmatch[0].rm_eo;
3415 30 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3416 0 : search_start++;
3417 30 : continue;
3418 : }
3419 :
3420 : /*
3421 : * Copy the text to the left of the match position. Note we are given
3422 : * character not byte indexes.
3423 : */
3424 4215 : if (pmatch[0].rm_so - data_pos > 0)
3425 : {
3426 : int chunk_len;
3427 :
3428 4127 : chunk_len = charlen_to_bytelen(start_ptr,
3429 4127 : pmatch[0].rm_so - data_pos);
3430 4127 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3431 :
3432 : /*
3433 : * Advance start_ptr over that text, to avoid multiple rescans of
3434 : * it if the replace_text contains multiple back-references.
3435 : */
3436 4127 : start_ptr += chunk_len;
3437 4127 : data_pos = pmatch[0].rm_so;
3438 : }
3439 :
3440 : /*
3441 : * Copy the replace_text, processing escapes if any are present.
3442 : */
3443 4215 : if (escape_status > 0)
3444 127 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3445 : start_ptr, data_pos);
3446 : else
3447 4088 : appendStringInfoText(&buf, replace_text);
3448 :
3449 : /* Advance start_ptr and data_pos over the matched text. */
3450 8430 : start_ptr += charlen_to_bytelen(start_ptr,
3451 4215 : pmatch[0].rm_eo - data_pos);
3452 4215 : data_pos = pmatch[0].rm_eo;
3453 :
3454 : /*
3455 : * If we only want to replace one occurrence, we're done.
3456 : */
3457 4215 : if (n > 0)
3458 1035 : break;
3459 :
3460 : /*
3461 : * Advance search position. Normally we start the next search at the
3462 : * end of the previous match; but if the match was of zero length, we
3463 : * have to advance by one character, or we'd just find the same match
3464 : * again.
3465 : */
3466 3180 : search_start = data_pos;
3467 3180 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3468 6 : search_start++;
3469 : }
3470 :
3471 : /*
3472 : * Copy the text to the right of the last match.
3473 : */
3474 9393 : if (data_pos < data_len)
3475 : {
3476 : int chunk_len;
3477 :
3478 8946 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3479 8946 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3480 : }
3481 :
3482 9393 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
3483 9393 : pfree(buf.data);
3484 9393 : pfree(data);
3485 :
3486 9393 : return ret_text;
3487 : }
3488 :
3489 : /*
3490 : * split_part
3491 : * parse input string based on provided field separator
3492 : * return N'th item (1 based, negative counts from end)
3493 : */
3494 : Datum
3495 75 : split_part(PG_FUNCTION_ARGS)
3496 : {
3497 75 : text *inputstring = PG_GETARG_TEXT_PP(0);
3498 75 : text *fldsep = PG_GETARG_TEXT_PP(1);
3499 75 : int fldnum = PG_GETARG_INT32(2);
3500 : int inputstring_len;
3501 : int fldsep_len;
3502 : TextPositionState state;
3503 : char *start_ptr;
3504 : char *end_ptr;
3505 : text *result_text;
3506 : bool found;
3507 :
3508 : /* field number is 1 based */
3509 75 : if (fldnum == 0)
3510 3 : ereport(ERROR,
3511 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3512 : errmsg("field position must not be zero")));
3513 :
3514 72 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3515 72 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3516 :
3517 : /* return empty string for empty input string */
3518 72 : if (inputstring_len < 1)
3519 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
3520 :
3521 : /* handle empty field separator */
3522 66 : if (fldsep_len < 1)
3523 : {
3524 : /* if first or last field, return input string, else empty string */
3525 12 : if (fldnum == 1 || fldnum == -1)
3526 6 : PG_RETURN_TEXT_P(inputstring);
3527 : else
3528 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
3529 : }
3530 :
3531 : /* find the first field separator */
3532 54 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
3533 :
3534 54 : found = text_position_next(&state);
3535 :
3536 : /* special case if fldsep not found at all */
3537 54 : if (!found)
3538 : {
3539 12 : text_position_cleanup(&state);
3540 : /* if first or last field, return input string, else empty string */
3541 12 : if (fldnum == 1 || fldnum == -1)
3542 6 : PG_RETURN_TEXT_P(inputstring);
3543 : else
3544 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
3545 : }
3546 :
3547 : /*
3548 : * take care of a negative field number (i.e. count from the right) by
3549 : * converting to a positive field number; we need total number of fields
3550 : */
3551 42 : if (fldnum < 0)
3552 : {
3553 : /* we found a fldsep, so there are at least two fields */
3554 21 : int numfields = 2;
3555 :
3556 27 : while (text_position_next(&state))
3557 6 : numfields++;
3558 :
3559 : /* special case of last field does not require an extra pass */
3560 21 : if (fldnum == -1)
3561 : {
3562 12 : start_ptr = text_position_get_match_ptr(&state) + state.last_match_len;
3563 12 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
3564 12 : text_position_cleanup(&state);
3565 12 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
3566 : end_ptr - start_ptr));
3567 : }
3568 :
3569 : /* else, convert fldnum to positive notation */
3570 9 : fldnum += numfields + 1;
3571 :
3572 : /* if nonexistent field, return empty string */
3573 9 : if (fldnum <= 0)
3574 : {
3575 3 : text_position_cleanup(&state);
3576 3 : PG_RETURN_TEXT_P(cstring_to_text(""));
3577 : }
3578 :
3579 : /* reset to pointing at first match, but now with positive fldnum */
3580 6 : text_position_reset(&state);
3581 6 : found = text_position_next(&state);
3582 : Assert(found);
3583 : }
3584 :
3585 : /* identify bounds of first field */
3586 27 : start_ptr = VARDATA_ANY(inputstring);
3587 27 : end_ptr = text_position_get_match_ptr(&state);
3588 :
3589 51 : while (found && --fldnum > 0)
3590 : {
3591 : /* identify bounds of next field */
3592 24 : start_ptr = end_ptr + state.last_match_len;
3593 24 : found = text_position_next(&state);
3594 24 : if (found)
3595 9 : end_ptr = text_position_get_match_ptr(&state);
3596 : }
3597 :
3598 27 : text_position_cleanup(&state);
3599 :
3600 27 : if (fldnum > 0)
3601 : {
3602 : /* N'th field separator not found */
3603 : /* if last field requested, return it, else empty string */
3604 15 : if (fldnum == 1)
3605 : {
3606 12 : int last_len = start_ptr - VARDATA_ANY(inputstring);
3607 :
3608 12 : result_text = cstring_to_text_with_len(start_ptr,
3609 : inputstring_len - last_len);
3610 : }
3611 : else
3612 3 : result_text = cstring_to_text("");
3613 : }
3614 : else
3615 : {
3616 : /* non-last field requested */
3617 12 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
3618 : }
3619 :
3620 27 : PG_RETURN_TEXT_P(result_text);
3621 : }
3622 :
3623 : /*
3624 : * Convenience function to return true when two text params are equal.
3625 : */
3626 : static bool
3627 192 : text_isequal(text *txt1, text *txt2, Oid collid)
3628 : {
3629 192 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
3630 : collid,
3631 : PointerGetDatum(txt1),
3632 : PointerGetDatum(txt2)));
3633 : }
3634 :
3635 : /*
3636 : * text_to_array
3637 : * parse input string and return text array of elements,
3638 : * based on provided field separator
3639 : */
3640 : Datum
3641 85 : text_to_array(PG_FUNCTION_ARGS)
3642 : {
3643 : SplitTextOutputData tstate;
3644 :
3645 : /* For array output, tstate should start as all zeroes */
3646 85 : memset(&tstate, 0, sizeof(tstate));
3647 :
3648 85 : if (!split_text(fcinfo, &tstate))
3649 3 : PG_RETURN_NULL();
3650 :
3651 82 : if (tstate.astate == NULL)
3652 3 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
3653 :
3654 79 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
3655 : CurrentMemoryContext));
3656 : }
3657 :
3658 : /*
3659 : * text_to_array_null
3660 : * parse input string and return text array of elements,
3661 : * based on provided field separator and null string
3662 : *
3663 : * This is a separate entry point only to prevent the regression tests from
3664 : * complaining about different argument sets for the same internal function.
3665 : */
3666 : Datum
3667 30 : text_to_array_null(PG_FUNCTION_ARGS)
3668 : {
3669 30 : return text_to_array(fcinfo);
3670 : }
3671 :
3672 : /*
3673 : * text_to_table
3674 : * parse input string and return table of elements,
3675 : * based on provided field separator
3676 : */
3677 : Datum
3678 42 : text_to_table(PG_FUNCTION_ARGS)
3679 : {
3680 42 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
3681 : SplitTextOutputData tstate;
3682 :
3683 42 : tstate.astate = NULL;
3684 42 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
3685 42 : tstate.tupstore = rsi->setResult;
3686 42 : tstate.tupdesc = rsi->setDesc;
3687 :
3688 42 : (void) split_text(fcinfo, &tstate);
3689 :
3690 42 : return (Datum) 0;
3691 : }
3692 :
3693 : /*
3694 : * text_to_table_null
3695 : * parse input string and return table of elements,
3696 : * based on provided field separator and null string
3697 : *
3698 : * This is a separate entry point only to prevent the regression tests from
3699 : * complaining about different argument sets for the same internal function.
3700 : */
3701 : Datum
3702 12 : text_to_table_null(PG_FUNCTION_ARGS)
3703 : {
3704 12 : return text_to_table(fcinfo);
3705 : }
3706 :
3707 : /*
3708 : * Common code for text_to_array, text_to_array_null, text_to_table
3709 : * and text_to_table_null functions.
3710 : *
3711 : * These are not strict so we have to test for null inputs explicitly.
3712 : * Returns false if result is to be null, else returns true.
3713 : *
3714 : * Note that if the result is valid but empty (zero elements), we return
3715 : * without changing *tstate --- caller must handle that case, too.
3716 : */
3717 : static bool
3718 127 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
3719 : {
3720 : text *inputstring;
3721 : text *fldsep;
3722 : text *null_string;
3723 127 : Oid collation = PG_GET_COLLATION();
3724 : int inputstring_len;
3725 : int fldsep_len;
3726 : char *start_ptr;
3727 : text *result_text;
3728 :
3729 : /* when input string is NULL, then result is NULL too */
3730 127 : if (PG_ARGISNULL(0))
3731 6 : return false;
3732 :
3733 121 : inputstring = PG_GETARG_TEXT_PP(0);
3734 :
3735 : /* fldsep can be NULL */
3736 121 : if (!PG_ARGISNULL(1))
3737 106 : fldsep = PG_GETARG_TEXT_PP(1);
3738 : else
3739 15 : fldsep = NULL;
3740 :
3741 : /* null_string can be NULL or omitted */
3742 121 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
3743 42 : null_string = PG_GETARG_TEXT_PP(2);
3744 : else
3745 79 : null_string = NULL;
3746 :
3747 121 : if (fldsep != NULL)
3748 : {
3749 : /*
3750 : * Normal case with non-null fldsep. Use the text_position machinery
3751 : * to search for occurrences of fldsep.
3752 : */
3753 : TextPositionState state;
3754 :
3755 106 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3756 106 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3757 :
3758 : /* return empty set for empty input string */
3759 106 : if (inputstring_len < 1)
3760 30 : return true;
3761 :
3762 : /* empty field separator: return input string as a one-element set */
3763 100 : if (fldsep_len < 1)
3764 : {
3765 24 : split_text_accum_result(tstate, inputstring,
3766 : null_string, collation);
3767 24 : return true;
3768 : }
3769 :
3770 76 : text_position_setup(inputstring, fldsep, collation, &state);
3771 :
3772 76 : start_ptr = VARDATA_ANY(inputstring);
3773 :
3774 : for (;;)
3775 256 : {
3776 : bool found;
3777 : char *end_ptr;
3778 : int chunk_len;
3779 :
3780 332 : CHECK_FOR_INTERRUPTS();
3781 :
3782 332 : found = text_position_next(&state);
3783 332 : if (!found)
3784 : {
3785 : /* fetch last field */
3786 76 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
3787 76 : end_ptr = NULL; /* not used, but some compilers complain */
3788 : }
3789 : else
3790 : {
3791 : /* fetch non-last field */
3792 256 : end_ptr = text_position_get_match_ptr(&state);
3793 256 : chunk_len = end_ptr - start_ptr;
3794 : }
3795 :
3796 : /* build a temp text datum to pass to split_text_accum_result */
3797 332 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3798 :
3799 : /* stash away this field */
3800 332 : split_text_accum_result(tstate, result_text,
3801 : null_string, collation);
3802 :
3803 332 : pfree(result_text);
3804 :
3805 332 : if (!found)
3806 76 : break;
3807 :
3808 256 : start_ptr = end_ptr + state.last_match_len;
3809 : }
3810 :
3811 76 : text_position_cleanup(&state);
3812 : }
3813 : else
3814 : {
3815 : const char *end_ptr;
3816 :
3817 : /*
3818 : * When fldsep is NULL, each character in the input string becomes a
3819 : * separate element in the result set. The separator is effectively
3820 : * the space between characters.
3821 : */
3822 15 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3823 :
3824 15 : start_ptr = VARDATA_ANY(inputstring);
3825 15 : end_ptr = start_ptr + inputstring_len;
3826 :
3827 126 : while (inputstring_len > 0)
3828 : {
3829 111 : int chunk_len = pg_mblen_range(start_ptr, end_ptr);
3830 :
3831 111 : CHECK_FOR_INTERRUPTS();
3832 :
3833 : /* build a temp text datum to pass to split_text_accum_result */
3834 111 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3835 :
3836 : /* stash away this field */
3837 111 : split_text_accum_result(tstate, result_text,
3838 : null_string, collation);
3839 :
3840 111 : pfree(result_text);
3841 :
3842 111 : start_ptr += chunk_len;
3843 111 : inputstring_len -= chunk_len;
3844 : }
3845 : }
3846 :
3847 91 : return true;
3848 : }
3849 :
3850 : /*
3851 : * Add text item to result set (table or array).
3852 : *
3853 : * This is also responsible for checking to see if the item matches
3854 : * the null_string, in which case we should emit NULL instead.
3855 : */
3856 : static void
3857 467 : split_text_accum_result(SplitTextOutputData *tstate,
3858 : text *field_value,
3859 : text *null_string,
3860 : Oid collation)
3861 : {
3862 467 : bool is_null = false;
3863 :
3864 467 : if (null_string && text_isequal(field_value, null_string, collation))
3865 36 : is_null = true;
3866 :
3867 467 : if (tstate->tupstore)
3868 : {
3869 : Datum values[1];
3870 : bool nulls[1];
3871 :
3872 114 : values[0] = PointerGetDatum(field_value);
3873 114 : nulls[0] = is_null;
3874 :
3875 114 : tuplestore_putvalues(tstate->tupstore,
3876 : tstate->tupdesc,
3877 : values,
3878 : nulls);
3879 : }
3880 : else
3881 : {
3882 353 : tstate->astate = accumArrayResult(tstate->astate,
3883 : PointerGetDatum(field_value),
3884 : is_null,
3885 : TEXTOID,
3886 : CurrentMemoryContext);
3887 : }
3888 467 : }
3889 :
3890 : /*
3891 : * array_to_text
3892 : * concatenate Cstring representation of input array elements
3893 : * using provided field separator
3894 : */
3895 : Datum
3896 38547 : array_to_text(PG_FUNCTION_ARGS)
3897 : {
3898 38547 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
3899 38547 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3900 :
3901 38547 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
3902 : }
3903 :
3904 : /*
3905 : * array_to_text_null
3906 : * concatenate Cstring representation of input array elements
3907 : * using provided field separator and null string
3908 : *
3909 : * This version is not strict so we have to test for null inputs explicitly.
3910 : */
3911 : Datum
3912 6 : array_to_text_null(PG_FUNCTION_ARGS)
3913 : {
3914 : ArrayType *v;
3915 : char *fldsep;
3916 : char *null_string;
3917 :
3918 : /* returns NULL when first or second parameter is NULL */
3919 6 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
3920 0 : PG_RETURN_NULL();
3921 :
3922 6 : v = PG_GETARG_ARRAYTYPE_P(0);
3923 6 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3924 :
3925 : /* NULL null string is passed through as a null pointer */
3926 6 : if (!PG_ARGISNULL(2))
3927 3 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
3928 : else
3929 3 : null_string = NULL;
3930 :
3931 6 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
3932 : }
3933 :
3934 : /*
3935 : * common code for array_to_text and array_to_text_null functions
3936 : */
3937 : static text *
3938 38562 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
3939 : const char *fldsep, const char *null_string)
3940 : {
3941 : text *result;
3942 : int nitems,
3943 : *dims,
3944 : ndims;
3945 : Oid element_type;
3946 : int typlen;
3947 : bool typbyval;
3948 : char typalign;
3949 : uint8 typalignby;
3950 : StringInfoData buf;
3951 38562 : bool printed = false;
3952 : char *p;
3953 : bits8 *bitmap;
3954 : int bitmask;
3955 : int i;
3956 : ArrayMetaState *my_extra;
3957 :
3958 38562 : ndims = ARR_NDIM(v);
3959 38562 : dims = ARR_DIMS(v);
3960 38562 : nitems = ArrayGetNItems(ndims, dims);
3961 :
3962 : /* if there are no elements, return an empty string */
3963 38562 : if (nitems == 0)
3964 25965 : return cstring_to_text_with_len("", 0);
3965 :
3966 12597 : element_type = ARR_ELEMTYPE(v);
3967 12597 : initStringInfo(&buf);
3968 :
3969 : /*
3970 : * We arrange to look up info about element type, including its output
3971 : * conversion proc, only once per series of calls, assuming the element
3972 : * type doesn't change underneath us.
3973 : */
3974 12597 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3975 12597 : if (my_extra == NULL)
3976 : {
3977 710 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
3978 : sizeof(ArrayMetaState));
3979 710 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3980 710 : my_extra->element_type = ~element_type;
3981 : }
3982 :
3983 12597 : if (my_extra->element_type != element_type)
3984 : {
3985 : /*
3986 : * Get info about element type, including its output conversion proc
3987 : */
3988 710 : get_type_io_data(element_type, IOFunc_output,
3989 : &my_extra->typlen, &my_extra->typbyval,
3990 : &my_extra->typalign, &my_extra->typdelim,
3991 : &my_extra->typioparam, &my_extra->typiofunc);
3992 710 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
3993 710 : fcinfo->flinfo->fn_mcxt);
3994 710 : my_extra->element_type = element_type;
3995 : }
3996 12597 : typlen = my_extra->typlen;
3997 12597 : typbyval = my_extra->typbyval;
3998 12597 : typalign = my_extra->typalign;
3999 12597 : typalignby = typalign_to_alignby(typalign);
4000 :
4001 12597 : p = ARR_DATA_PTR(v);
4002 12597 : bitmap = ARR_NULLBITMAP(v);
4003 12597 : bitmask = 1;
4004 :
4005 42828 : for (i = 0; i < nitems; i++)
4006 : {
4007 : Datum itemvalue;
4008 : char *value;
4009 :
4010 : /* Get source element, checking for NULL */
4011 30231 : if (bitmap && (*bitmap & bitmask) == 0)
4012 : {
4013 : /* if null_string is NULL, we just ignore null elements */
4014 9 : if (null_string != NULL)
4015 : {
4016 3 : if (printed)
4017 3 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
4018 : else
4019 0 : appendStringInfoString(&buf, null_string);
4020 3 : printed = true;
4021 : }
4022 : }
4023 : else
4024 : {
4025 30222 : itemvalue = fetch_att(p, typbyval, typlen);
4026 :
4027 30222 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
4028 :
4029 30222 : if (printed)
4030 17625 : appendStringInfo(&buf, "%s%s", fldsep, value);
4031 : else
4032 12597 : appendStringInfoString(&buf, value);
4033 30222 : printed = true;
4034 :
4035 30222 : p = att_addlength_pointer(p, typlen, p);
4036 30222 : p = (char *) att_nominal_alignby(p, typalignby);
4037 : }
4038 :
4039 : /* advance bitmap pointer if any */
4040 30231 : if (bitmap)
4041 : {
4042 54 : bitmask <<= 1;
4043 54 : if (bitmask == 0x100)
4044 : {
4045 0 : bitmap++;
4046 0 : bitmask = 1;
4047 : }
4048 : }
4049 : }
4050 :
4051 12597 : result = cstring_to_text_with_len(buf.data, buf.len);
4052 12597 : pfree(buf.data);
4053 :
4054 12597 : return result;
4055 : }
4056 :
4057 : /*
4058 : * Workhorse for to_bin, to_oct, and to_hex. Note that base must be > 1 and <=
4059 : * 16.
4060 : */
4061 : static inline text *
4062 19375 : convert_to_base(uint64 value, int base)
4063 : {
4064 19375 : const char *digits = "0123456789abcdef";
4065 :
4066 : /* We size the buffer for to_bin's longest possible return value. */
4067 : char buf[sizeof(uint64) * BITS_PER_BYTE];
4068 19375 : char *const end = buf + sizeof(buf);
4069 19375 : char *ptr = end;
4070 :
4071 : Assert(base > 1);
4072 : Assert(base <= 16);
4073 :
4074 : do
4075 : {
4076 37987 : *--ptr = digits[value % base];
4077 37987 : value /= base;
4078 37987 : } while (ptr > buf && value);
4079 :
4080 19375 : return cstring_to_text_with_len(ptr, end - ptr);
4081 : }
4082 :
4083 : /*
4084 : * Convert an integer to a string containing a base-2 (binary) representation
4085 : * of the number.
4086 : */
4087 : Datum
4088 6 : to_bin32(PG_FUNCTION_ARGS)
4089 : {
4090 6 : uint64 value = (uint32) PG_GETARG_INT32(0);
4091 :
4092 6 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4093 : }
4094 : Datum
4095 6 : to_bin64(PG_FUNCTION_ARGS)
4096 : {
4097 6 : uint64 value = (uint64) PG_GETARG_INT64(0);
4098 :
4099 6 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4100 : }
4101 :
4102 : /*
4103 : * Convert an integer to a string containing a base-8 (oct) representation of
4104 : * the number.
4105 : */
4106 : Datum
4107 6 : to_oct32(PG_FUNCTION_ARGS)
4108 : {
4109 6 : uint64 value = (uint32) PG_GETARG_INT32(0);
4110 :
4111 6 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4112 : }
4113 : Datum
4114 6 : to_oct64(PG_FUNCTION_ARGS)
4115 : {
4116 6 : uint64 value = (uint64) PG_GETARG_INT64(0);
4117 :
4118 6 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4119 : }
4120 :
4121 : /*
4122 : * Convert an integer to a string containing a base-16 (hex) representation of
4123 : * the number.
4124 : */
4125 : Datum
4126 19345 : to_hex32(PG_FUNCTION_ARGS)
4127 : {
4128 19345 : uint64 value = (uint32) PG_GETARG_INT32(0);
4129 :
4130 19345 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4131 : }
4132 : Datum
4133 6 : to_hex64(PG_FUNCTION_ARGS)
4134 : {
4135 6 : uint64 value = (uint64) PG_GETARG_INT64(0);
4136 :
4137 6 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4138 : }
4139 :
4140 : /*
4141 : * Return the size of a datum, possibly compressed
4142 : *
4143 : * Works on any data type
4144 : */
4145 : Datum
4146 61 : pg_column_size(PG_FUNCTION_ARGS)
4147 : {
4148 61 : Datum value = PG_GETARG_DATUM(0);
4149 : int32 result;
4150 : int typlen;
4151 :
4152 : /* On first call, get the input type's typlen, and save at *fn_extra */
4153 61 : if (fcinfo->flinfo->fn_extra == NULL)
4154 : {
4155 : /* Lookup the datatype of the supplied argument */
4156 61 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4157 :
4158 61 : typlen = get_typlen(argtypeid);
4159 61 : if (typlen == 0) /* should not happen */
4160 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4161 :
4162 61 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4163 : sizeof(int));
4164 61 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4165 : }
4166 : else
4167 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4168 :
4169 61 : if (typlen == -1)
4170 : {
4171 : /* varlena type, possibly toasted */
4172 61 : result = toast_datum_size(value);
4173 : }
4174 0 : else if (typlen == -2)
4175 : {
4176 : /* cstring */
4177 0 : result = strlen(DatumGetCString(value)) + 1;
4178 : }
4179 : else
4180 : {
4181 : /* ordinary fixed-width type */
4182 0 : result = typlen;
4183 : }
4184 :
4185 61 : PG_RETURN_INT32(result);
4186 : }
4187 :
4188 : /*
4189 : * Return the compression method stored in the compressed attribute. Return
4190 : * NULL for non varlena type or uncompressed data.
4191 : */
4192 : Datum
4193 96 : pg_column_compression(PG_FUNCTION_ARGS)
4194 : {
4195 : int typlen;
4196 : char *result;
4197 : ToastCompressionId cmid;
4198 :
4199 : /* On first call, get the input type's typlen, and save at *fn_extra */
4200 96 : if (fcinfo->flinfo->fn_extra == NULL)
4201 : {
4202 : /* Lookup the datatype of the supplied argument */
4203 78 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4204 :
4205 78 : typlen = get_typlen(argtypeid);
4206 78 : if (typlen == 0) /* should not happen */
4207 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4208 :
4209 78 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4210 : sizeof(int));
4211 78 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4212 : }
4213 : else
4214 18 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4215 :
4216 96 : if (typlen != -1)
4217 0 : PG_RETURN_NULL();
4218 :
4219 : /* get the compression method id stored in the compressed varlena */
4220 96 : cmid = toast_get_compression_id((varlena *)
4221 96 : DatumGetPointer(PG_GETARG_DATUM(0)));
4222 96 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
4223 21 : PG_RETURN_NULL();
4224 :
4225 : /* convert compression method id to compression method name */
4226 75 : switch (cmid)
4227 : {
4228 42 : case TOAST_PGLZ_COMPRESSION_ID:
4229 42 : result = "pglz";
4230 42 : break;
4231 33 : case TOAST_LZ4_COMPRESSION_ID:
4232 33 : result = "lz4";
4233 33 : break;
4234 0 : default:
4235 0 : elog(ERROR, "invalid compression method id %d", cmid);
4236 : }
4237 :
4238 75 : PG_RETURN_TEXT_P(cstring_to_text(result));
4239 : }
4240 :
4241 : /*
4242 : * Return the chunk_id of the on-disk TOASTed value. Return NULL if the value
4243 : * is un-TOASTed or not on-disk.
4244 : */
4245 : Datum
4246 26 : pg_column_toast_chunk_id(PG_FUNCTION_ARGS)
4247 : {
4248 : int typlen;
4249 : varlena *attr;
4250 : varatt_external toast_pointer;
4251 :
4252 : /* On first call, get the input type's typlen, and save at *fn_extra */
4253 26 : if (fcinfo->flinfo->fn_extra == NULL)
4254 : {
4255 : /* Lookup the datatype of the supplied argument */
4256 20 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4257 :
4258 20 : typlen = get_typlen(argtypeid);
4259 20 : if (typlen == 0) /* should not happen */
4260 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4261 :
4262 20 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4263 : sizeof(int));
4264 20 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4265 : }
4266 : else
4267 6 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4268 :
4269 26 : if (typlen != -1)
4270 0 : PG_RETURN_NULL();
4271 :
4272 26 : attr = (varlena *) DatumGetPointer(PG_GETARG_DATUM(0));
4273 :
4274 26 : if (!VARATT_IS_EXTERNAL_ONDISK(attr))
4275 6 : PG_RETURN_NULL();
4276 :
4277 20 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
4278 :
4279 20 : PG_RETURN_OID(toast_pointer.va_valueid);
4280 : }
4281 :
4282 : /*
4283 : * string_agg - Concatenates values and returns string.
4284 : *
4285 : * Syntax: string_agg(value text, delimiter text) RETURNS text
4286 : *
4287 : * Note: Any NULL values are ignored. The first-call delimiter isn't
4288 : * actually used at all, and on subsequent calls the delimiter precedes
4289 : * the associated value.
4290 : */
4291 :
4292 : /* subroutine to initialize state */
4293 : static StringInfo
4294 1216 : makeStringAggState(FunctionCallInfo fcinfo)
4295 : {
4296 : StringInfo state;
4297 : MemoryContext aggcontext;
4298 : MemoryContext oldcontext;
4299 :
4300 1216 : if (!AggCheckCallContext(fcinfo, &aggcontext))
4301 : {
4302 : /* cannot be called directly because of internal-type argument */
4303 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
4304 : }
4305 :
4306 : /*
4307 : * Create state in aggregate context. It'll stay there across subsequent
4308 : * calls.
4309 : */
4310 1216 : oldcontext = MemoryContextSwitchTo(aggcontext);
4311 1216 : state = makeStringInfo();
4312 1216 : MemoryContextSwitchTo(oldcontext);
4313 :
4314 1216 : return state;
4315 : }
4316 :
4317 : Datum
4318 547535 : string_agg_transfn(PG_FUNCTION_ARGS)
4319 : {
4320 : StringInfo state;
4321 :
4322 547535 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4323 :
4324 : /* Append the value unless null, preceding it with the delimiter. */
4325 547535 : if (!PG_ARGISNULL(1))
4326 : {
4327 540011 : text *value = PG_GETARG_TEXT_PP(1);
4328 540011 : bool isfirst = false;
4329 :
4330 : /*
4331 : * You might think we can just throw away the first delimiter, however
4332 : * we must keep it as we may be a parallel worker doing partial
4333 : * aggregation building a state to send to the main process. We need
4334 : * to keep the delimiter of every aggregation so that the combine
4335 : * function can properly join up the strings of two separately
4336 : * partially aggregated results. The first delimiter is only stripped
4337 : * off in the final function. To know how much to strip off the front
4338 : * of the string, we store the length of the first delimiter in the
4339 : * StringInfo's cursor field, which we don't otherwise need here.
4340 : */
4341 540011 : if (state == NULL)
4342 : {
4343 1036 : state = makeStringAggState(fcinfo);
4344 1036 : isfirst = true;
4345 : }
4346 :
4347 540011 : if (!PG_ARGISNULL(2))
4348 : {
4349 540011 : text *delim = PG_GETARG_TEXT_PP(2);
4350 :
4351 540011 : appendStringInfoText(state, delim);
4352 540011 : if (isfirst)
4353 1036 : state->cursor = VARSIZE_ANY_EXHDR(delim);
4354 : }
4355 :
4356 540011 : appendStringInfoText(state, value);
4357 : }
4358 :
4359 : /*
4360 : * The transition type for string_agg() is declared to be "internal",
4361 : * which is a pass-by-value type the same size as a pointer.
4362 : */
4363 547535 : if (state)
4364 547490 : PG_RETURN_POINTER(state);
4365 45 : PG_RETURN_NULL();
4366 : }
4367 :
4368 : /*
4369 : * string_agg_combine
4370 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
4371 : */
4372 : Datum
4373 120 : string_agg_combine(PG_FUNCTION_ARGS)
4374 : {
4375 : StringInfo state1;
4376 : StringInfo state2;
4377 : MemoryContext agg_context;
4378 :
4379 120 : if (!AggCheckCallContext(fcinfo, &agg_context))
4380 0 : elog(ERROR, "aggregate function called in non-aggregate context");
4381 :
4382 120 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4383 120 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
4384 :
4385 120 : if (state2 == NULL)
4386 : {
4387 : /*
4388 : * NULL state2 is easy, just return state1, which we know is already
4389 : * in the agg_context
4390 : */
4391 0 : if (state1 == NULL)
4392 0 : PG_RETURN_NULL();
4393 0 : PG_RETURN_POINTER(state1);
4394 : }
4395 :
4396 120 : if (state1 == NULL)
4397 : {
4398 : /* We must copy state2's data into the agg_context */
4399 : MemoryContext old_context;
4400 :
4401 60 : old_context = MemoryContextSwitchTo(agg_context);
4402 60 : state1 = makeStringAggState(fcinfo);
4403 60 : appendBinaryStringInfo(state1, state2->data, state2->len);
4404 60 : state1->cursor = state2->cursor;
4405 60 : MemoryContextSwitchTo(old_context);
4406 : }
4407 60 : else if (state2->len > 0)
4408 : {
4409 : /* Combine ... state1->cursor does not change in this case */
4410 60 : appendBinaryStringInfo(state1, state2->data, state2->len);
4411 : }
4412 :
4413 120 : PG_RETURN_POINTER(state1);
4414 : }
4415 :
4416 : /*
4417 : * string_agg_serialize
4418 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
4419 : *
4420 : * This is strict, so we need not handle NULL input
4421 : */
4422 : Datum
4423 120 : string_agg_serialize(PG_FUNCTION_ARGS)
4424 : {
4425 : StringInfo state;
4426 : StringInfoData buf;
4427 : bytea *result;
4428 :
4429 : /* cannot be called directly because of internal-type argument */
4430 : Assert(AggCheckCallContext(fcinfo, NULL));
4431 :
4432 120 : state = (StringInfo) PG_GETARG_POINTER(0);
4433 :
4434 120 : pq_begintypsend(&buf);
4435 :
4436 : /* cursor */
4437 120 : pq_sendint(&buf, state->cursor, 4);
4438 :
4439 : /* data */
4440 120 : pq_sendbytes(&buf, state->data, state->len);
4441 :
4442 120 : result = pq_endtypsend(&buf);
4443 :
4444 120 : PG_RETURN_BYTEA_P(result);
4445 : }
4446 :
4447 : /*
4448 : * string_agg_deserialize
4449 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
4450 : *
4451 : * This is strict, so we need not handle NULL input
4452 : */
4453 : Datum
4454 120 : string_agg_deserialize(PG_FUNCTION_ARGS)
4455 : {
4456 : bytea *sstate;
4457 : StringInfo result;
4458 : StringInfoData buf;
4459 : char *data;
4460 : int datalen;
4461 :
4462 : /* cannot be called directly because of internal-type argument */
4463 : Assert(AggCheckCallContext(fcinfo, NULL));
4464 :
4465 120 : sstate = PG_GETARG_BYTEA_PP(0);
4466 :
4467 : /*
4468 : * Initialize a StringInfo so that we can "receive" it using the standard
4469 : * recv-function infrastructure.
4470 : */
4471 120 : initReadOnlyStringInfo(&buf, VARDATA_ANY(sstate),
4472 120 : VARSIZE_ANY_EXHDR(sstate));
4473 :
4474 120 : result = makeStringAggState(fcinfo);
4475 :
4476 : /* cursor */
4477 120 : result->cursor = pq_getmsgint(&buf, 4);
4478 :
4479 : /* data */
4480 120 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
4481 120 : data = (char *) pq_getmsgbytes(&buf, datalen);
4482 120 : appendBinaryStringInfo(result, data, datalen);
4483 :
4484 120 : pq_getmsgend(&buf);
4485 :
4486 120 : PG_RETURN_POINTER(result);
4487 : }
4488 :
4489 : Datum
4490 1048 : string_agg_finalfn(PG_FUNCTION_ARGS)
4491 : {
4492 : StringInfo state;
4493 :
4494 : /* cannot be called directly because of internal-type argument */
4495 : Assert(AggCheckCallContext(fcinfo, NULL));
4496 :
4497 1048 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4498 :
4499 1048 : if (state != NULL)
4500 : {
4501 : /* As per comment in transfn, strip data before the cursor position */
4502 1006 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
4503 : state->len - state->cursor));
4504 : }
4505 : else
4506 42 : PG_RETURN_NULL();
4507 : }
4508 :
4509 : /*
4510 : * Prepare cache with fmgr info for the output functions of the datatypes of
4511 : * the arguments of a concat-like function, beginning with argument "argidx".
4512 : * (Arguments before that will have corresponding slots in the resulting
4513 : * FmgrInfo array, but we don't fill those slots.)
4514 : */
4515 : static FmgrInfo *
4516 53 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
4517 : {
4518 : FmgrInfo *foutcache;
4519 : int i;
4520 :
4521 : /* We keep the info in fn_mcxt so it survives across calls */
4522 53 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4523 53 : PG_NARGS() * sizeof(FmgrInfo));
4524 :
4525 200 : for (i = argidx; i < PG_NARGS(); i++)
4526 : {
4527 : Oid valtype;
4528 : Oid typOutput;
4529 : bool typIsVarlena;
4530 :
4531 147 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4532 147 : if (!OidIsValid(valtype))
4533 0 : elog(ERROR, "could not determine data type of concat() input");
4534 :
4535 147 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4536 147 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
4537 : }
4538 :
4539 53 : fcinfo->flinfo->fn_extra = foutcache;
4540 :
4541 53 : return foutcache;
4542 : }
4543 :
4544 : /*
4545 : * Implementation of both concat() and concat_ws().
4546 : *
4547 : * sepstr is the separator string to place between values.
4548 : * argidx identifies the first argument to concatenate (counting from zero);
4549 : * note that this must be constant across any one series of calls.
4550 : *
4551 : * Returns NULL if result should be NULL, else text value.
4552 : */
4553 : static text *
4554 132 : concat_internal(const char *sepstr, int argidx,
4555 : FunctionCallInfo fcinfo)
4556 : {
4557 : text *result;
4558 : StringInfoData str;
4559 : FmgrInfo *foutcache;
4560 132 : bool first_arg = true;
4561 : int i;
4562 :
4563 : /*
4564 : * concat(VARIADIC some-array) is essentially equivalent to
4565 : * array_to_text(), ie concat the array elements with the given separator.
4566 : * So we just pass the case off to that code.
4567 : */
4568 132 : if (get_fn_expr_variadic(fcinfo->flinfo))
4569 : {
4570 : ArrayType *arr;
4571 :
4572 : /* Should have just the one argument */
4573 : Assert(argidx == PG_NARGS() - 1);
4574 :
4575 : /* concat(VARIADIC NULL) is defined as NULL */
4576 15 : if (PG_ARGISNULL(argidx))
4577 6 : return NULL;
4578 :
4579 : /*
4580 : * Non-null argument had better be an array. We assume that any call
4581 : * context that could let get_fn_expr_variadic return true will have
4582 : * checked that a VARIADIC-labeled parameter actually is an array. So
4583 : * it should be okay to just Assert that it's an array rather than
4584 : * doing a full-fledged error check.
4585 : */
4586 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
4587 :
4588 : /* OK, safe to fetch the array value */
4589 9 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
4590 :
4591 : /*
4592 : * And serialize the array. We tell array_to_text to ignore null
4593 : * elements, which matches the behavior of the loop below.
4594 : */
4595 9 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4596 : }
4597 :
4598 : /* Normal case without explicit VARIADIC marker */
4599 117 : initStringInfo(&str);
4600 :
4601 : /* Get output function info, building it if first time through */
4602 117 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
4603 117 : if (foutcache == NULL)
4604 53 : foutcache = build_concat_foutcache(fcinfo, argidx);
4605 :
4606 411 : for (i = argidx; i < PG_NARGS(); i++)
4607 : {
4608 294 : if (!PG_ARGISNULL(i))
4609 : {
4610 255 : Datum value = PG_GETARG_DATUM(i);
4611 :
4612 : /* add separator if appropriate */
4613 255 : if (first_arg)
4614 114 : first_arg = false;
4615 : else
4616 141 : appendStringInfoString(&str, sepstr);
4617 :
4618 : /* call the appropriate type output function, append the result */
4619 255 : appendStringInfoString(&str,
4620 255 : OutputFunctionCall(&foutcache[i], value));
4621 : }
4622 : }
4623 :
4624 117 : result = cstring_to_text_with_len(str.data, str.len);
4625 117 : pfree(str.data);
4626 :
4627 117 : return result;
4628 : }
4629 :
4630 : /*
4631 : * Concatenate all arguments. NULL arguments are ignored.
4632 : */
4633 : Datum
4634 93 : text_concat(PG_FUNCTION_ARGS)
4635 : {
4636 : text *result;
4637 :
4638 93 : result = concat_internal("", 0, fcinfo);
4639 93 : if (result == NULL)
4640 3 : PG_RETURN_NULL();
4641 90 : PG_RETURN_TEXT_P(result);
4642 : }
4643 :
4644 : /*
4645 : * Concatenate all but first argument value with separators. The first
4646 : * parameter is used as the separator. NULL arguments are ignored.
4647 : */
4648 : Datum
4649 42 : text_concat_ws(PG_FUNCTION_ARGS)
4650 : {
4651 : char *sep;
4652 : text *result;
4653 :
4654 : /* return NULL when separator is NULL */
4655 42 : if (PG_ARGISNULL(0))
4656 3 : PG_RETURN_NULL();
4657 39 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
4658 :
4659 39 : result = concat_internal(sep, 1, fcinfo);
4660 39 : if (result == NULL)
4661 3 : PG_RETURN_NULL();
4662 36 : PG_RETURN_TEXT_P(result);
4663 : }
4664 :
4665 : /*
4666 : * Return first n characters in the string. When n is negative,
4667 : * return all but last |n| characters.
4668 : */
4669 : Datum
4670 1074 : text_left(PG_FUNCTION_ARGS)
4671 : {
4672 1074 : int n = PG_GETARG_INT32(1);
4673 :
4674 1074 : if (n < 0)
4675 : {
4676 15 : text *str = PG_GETARG_TEXT_PP(0);
4677 15 : const char *p = VARDATA_ANY(str);
4678 15 : int len = VARSIZE_ANY_EXHDR(str);
4679 : int rlen;
4680 :
4681 15 : n = pg_mbstrlen_with_len(p, len) + n;
4682 15 : rlen = pg_mbcharcliplen(p, len, n);
4683 15 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
4684 : }
4685 : else
4686 1059 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
4687 : }
4688 :
4689 : /*
4690 : * Return last n characters in the string. When n is negative,
4691 : * return all but first |n| characters.
4692 : */
4693 : Datum
4694 33 : text_right(PG_FUNCTION_ARGS)
4695 : {
4696 33 : text *str = PG_GETARG_TEXT_PP(0);
4697 33 : const char *p = VARDATA_ANY(str);
4698 33 : int len = VARSIZE_ANY_EXHDR(str);
4699 33 : int n = PG_GETARG_INT32(1);
4700 : int off;
4701 :
4702 33 : if (n < 0)
4703 15 : n = -n;
4704 : else
4705 18 : n = pg_mbstrlen_with_len(p, len) - n;
4706 33 : off = pg_mbcharcliplen(p, len, n);
4707 :
4708 33 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4709 : }
4710 :
4711 : /*
4712 : * Return reversed string
4713 : */
4714 : Datum
4715 21 : text_reverse(PG_FUNCTION_ARGS)
4716 : {
4717 21 : text *str = PG_GETARG_TEXT_PP(0);
4718 21 : const char *p = VARDATA_ANY(str);
4719 21 : int len = VARSIZE_ANY_EXHDR(str);
4720 21 : const char *endp = p + len;
4721 : text *result;
4722 : char *dst;
4723 :
4724 21 : result = palloc(len + VARHDRSZ);
4725 21 : dst = (char *) VARDATA(result) + len;
4726 21 : SET_VARSIZE(result, len + VARHDRSZ);
4727 :
4728 21 : if (pg_database_encoding_max_length() > 1)
4729 : {
4730 : /* multibyte version */
4731 162 : while (p < endp)
4732 : {
4733 : int sz;
4734 :
4735 144 : sz = pg_mblen_range(p, endp);
4736 141 : dst -= sz;
4737 141 : memcpy(dst, p, sz);
4738 141 : p += sz;
4739 : }
4740 : }
4741 : else
4742 : {
4743 : /* single byte version */
4744 0 : while (p < endp)
4745 0 : *(--dst) = *p++;
4746 : }
4747 :
4748 18 : PG_RETURN_TEXT_P(result);
4749 : }
4750 :
4751 :
4752 : /*
4753 : * Support macros for text_format()
4754 : */
4755 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4756 :
4757 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4758 : do { \
4759 : if (++(ptr) >= (end_ptr)) \
4760 : ereport(ERROR, \
4761 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4762 : errmsg("unterminated format() type specifier"), \
4763 : errhint("For a single \"%%\" use \"%%%%\"."))); \
4764 : } while (0)
4765 :
4766 : /*
4767 : * Returns a formatted string
4768 : */
4769 : Datum
4770 16612 : text_format(PG_FUNCTION_ARGS)
4771 : {
4772 : text *fmt;
4773 : StringInfoData str;
4774 : const char *cp;
4775 : const char *start_ptr;
4776 : const char *end_ptr;
4777 : text *result;
4778 : int arg;
4779 : bool funcvariadic;
4780 : int nargs;
4781 16612 : Datum *elements = NULL;
4782 16612 : bool *nulls = NULL;
4783 16612 : Oid element_type = InvalidOid;
4784 16612 : Oid prev_type = InvalidOid;
4785 16612 : Oid prev_width_type = InvalidOid;
4786 : FmgrInfo typoutputfinfo;
4787 : FmgrInfo typoutputinfo_width;
4788 :
4789 : /* When format string is null, immediately return null */
4790 16612 : if (PG_ARGISNULL(0))
4791 3 : PG_RETURN_NULL();
4792 :
4793 : /* If argument is marked VARIADIC, expand array into elements */
4794 16609 : if (get_fn_expr_variadic(fcinfo->flinfo))
4795 : {
4796 : ArrayType *arr;
4797 : int16 elmlen;
4798 : bool elmbyval;
4799 : char elmalign;
4800 : int nitems;
4801 :
4802 : /* Should have just the one argument */
4803 : Assert(PG_NARGS() == 2);
4804 :
4805 : /* If argument is NULL, we treat it as zero-length array */
4806 24 : if (PG_ARGISNULL(1))
4807 3 : nitems = 0;
4808 : else
4809 : {
4810 : /*
4811 : * Non-null argument had better be an array. We assume that any
4812 : * call context that could let get_fn_expr_variadic return true
4813 : * will have checked that a VARIADIC-labeled parameter actually is
4814 : * an array. So it should be okay to just Assert that it's an
4815 : * array rather than doing a full-fledged error check.
4816 : */
4817 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
4818 :
4819 : /* OK, safe to fetch the array value */
4820 21 : arr = PG_GETARG_ARRAYTYPE_P(1);
4821 :
4822 : /* Get info about array element type */
4823 21 : element_type = ARR_ELEMTYPE(arr);
4824 21 : get_typlenbyvalalign(element_type,
4825 : &elmlen, &elmbyval, &elmalign);
4826 :
4827 : /* Extract all array elements */
4828 21 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
4829 : &elements, &nulls, &nitems);
4830 : }
4831 :
4832 24 : nargs = nitems + 1;
4833 24 : funcvariadic = true;
4834 : }
4835 : else
4836 : {
4837 : /* Non-variadic case, we'll process the arguments individually */
4838 16585 : nargs = PG_NARGS();
4839 16585 : funcvariadic = false;
4840 : }
4841 :
4842 : /* Setup for main loop. */
4843 16609 : fmt = PG_GETARG_TEXT_PP(0);
4844 16609 : start_ptr = VARDATA_ANY(fmt);
4845 16609 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
4846 16609 : initStringInfo(&str);
4847 16609 : arg = 1; /* next argument position to print */
4848 :
4849 : /* Scan format string, looking for conversion specifiers. */
4850 506704 : for (cp = start_ptr; cp < end_ptr; cp++)
4851 : {
4852 : int argpos;
4853 : int widthpos;
4854 : int flags;
4855 : int width;
4856 : Datum value;
4857 : bool isNull;
4858 : Oid typid;
4859 :
4860 : /*
4861 : * If it's not the start of a conversion specifier, just copy it to
4862 : * the output buffer.
4863 : */
4864 490125 : if (*cp != '%')
4865 : {
4866 457187 : appendStringInfoCharMacro(&str, *cp);
4867 457196 : continue;
4868 : }
4869 :
4870 32938 : ADVANCE_PARSE_POINTER(cp, end_ptr);
4871 :
4872 : /* Easy case: %% outputs a single % */
4873 32938 : if (*cp == '%')
4874 : {
4875 9 : appendStringInfoCharMacro(&str, *cp);
4876 9 : continue;
4877 : }
4878 :
4879 : /* Parse the optional portions of the format specifier */
4880 32929 : cp = text_format_parse_format(cp, end_ptr,
4881 : &argpos, &widthpos,
4882 : &flags, &width);
4883 :
4884 : /*
4885 : * Next we should see the main conversion specifier. Whether or not
4886 : * an argument position was present, it's known that at least one
4887 : * character remains in the string at this point. Experience suggests
4888 : * that it's worth checking that that character is one of the expected
4889 : * ones before we try to fetch arguments, so as to produce the least
4890 : * confusing response to a mis-formatted specifier.
4891 : */
4892 32917 : if (strchr("sIL", *cp) == NULL)
4893 3 : ereport(ERROR,
4894 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4895 : errmsg("unrecognized format() type specifier \"%.*s\"",
4896 : pg_mblen_range(cp, end_ptr), cp),
4897 : errhint("For a single \"%%\" use \"%%%%\".")));
4898 :
4899 : /* If indirect width was specified, get its value */
4900 32914 : if (widthpos >= 0)
4901 : {
4902 : /* Collect the specified or next argument position */
4903 21 : if (widthpos > 0)
4904 18 : arg = widthpos;
4905 21 : if (arg >= nargs)
4906 0 : ereport(ERROR,
4907 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4908 : errmsg("too few arguments for format()")));
4909 :
4910 : /* Get the value and type of the selected argument */
4911 21 : if (!funcvariadic)
4912 : {
4913 21 : value = PG_GETARG_DATUM(arg);
4914 21 : isNull = PG_ARGISNULL(arg);
4915 21 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4916 : }
4917 : else
4918 : {
4919 0 : value = elements[arg - 1];
4920 0 : isNull = nulls[arg - 1];
4921 0 : typid = element_type;
4922 : }
4923 21 : if (!OidIsValid(typid))
4924 0 : elog(ERROR, "could not determine data type of format() input");
4925 :
4926 21 : arg++;
4927 :
4928 : /* We can treat NULL width the same as zero */
4929 21 : if (isNull)
4930 3 : width = 0;
4931 18 : else if (typid == INT4OID)
4932 18 : width = DatumGetInt32(value);
4933 0 : else if (typid == INT2OID)
4934 0 : width = DatumGetInt16(value);
4935 : else
4936 : {
4937 : /* For less-usual datatypes, convert to text then to int */
4938 : char *str;
4939 :
4940 0 : if (typid != prev_width_type)
4941 : {
4942 : Oid typoutputfunc;
4943 : bool typIsVarlena;
4944 :
4945 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4946 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
4947 0 : prev_width_type = typid;
4948 : }
4949 :
4950 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
4951 :
4952 : /* pg_strtoint32 will complain about bad data or overflow */
4953 0 : width = pg_strtoint32(str);
4954 :
4955 0 : pfree(str);
4956 : }
4957 : }
4958 :
4959 : /* Collect the specified or next argument position */
4960 32914 : if (argpos > 0)
4961 66 : arg = argpos;
4962 32914 : if (arg >= nargs)
4963 12 : ereport(ERROR,
4964 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4965 : errmsg("too few arguments for format()")));
4966 :
4967 : /* Get the value and type of the selected argument */
4968 32902 : if (!funcvariadic)
4969 : {
4970 32266 : value = PG_GETARG_DATUM(arg);
4971 32266 : isNull = PG_ARGISNULL(arg);
4972 32266 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4973 : }
4974 : else
4975 : {
4976 636 : value = elements[arg - 1];
4977 636 : isNull = nulls[arg - 1];
4978 636 : typid = element_type;
4979 : }
4980 32902 : if (!OidIsValid(typid))
4981 0 : elog(ERROR, "could not determine data type of format() input");
4982 :
4983 32902 : arg++;
4984 :
4985 : /*
4986 : * Get the appropriate typOutput function, reusing previous one if
4987 : * same type as previous argument. That's particularly useful in the
4988 : * variadic-array case, but often saves work even for ordinary calls.
4989 : */
4990 32902 : if (typid != prev_type)
4991 : {
4992 : Oid typoutputfunc;
4993 : bool typIsVarlena;
4994 :
4995 17143 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4996 17143 : fmgr_info(typoutputfunc, &typoutputfinfo);
4997 17143 : prev_type = typid;
4998 : }
4999 :
5000 : /*
5001 : * And now we can format the value.
5002 : */
5003 32902 : switch (*cp)
5004 : {
5005 32902 : case 's':
5006 : case 'I':
5007 : case 'L':
5008 32902 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
5009 : value, isNull,
5010 : flags, width);
5011 32899 : break;
5012 0 : default:
5013 : /* should not get here, because of previous check */
5014 0 : ereport(ERROR,
5015 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5016 : errmsg("unrecognized format() type specifier \"%.*s\"",
5017 : pg_mblen_range(cp, end_ptr), cp),
5018 : errhint("For a single \"%%\" use \"%%%%\".")));
5019 : break;
5020 : }
5021 : }
5022 :
5023 : /* Don't need deconstruct_array results anymore. */
5024 16579 : if (elements != NULL)
5025 21 : pfree(elements);
5026 16579 : if (nulls != NULL)
5027 21 : pfree(nulls);
5028 :
5029 : /* Generate results. */
5030 16579 : result = cstring_to_text_with_len(str.data, str.len);
5031 16579 : pfree(str.data);
5032 :
5033 16579 : PG_RETURN_TEXT_P(result);
5034 : }
5035 :
5036 : /*
5037 : * Parse contiguous digits as a decimal number.
5038 : *
5039 : * Returns true if some digits could be parsed.
5040 : * The value is returned into *value, and *ptr is advanced to the next
5041 : * character to be parsed.
5042 : *
5043 : * Note parsing invariant: at least one character is known available before
5044 : * string end (end_ptr) at entry, and this is still true at exit.
5045 : */
5046 : static bool
5047 65840 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5048 : {
5049 65840 : bool found = false;
5050 65840 : const char *cp = *ptr;
5051 65840 : int val = 0;
5052 :
5053 65996 : while (*cp >= '0' && *cp <= '9')
5054 : {
5055 159 : int8 digit = (*cp - '0');
5056 :
5057 159 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5058 159 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5059 0 : ereport(ERROR,
5060 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5061 : errmsg("number is out of range")));
5062 159 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5063 156 : found = true;
5064 : }
5065 :
5066 65837 : *ptr = cp;
5067 65837 : *value = val;
5068 :
5069 65837 : return found;
5070 : }
5071 :
5072 : /*
5073 : * Parse a format specifier (generally following the SUS printf spec).
5074 : *
5075 : * We have already advanced over the initial '%', and we are looking for
5076 : * [argpos][flags][width]type (but the type character is not consumed here).
5077 : *
5078 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5079 : * Output parameters:
5080 : * argpos: argument position for value to be printed. -1 means unspecified.
5081 : * widthpos: argument position for width. Zero means the argument position
5082 : * was unspecified (ie, take the next arg) and -1 means no width
5083 : * argument (width was omitted or specified as a constant).
5084 : * flags: bitmask of flags.
5085 : * width: directly-specified width value. Zero means the width was omitted
5086 : * (note it's not necessary to distinguish this case from an explicit
5087 : * zero width value).
5088 : *
5089 : * The function result is the next character position to be parsed, ie, the
5090 : * location where the type character is/should be.
5091 : *
5092 : * Note parsing invariant: at least one character is known available before
5093 : * string end (end_ptr) at entry, and this is still true at exit.
5094 : */
5095 : static const char *
5096 32929 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5097 : int *argpos, int *widthpos,
5098 : int *flags, int *width)
5099 : {
5100 32929 : const char *cp = start_ptr;
5101 : int n;
5102 :
5103 : /* set defaults for output parameters */
5104 32929 : *argpos = -1;
5105 32929 : *widthpos = -1;
5106 32929 : *flags = 0;
5107 32929 : *width = 0;
5108 :
5109 : /* try to identify first number */
5110 32929 : if (text_format_parse_digits(&cp, end_ptr, &n))
5111 : {
5112 87 : if (*cp != '$')
5113 : {
5114 : /* Must be just a width and a type, so we're done */
5115 12 : *width = n;
5116 12 : return cp;
5117 : }
5118 : /* The number was argument position */
5119 75 : *argpos = n;
5120 : /* Explicit 0 for argument index is immediately refused */
5121 75 : if (n == 0)
5122 3 : ereport(ERROR,
5123 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5124 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5125 72 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5126 : }
5127 :
5128 : /* Handle flags (only minus is supported now) */
5129 32926 : while (*cp == '-')
5130 : {
5131 15 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5132 15 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5133 : }
5134 :
5135 32911 : if (*cp == '*')
5136 : {
5137 : /* Handle indirect width */
5138 24 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5139 24 : if (text_format_parse_digits(&cp, end_ptr, &n))
5140 : {
5141 : /* number in this position must be closed by $ */
5142 21 : if (*cp != '$')
5143 0 : ereport(ERROR,
5144 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5145 : errmsg("width argument position must be ended by \"$\"")));
5146 : /* The number was width argument position */
5147 21 : *widthpos = n;
5148 : /* Explicit 0 for argument index is immediately refused */
5149 21 : if (n == 0)
5150 3 : ereport(ERROR,
5151 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5152 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5153 18 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5154 : }
5155 : else
5156 3 : *widthpos = 0; /* width's argument position is unspecified */
5157 : }
5158 : else
5159 : {
5160 : /* Check for direct width specification */
5161 32887 : if (text_format_parse_digits(&cp, end_ptr, &n))
5162 15 : *width = n;
5163 : }
5164 :
5165 : /* cp should now be pointing at type character */
5166 32905 : return cp;
5167 : }
5168 :
5169 : /*
5170 : * Format a %s, %I, or %L conversion
5171 : */
5172 : static void
5173 32902 : text_format_string_conversion(StringInfo buf, char conversion,
5174 : FmgrInfo *typOutputInfo,
5175 : Datum value, bool isNull,
5176 : int flags, int width)
5177 : {
5178 : char *str;
5179 :
5180 : /* Handle NULL arguments before trying to stringify the value. */
5181 32902 : if (isNull)
5182 : {
5183 171 : if (conversion == 's')
5184 135 : text_format_append_string(buf, "", flags, width);
5185 36 : else if (conversion == 'L')
5186 33 : text_format_append_string(buf, "NULL", flags, width);
5187 3 : else if (conversion == 'I')
5188 3 : ereport(ERROR,
5189 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5190 : errmsg("null values cannot be formatted as an SQL identifier")));
5191 168 : return;
5192 : }
5193 :
5194 : /* Stringify. */
5195 32731 : str = OutputFunctionCall(typOutputInfo, value);
5196 :
5197 : /* Escape. */
5198 32731 : if (conversion == 'I')
5199 : {
5200 : /* quote_identifier may or may not allocate a new string. */
5201 2453 : text_format_append_string(buf, quote_identifier(str), flags, width);
5202 : }
5203 30278 : else if (conversion == 'L')
5204 : {
5205 1626 : char *qstr = quote_literal_cstr(str);
5206 :
5207 1626 : text_format_append_string(buf, qstr, flags, width);
5208 : /* quote_literal_cstr() always allocates a new string */
5209 1626 : pfree(qstr);
5210 : }
5211 : else
5212 28652 : text_format_append_string(buf, str, flags, width);
5213 :
5214 : /* Cleanup. */
5215 32731 : pfree(str);
5216 : }
5217 :
5218 : /*
5219 : * Append str to buf, padding as directed by flags/width
5220 : */
5221 : static void
5222 32899 : text_format_append_string(StringInfo buf, const char *str,
5223 : int flags, int width)
5224 : {
5225 32899 : bool align_to_left = false;
5226 : int len;
5227 :
5228 : /* fast path for typical easy case */
5229 32899 : if (width == 0)
5230 : {
5231 32857 : appendStringInfoString(buf, str);
5232 32857 : return;
5233 : }
5234 :
5235 42 : if (width < 0)
5236 : {
5237 : /* Negative width: implicit '-' flag, then take absolute value */
5238 3 : align_to_left = true;
5239 : /* -INT_MIN is undefined */
5240 3 : if (width <= INT_MIN)
5241 0 : ereport(ERROR,
5242 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5243 : errmsg("number is out of range")));
5244 3 : width = -width;
5245 : }
5246 39 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
5247 12 : align_to_left = true;
5248 :
5249 42 : len = pg_mbstrlen(str);
5250 42 : if (align_to_left)
5251 : {
5252 : /* left justify */
5253 15 : appendStringInfoString(buf, str);
5254 15 : if (len < width)
5255 15 : appendStringInfoSpaces(buf, width - len);
5256 : }
5257 : else
5258 : {
5259 : /* right justify */
5260 27 : if (len < width)
5261 27 : appendStringInfoSpaces(buf, width - len);
5262 27 : appendStringInfoString(buf, str);
5263 : }
5264 : }
5265 :
5266 : /*
5267 : * text_format_nv - nonvariadic wrapper for text_format function.
5268 : *
5269 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5270 : * which checks that all built-in functions that share the implementing C
5271 : * function take the same number of arguments.
5272 : */
5273 : Datum
5274 1905 : text_format_nv(PG_FUNCTION_ARGS)
5275 : {
5276 1905 : return text_format(fcinfo);
5277 : }
5278 :
5279 : /*
5280 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
5281 : * for this use case.
5282 : */
5283 : static inline bool
5284 0 : rest_of_char_same(const char *s1, const char *s2, int len)
5285 : {
5286 0 : while (len > 0)
5287 : {
5288 0 : len--;
5289 0 : if (s1[len] != s2[len])
5290 0 : return false;
5291 : }
5292 0 : return true;
5293 : }
5294 :
5295 : /* Expand each Levenshtein distance variant */
5296 : #include "levenshtein.c"
5297 : #define LEVENSHTEIN_LESS_EQUAL
5298 : #include "levenshtein.c"
5299 :
5300 :
5301 : /*
5302 : * The following *ClosestMatch() functions can be used to determine whether a
5303 : * user-provided string resembles any known valid values, which is useful for
5304 : * providing hints in log messages, among other things. Use these functions
5305 : * like so:
5306 : *
5307 : * initClosestMatch(&state, source_string, max_distance);
5308 : *
5309 : * for (int i = 0; i < num_valid_strings; i++)
5310 : * updateClosestMatch(&state, valid_strings[i]);
5311 : *
5312 : * closestMatch = getClosestMatch(&state);
5313 : */
5314 :
5315 : /*
5316 : * Initialize the given state with the source string and maximum Levenshtein
5317 : * distance to consider.
5318 : */
5319 : void
5320 39 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
5321 : {
5322 : Assert(state);
5323 : Assert(max_d >= 0);
5324 :
5325 39 : state->source = source;
5326 39 : state->min_d = -1;
5327 39 : state->max_d = max_d;
5328 39 : state->match = NULL;
5329 39 : }
5330 :
5331 : /*
5332 : * If the candidate string is a closer match than the current one saved (or
5333 : * there is no match saved), save it as the closest match.
5334 : *
5335 : * If the source or candidate string is NULL, empty, or too long, this function
5336 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
5337 : * allowed or more than half the characters are different, no action is taken.
5338 : */
5339 : void
5340 402 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
5341 : {
5342 : int dist;
5343 :
5344 : Assert(state);
5345 :
5346 402 : if (state->source == NULL || state->source[0] == '\0' ||
5347 402 : candidate == NULL || candidate[0] == '\0')
5348 0 : return;
5349 :
5350 : /*
5351 : * To avoid ERROR-ing, we check the lengths here instead of setting
5352 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
5353 : */
5354 402 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
5355 402 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
5356 0 : return;
5357 :
5358 402 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
5359 402 : candidate, strlen(candidate), 1, 1, 1,
5360 : state->max_d, true);
5361 402 : if (dist <= state->max_d &&
5362 31 : dist <= strlen(state->source) / 2 &&
5363 7 : (state->min_d == -1 || dist < state->min_d))
5364 : {
5365 7 : state->min_d = dist;
5366 7 : state->match = candidate;
5367 : }
5368 : }
5369 :
5370 : /*
5371 : * Return the closest match. If no suitable candidates were provided via
5372 : * updateClosestMatch(), return NULL.
5373 : */
5374 : const char *
5375 39 : getClosestMatch(ClosestMatchState *state)
5376 : {
5377 : Assert(state);
5378 :
5379 39 : return state->match;
5380 : }
5381 :
5382 :
5383 : /*
5384 : * Unicode support
5385 : */
5386 :
5387 : static UnicodeNormalizationForm
5388 105 : unicode_norm_form_from_string(const char *formstr)
5389 : {
5390 105 : UnicodeNormalizationForm form = -1;
5391 :
5392 : /*
5393 : * Might as well check this while we're here.
5394 : */
5395 105 : if (GetDatabaseEncoding() != PG_UTF8)
5396 0 : ereport(ERROR,
5397 : (errcode(ERRCODE_SYNTAX_ERROR),
5398 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
5399 :
5400 105 : if (pg_strcasecmp(formstr, "NFC") == 0)
5401 33 : form = UNICODE_NFC;
5402 72 : else if (pg_strcasecmp(formstr, "NFD") == 0)
5403 30 : form = UNICODE_NFD;
5404 42 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
5405 18 : form = UNICODE_NFKC;
5406 24 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
5407 18 : form = UNICODE_NFKD;
5408 : else
5409 6 : ereport(ERROR,
5410 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5411 : errmsg("invalid normalization form: %s", formstr)));
5412 :
5413 99 : return form;
5414 : }
5415 :
5416 : /*
5417 : * Returns version of Unicode used by Postgres in "major.minor" format (the
5418 : * same format as the Unicode version reported by ICU). The third component
5419 : * ("update version") never involves additions to the character repertoire and
5420 : * is unimportant for most purposes.
5421 : *
5422 : * See: https://unicode.org/versions/
5423 : */
5424 : Datum
5425 17 : unicode_version(PG_FUNCTION_ARGS)
5426 : {
5427 17 : PG_RETURN_TEXT_P(cstring_to_text(PG_UNICODE_VERSION));
5428 : }
5429 :
5430 : /*
5431 : * Returns version of Unicode used by ICU, if enabled; otherwise NULL.
5432 : */
5433 : Datum
5434 1 : icu_unicode_version(PG_FUNCTION_ARGS)
5435 : {
5436 1 : const char *version = pg_icu_unicode_version();
5437 :
5438 1 : if (version)
5439 1 : PG_RETURN_TEXT_P(cstring_to_text(version));
5440 : else
5441 0 : PG_RETURN_NULL();
5442 : }
5443 :
5444 : /*
5445 : * Check whether the string contains only assigned Unicode code
5446 : * points. Requires that the database encoding is UTF-8.
5447 : */
5448 : Datum
5449 6 : unicode_assigned(PG_FUNCTION_ARGS)
5450 : {
5451 6 : text *input = PG_GETARG_TEXT_PP(0);
5452 : unsigned char *p;
5453 : int size;
5454 :
5455 6 : if (GetDatabaseEncoding() != PG_UTF8)
5456 0 : ereport(ERROR,
5457 : (errmsg("Unicode categorization can only be performed if server encoding is UTF8")));
5458 :
5459 : /* convert to char32_t */
5460 6 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5461 6 : p = (unsigned char *) VARDATA_ANY(input);
5462 24 : for (int i = 0; i < size; i++)
5463 : {
5464 21 : char32_t uchar = utf8_to_unicode(p);
5465 21 : int category = unicode_category(uchar);
5466 :
5467 21 : if (category == PG_U_UNASSIGNED)
5468 3 : PG_RETURN_BOOL(false);
5469 :
5470 18 : p += pg_utf_mblen(p);
5471 : }
5472 :
5473 3 : PG_RETURN_BOOL(true);
5474 : }
5475 :
5476 : Datum
5477 36 : unicode_normalize_func(PG_FUNCTION_ARGS)
5478 : {
5479 36 : text *input = PG_GETARG_TEXT_PP(0);
5480 36 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5481 : UnicodeNormalizationForm form;
5482 : int size;
5483 : char32_t *input_chars;
5484 : char32_t *output_chars;
5485 : unsigned char *p;
5486 : text *result;
5487 : int i;
5488 :
5489 36 : form = unicode_norm_form_from_string(formstr);
5490 :
5491 : /* convert to char32_t */
5492 33 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5493 33 : input_chars = palloc((size + 1) * sizeof(char32_t));
5494 33 : p = (unsigned char *) VARDATA_ANY(input);
5495 144 : for (i = 0; i < size; i++)
5496 : {
5497 111 : input_chars[i] = utf8_to_unicode(p);
5498 111 : p += pg_utf_mblen(p);
5499 : }
5500 33 : input_chars[i] = (char32_t) '\0';
5501 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5502 :
5503 : /* action */
5504 33 : output_chars = unicode_normalize(form, input_chars);
5505 :
5506 : /* convert back to UTF-8 string */
5507 33 : size = 0;
5508 153 : for (char32_t *wp = output_chars; *wp; wp++)
5509 : {
5510 : unsigned char buf[4];
5511 :
5512 120 : unicode_to_utf8(*wp, buf);
5513 120 : size += pg_utf_mblen(buf);
5514 : }
5515 :
5516 33 : result = palloc(size + VARHDRSZ);
5517 33 : SET_VARSIZE(result, size + VARHDRSZ);
5518 :
5519 33 : p = (unsigned char *) VARDATA_ANY(result);
5520 153 : for (char32_t *wp = output_chars; *wp; wp++)
5521 : {
5522 120 : unicode_to_utf8(*wp, p);
5523 120 : p += pg_utf_mblen(p);
5524 : }
5525 : Assert((char *) p == (char *) result + size + VARHDRSZ);
5526 :
5527 33 : PG_RETURN_TEXT_P(result);
5528 : }
5529 :
5530 : /*
5531 : * Check whether the string is in the specified Unicode normalization form.
5532 : *
5533 : * This is done by converting the string to the specified normal form and then
5534 : * comparing that to the original string. To speed that up, we also apply the
5535 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
5536 : * answer for many strings by just scanning the string once.
5537 : *
5538 : * This function should generally be optimized for the case where the string
5539 : * is in fact normalized. In that case, we'll end up looking at the entire
5540 : * string, so it's probably not worth doing any incremental conversion etc.
5541 : */
5542 : Datum
5543 69 : unicode_is_normalized(PG_FUNCTION_ARGS)
5544 : {
5545 69 : text *input = PG_GETARG_TEXT_PP(0);
5546 69 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5547 : UnicodeNormalizationForm form;
5548 : int size;
5549 : char32_t *input_chars;
5550 : char32_t *output_chars;
5551 : unsigned char *p;
5552 : int i;
5553 : UnicodeNormalizationQC quickcheck;
5554 : int output_size;
5555 : bool result;
5556 :
5557 69 : form = unicode_norm_form_from_string(formstr);
5558 :
5559 : /* convert to char32_t */
5560 66 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5561 66 : input_chars = palloc((size + 1) * sizeof(char32_t));
5562 66 : p = (unsigned char *) VARDATA_ANY(input);
5563 252 : for (i = 0; i < size; i++)
5564 : {
5565 186 : input_chars[i] = utf8_to_unicode(p);
5566 186 : p += pg_utf_mblen(p);
5567 : }
5568 66 : input_chars[i] = (char32_t) '\0';
5569 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5570 :
5571 : /* quick check (see UAX #15) */
5572 66 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
5573 66 : if (quickcheck == UNICODE_NORM_QC_YES)
5574 21 : PG_RETURN_BOOL(true);
5575 45 : else if (quickcheck == UNICODE_NORM_QC_NO)
5576 6 : PG_RETURN_BOOL(false);
5577 :
5578 : /* normalize and compare with original */
5579 39 : output_chars = unicode_normalize(form, input_chars);
5580 :
5581 39 : output_size = 0;
5582 162 : for (char32_t *wp = output_chars; *wp; wp++)
5583 123 : output_size++;
5584 :
5585 57 : result = (size == output_size) &&
5586 18 : (memcmp(input_chars, output_chars, size * sizeof(char32_t)) == 0);
5587 :
5588 39 : PG_RETURN_BOOL(result);
5589 : }
5590 :
5591 : /*
5592 : * Check if first n chars are hexadecimal digits
5593 : */
5594 : static bool
5595 78 : isxdigits_n(const char *instr, size_t n)
5596 : {
5597 330 : for (size_t i = 0; i < n; i++)
5598 285 : if (!isxdigit((unsigned char) instr[i]))
5599 33 : return false;
5600 :
5601 45 : return true;
5602 : }
5603 :
5604 : static unsigned int
5605 252 : hexval(unsigned char c)
5606 : {
5607 252 : if (c >= '0' && c <= '9')
5608 192 : return c - '0';
5609 60 : if (c >= 'a' && c <= 'f')
5610 30 : return c - 'a' + 0xA;
5611 30 : if (c >= 'A' && c <= 'F')
5612 30 : return c - 'A' + 0xA;
5613 0 : elog(ERROR, "invalid hexadecimal digit");
5614 : return 0; /* not reached */
5615 : }
5616 :
5617 : /*
5618 : * Translate string with hexadecimal digits to number
5619 : */
5620 : static unsigned int
5621 45 : hexval_n(const char *instr, size_t n)
5622 : {
5623 45 : unsigned int result = 0;
5624 :
5625 297 : for (size_t i = 0; i < n; i++)
5626 252 : result += hexval(instr[i]) << (4 * (n - i - 1));
5627 :
5628 45 : return result;
5629 : }
5630 :
5631 : /*
5632 : * Replaces Unicode escape sequences by Unicode characters
5633 : */
5634 : Datum
5635 33 : unistr(PG_FUNCTION_ARGS)
5636 : {
5637 33 : text *input_text = PG_GETARG_TEXT_PP(0);
5638 : char *instr;
5639 : int len;
5640 : StringInfoData str;
5641 : text *result;
5642 33 : char16_t pair_first = 0;
5643 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
5644 :
5645 33 : instr = VARDATA_ANY(input_text);
5646 33 : len = VARSIZE_ANY_EXHDR(input_text);
5647 :
5648 33 : initStringInfo(&str);
5649 :
5650 255 : while (len > 0)
5651 : {
5652 243 : if (instr[0] == '\\')
5653 : {
5654 51 : if (len >= 2 &&
5655 51 : instr[1] == '\\')
5656 : {
5657 3 : if (pair_first)
5658 0 : goto invalid_pair;
5659 3 : appendStringInfoChar(&str, '\\');
5660 3 : instr += 2;
5661 3 : len -= 2;
5662 : }
5663 48 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
5664 33 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
5665 15 : {
5666 : char32_t unicode;
5667 21 : int offset = instr[1] == 'u' ? 2 : 1;
5668 :
5669 21 : unicode = hexval_n(instr + offset, 4);
5670 :
5671 21 : if (!is_valid_unicode_codepoint(unicode))
5672 0 : ereport(ERROR,
5673 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5674 : errmsg("invalid Unicode code point: %04X", unicode));
5675 :
5676 21 : if (pair_first)
5677 : {
5678 6 : if (is_utf16_surrogate_second(unicode))
5679 : {
5680 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5681 0 : pair_first = 0;
5682 : }
5683 : else
5684 6 : goto invalid_pair;
5685 : }
5686 15 : else if (is_utf16_surrogate_second(unicode))
5687 0 : goto invalid_pair;
5688 :
5689 15 : if (is_utf16_surrogate_first(unicode))
5690 9 : pair_first = unicode;
5691 : else
5692 : {
5693 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5694 6 : appendStringInfoString(&str, cbuf);
5695 : }
5696 :
5697 15 : instr += 4 + offset;
5698 15 : len -= 4 + offset;
5699 : }
5700 27 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
5701 6 : {
5702 : char32_t unicode;
5703 :
5704 12 : unicode = hexval_n(instr + 2, 6);
5705 :
5706 12 : if (!is_valid_unicode_codepoint(unicode))
5707 3 : ereport(ERROR,
5708 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5709 : errmsg("invalid Unicode code point: %04X", unicode));
5710 :
5711 9 : if (pair_first)
5712 : {
5713 3 : if (is_utf16_surrogate_second(unicode))
5714 : {
5715 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5716 0 : pair_first = 0;
5717 : }
5718 : else
5719 3 : goto invalid_pair;
5720 : }
5721 6 : else if (is_utf16_surrogate_second(unicode))
5722 0 : goto invalid_pair;
5723 :
5724 6 : if (is_utf16_surrogate_first(unicode))
5725 3 : pair_first = unicode;
5726 : else
5727 : {
5728 3 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5729 3 : appendStringInfoString(&str, cbuf);
5730 : }
5731 :
5732 6 : instr += 8;
5733 6 : len -= 8;
5734 : }
5735 15 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
5736 6 : {
5737 : char32_t unicode;
5738 :
5739 12 : unicode = hexval_n(instr + 2, 8);
5740 :
5741 12 : if (!is_valid_unicode_codepoint(unicode))
5742 3 : ereport(ERROR,
5743 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5744 : errmsg("invalid Unicode code point: %04X", unicode));
5745 :
5746 9 : if (pair_first)
5747 : {
5748 3 : if (is_utf16_surrogate_second(unicode))
5749 : {
5750 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5751 0 : pair_first = 0;
5752 : }
5753 : else
5754 3 : goto invalid_pair;
5755 : }
5756 6 : else if (is_utf16_surrogate_second(unicode))
5757 0 : goto invalid_pair;
5758 :
5759 6 : if (is_utf16_surrogate_first(unicode))
5760 3 : pair_first = unicode;
5761 : else
5762 : {
5763 3 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5764 3 : appendStringInfoString(&str, cbuf);
5765 : }
5766 :
5767 6 : instr += 10;
5768 6 : len -= 10;
5769 : }
5770 : else
5771 3 : ereport(ERROR,
5772 : (errcode(ERRCODE_SYNTAX_ERROR),
5773 : errmsg("invalid Unicode escape"),
5774 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
5775 : }
5776 : else
5777 : {
5778 192 : if (pair_first)
5779 0 : goto invalid_pair;
5780 :
5781 192 : appendStringInfoChar(&str, *instr++);
5782 192 : len--;
5783 : }
5784 : }
5785 :
5786 : /* unfinished surrogate pair? */
5787 12 : if (pair_first)
5788 3 : goto invalid_pair;
5789 :
5790 9 : result = cstring_to_text_with_len(str.data, str.len);
5791 9 : pfree(str.data);
5792 :
5793 9 : PG_RETURN_TEXT_P(result);
5794 :
5795 15 : invalid_pair:
5796 15 : ereport(ERROR,
5797 : (errcode(ERRCODE_SYNTAX_ERROR),
5798 : errmsg("invalid Unicode surrogate pair")));
5799 : PG_RETURN_NULL(); /* keep compiler quiet */
5800 : }
|
