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 struct 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 text *text_overlay(text *t1, text *t2, int sp, int sl);
137 : static int text_position(text *t1, text *t2, Oid collid);
138 : static void text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state);
139 : static bool text_position_next(TextPositionState *state);
140 : static char *text_position_next_internal(char *start_ptr, TextPositionState *state);
141 : static char *text_position_get_match_ptr(TextPositionState *state);
142 : static int text_position_get_match_pos(TextPositionState *state);
143 : static void text_position_cleanup(TextPositionState *state);
144 : static void check_collation_set(Oid collid);
145 : static int text_cmp(text *arg1, text *arg2, Oid collid);
146 : static void appendStringInfoText(StringInfo str, const text *t);
147 : static bool split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate);
148 : static void split_text_accum_result(SplitTextOutputData *tstate,
149 : text *field_value,
150 : text *null_string,
151 : Oid collation);
152 : static text *array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
153 : const char *fldsep, const char *null_string);
154 : static StringInfo makeStringAggState(FunctionCallInfo fcinfo);
155 : static bool text_format_parse_digits(const char **ptr, const char *end_ptr,
156 : int *value);
157 : static const char *text_format_parse_format(const char *start_ptr,
158 : const char *end_ptr,
159 : int *argpos, int *widthpos,
160 : int *flags, int *width);
161 : static void text_format_string_conversion(StringInfo buf, char conversion,
162 : FmgrInfo *typOutputInfo,
163 : Datum value, bool isNull,
164 : int flags, int width);
165 : static void text_format_append_string(StringInfo buf, const char *str,
166 : int flags, int width);
167 :
168 :
169 : /*****************************************************************************
170 : * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
171 : *****************************************************************************/
172 :
173 : /*
174 : * cstring_to_text
175 : *
176 : * Create a text value from a null-terminated C string.
177 : *
178 : * The new text value is freshly palloc'd with a full-size VARHDR.
179 : */
180 : text *
181 25654704 : cstring_to_text(const char *s)
182 : {
183 25654704 : return cstring_to_text_with_len(s, strlen(s));
184 : }
185 :
186 : /*
187 : * cstring_to_text_with_len
188 : *
189 : * Same as cstring_to_text except the caller specifies the string length;
190 : * the string need not be null_terminated.
191 : */
192 : text *
193 28385968 : cstring_to_text_with_len(const char *s, int len)
194 : {
195 28385968 : text *result = (text *) palloc(len + VARHDRSZ);
196 :
197 28385968 : SET_VARSIZE(result, len + VARHDRSZ);
198 28385968 : memcpy(VARDATA(result), s, len);
199 :
200 28385968 : return result;
201 : }
202 :
203 : /*
204 : * text_to_cstring
205 : *
206 : * Create a palloc'd, null-terminated C string from a text value.
207 : *
208 : * We support being passed a compressed or toasted text value.
209 : * This is a bit bogus since such values shouldn't really be referred to as
210 : * "text *", but it seems useful for robustness. If we didn't handle that
211 : * case here, we'd need another routine that did, anyway.
212 : */
213 : char *
214 18417214 : text_to_cstring(const text *t)
215 : {
216 : /* must cast away the const, unfortunately */
217 18417214 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
218 18417214 : int len = VARSIZE_ANY_EXHDR(tunpacked);
219 : char *result;
220 :
221 18417214 : result = (char *) palloc(len + 1);
222 18417214 : memcpy(result, VARDATA_ANY(tunpacked), len);
223 18417214 : result[len] = '\0';
224 :
225 18417214 : if (tunpacked != t)
226 45908 : pfree(tunpacked);
227 :
228 18417214 : return result;
229 : }
230 :
231 : /*
232 : * text_to_cstring_buffer
233 : *
234 : * Copy a text value into a caller-supplied buffer of size dst_len.
235 : *
236 : * The text string is truncated if necessary to fit. The result is
237 : * guaranteed null-terminated (unless dst_len == 0).
238 : *
239 : * We support being passed a compressed or toasted text value.
240 : * This is a bit bogus since such values shouldn't really be referred to as
241 : * "text *", but it seems useful for robustness. If we didn't handle that
242 : * case here, we'd need another routine that did, anyway.
243 : */
244 : void
245 1006 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
246 : {
247 : /* must cast away the const, unfortunately */
248 1006 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
249 1006 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
250 :
251 1006 : if (dst_len > 0)
252 : {
253 1006 : dst_len--;
254 1006 : if (dst_len >= src_len)
255 1006 : dst_len = src_len;
256 : else /* ensure truncation is encoding-safe */
257 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
258 1006 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
259 1006 : dst[dst_len] = '\0';
260 : }
261 :
262 1006 : if (srcunpacked != src)
263 0 : pfree(srcunpacked);
264 1006 : }
265 :
266 :
267 : /*****************************************************************************
268 : * USER I/O ROUTINES *
269 : *****************************************************************************/
270 :
271 : /*
272 : * textin - converts cstring to internal representation
273 : */
274 : Datum
275 22327506 : textin(PG_FUNCTION_ARGS)
276 : {
277 22327506 : char *inputText = PG_GETARG_CSTRING(0);
278 :
279 22327506 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
280 : }
281 :
282 : /*
283 : * textout - converts internal representation to cstring
284 : */
285 : Datum
286 8274446 : textout(PG_FUNCTION_ARGS)
287 : {
288 8274446 : Datum txt = PG_GETARG_DATUM(0);
289 :
290 8274446 : PG_RETURN_CSTRING(TextDatumGetCString(txt));
291 : }
292 :
293 : /*
294 : * textrecv - converts external binary format to text
295 : */
296 : Datum
297 48 : textrecv(PG_FUNCTION_ARGS)
298 : {
299 48 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
300 : text *result;
301 : char *str;
302 : int nbytes;
303 :
304 48 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
305 :
306 48 : result = cstring_to_text_with_len(str, nbytes);
307 48 : pfree(str);
308 48 : PG_RETURN_TEXT_P(result);
309 : }
310 :
311 : /*
312 : * textsend - converts text to binary format
313 : */
314 : Datum
315 4720 : textsend(PG_FUNCTION_ARGS)
316 : {
317 4720 : text *t = PG_GETARG_TEXT_PP(0);
318 : StringInfoData buf;
319 :
320 4720 : pq_begintypsend(&buf);
321 4720 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
322 4720 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
323 : }
324 :
325 :
326 : /*
327 : * unknownin - converts cstring to internal representation
328 : */
329 : Datum
330 0 : unknownin(PG_FUNCTION_ARGS)
331 : {
332 0 : char *str = PG_GETARG_CSTRING(0);
333 :
334 : /* representation is same as cstring */
335 0 : PG_RETURN_CSTRING(pstrdup(str));
336 : }
337 :
338 : /*
339 : * unknownout - converts internal representation to cstring
340 : */
341 : Datum
342 940 : unknownout(PG_FUNCTION_ARGS)
343 : {
344 : /* representation is same as cstring */
345 940 : char *str = PG_GETARG_CSTRING(0);
346 :
347 940 : PG_RETURN_CSTRING(pstrdup(str));
348 : }
349 :
350 : /*
351 : * unknownrecv - converts external binary format to unknown
352 : */
353 : Datum
354 0 : unknownrecv(PG_FUNCTION_ARGS)
355 : {
356 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
357 : char *str;
358 : int nbytes;
359 :
360 0 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
361 : /* representation is same as cstring */
362 0 : PG_RETURN_CSTRING(str);
363 : }
364 :
365 : /*
366 : * unknownsend - converts unknown to binary format
367 : */
368 : Datum
369 0 : unknownsend(PG_FUNCTION_ARGS)
370 : {
371 : /* representation is same as cstring */
372 0 : char *str = PG_GETARG_CSTRING(0);
373 : StringInfoData buf;
374 :
375 0 : pq_begintypsend(&buf);
376 0 : pq_sendtext(&buf, str, strlen(str));
377 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
378 : }
379 :
380 :
381 : /* ========== PUBLIC ROUTINES ========== */
382 :
383 : /*
384 : * textlen -
385 : * returns the logical length of a text*
386 : * (which is less than the VARSIZE of the text*)
387 : */
388 : Datum
389 430872 : textlen(PG_FUNCTION_ARGS)
390 : {
391 430872 : Datum str = PG_GETARG_DATUM(0);
392 :
393 : /* try to avoid decompressing argument */
394 430872 : PG_RETURN_INT32(text_length(str));
395 : }
396 :
397 : /*
398 : * text_length -
399 : * Does the real work for textlen()
400 : *
401 : * This is broken out so it can be called directly by other string processing
402 : * functions. Note that the argument is passed as a Datum, to indicate that
403 : * it may still be in compressed form. We can avoid decompressing it at all
404 : * in some cases.
405 : */
406 : static int32
407 430884 : text_length(Datum str)
408 : {
409 : /* fastpath when max encoding length is one */
410 430884 : if (pg_database_encoding_max_length() == 1)
411 20 : return (toast_raw_datum_size(str) - VARHDRSZ);
412 : else
413 : {
414 430864 : text *t = DatumGetTextPP(str);
415 :
416 430864 : return (pg_mbstrlen_with_len(VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t)));
417 : }
418 : }
419 :
420 : /*
421 : * textoctetlen -
422 : * returns the physical length of a text*
423 : * (which is less than the VARSIZE of the text*)
424 : */
425 : Datum
426 70 : textoctetlen(PG_FUNCTION_ARGS)
427 : {
428 70 : Datum str = PG_GETARG_DATUM(0);
429 :
430 : /* We need not detoast the input at all */
431 70 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
432 : }
433 :
434 : /*
435 : * textcat -
436 : * takes two text* and returns a text* that is the concatenation of
437 : * the two.
438 : *
439 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
440 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
441 : * Allocate space for output in all cases.
442 : * XXX - thomas 1997-07-10
443 : */
444 : Datum
445 1962158 : textcat(PG_FUNCTION_ARGS)
446 : {
447 1962158 : text *t1 = PG_GETARG_TEXT_PP(0);
448 1962158 : text *t2 = PG_GETARG_TEXT_PP(1);
449 :
450 1962158 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
451 : }
452 :
453 : /*
454 : * text_catenate
455 : * Guts of textcat(), broken out so it can be used by other functions
456 : *
457 : * Arguments can be in short-header form, but not compressed or out-of-line
458 : */
459 : static text *
460 1962238 : text_catenate(text *t1, text *t2)
461 : {
462 : text *result;
463 : int len1,
464 : len2,
465 : len;
466 : char *ptr;
467 :
468 1962238 : len1 = VARSIZE_ANY_EXHDR(t1);
469 1962238 : len2 = VARSIZE_ANY_EXHDR(t2);
470 :
471 : /* paranoia ... probably should throw error instead? */
472 1962238 : if (len1 < 0)
473 0 : len1 = 0;
474 1962238 : if (len2 < 0)
475 0 : len2 = 0;
476 :
477 1962238 : len = len1 + len2 + VARHDRSZ;
478 1962238 : result = (text *) palloc(len);
479 :
480 : /* Set size of result string... */
481 1962238 : SET_VARSIZE(result, len);
482 :
483 : /* Fill data field of result string... */
484 1962238 : ptr = VARDATA(result);
485 1962238 : if (len1 > 0)
486 1961414 : memcpy(ptr, VARDATA_ANY(t1), len1);
487 1962238 : if (len2 > 0)
488 1962028 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
489 :
490 1962238 : return result;
491 : }
492 :
493 : /*
494 : * charlen_to_bytelen()
495 : * Compute the number of bytes occupied by n characters starting at *p
496 : *
497 : * It is caller's responsibility that there actually are n characters;
498 : * the string need not be null-terminated.
499 : */
500 : static int
501 17274 : charlen_to_bytelen(const char *p, int n)
502 : {
503 17274 : if (pg_database_encoding_max_length() == 1)
504 : {
505 : /* Optimization for single-byte encodings */
506 180 : return n;
507 : }
508 : else
509 : {
510 : const char *s;
511 :
512 6064214 : for (s = p; n > 0; n--)
513 6047120 : s += pg_mblen(s);
514 :
515 17094 : return s - p;
516 : }
517 : }
518 :
519 : /*
520 : * text_substr()
521 : * Return a substring starting at the specified position.
522 : * - thomas 1997-12-31
523 : *
524 : * Input:
525 : * - string
526 : * - starting position (is one-based)
527 : * - string length
528 : *
529 : * If the starting position is zero or less, then return from the start of the string
530 : * adjusting the length to be consistent with the "negative start" per SQL.
531 : * If the length is less than zero, return the remaining string.
532 : *
533 : * Added multibyte support.
534 : * - Tatsuo Ishii 1998-4-21
535 : * Changed behavior if starting position is less than one to conform to SQL behavior.
536 : * Formerly returned the entire string; now returns a portion.
537 : * - Thomas Lockhart 1998-12-10
538 : * Now uses faster TOAST-slicing interface
539 : * - John Gray 2002-02-22
540 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
541 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
542 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
543 : * S > LC and < LC + 4 sometimes garbage characters are returned.
544 : * - Joe Conway 2002-08-10
545 : */
546 : Datum
547 661594 : text_substr(PG_FUNCTION_ARGS)
548 : {
549 661594 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
550 : PG_GETARG_INT32(1),
551 : PG_GETARG_INT32(2),
552 : false));
553 : }
554 :
555 : /*
556 : * text_substr_no_len -
557 : * Wrapper to avoid opr_sanity failure due to
558 : * one function accepting a different number of args.
559 : */
560 : Datum
561 36 : text_substr_no_len(PG_FUNCTION_ARGS)
562 : {
563 36 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
564 : PG_GETARG_INT32(1),
565 : -1, true));
566 : }
567 :
568 : /*
569 : * text_substring -
570 : * Does the real work for text_substr() and text_substr_no_len()
571 : *
572 : * This is broken out so it can be called directly by other string processing
573 : * functions. Note that the argument is passed as a Datum, to indicate that
574 : * it may still be in compressed/toasted form. We can avoid detoasting all
575 : * of it in some cases.
576 : *
577 : * The result is always a freshly palloc'd datum.
578 : */
579 : static text *
580 701742 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
581 : {
582 701742 : int32 eml = pg_database_encoding_max_length();
583 701742 : int32 S = start; /* start position */
584 : int32 S1; /* adjusted start position */
585 : int32 L1; /* adjusted substring length */
586 : int32 E; /* end position */
587 :
588 : /*
589 : * SQL99 says S can be zero or negative (which we don't document), but we
590 : * still must fetch from the start of the string.
591 : * https://www.postgresql.org/message-id/170905442373.643.11536838320909376197%40wrigleys.postgresql.org
592 : */
593 701742 : S1 = Max(S, 1);
594 :
595 : /* life is easy if the encoding max length is 1 */
596 701742 : if (eml == 1)
597 : {
598 22 : if (length_not_specified) /* special case - get length to end of
599 : * string */
600 0 : L1 = -1;
601 22 : else if (length < 0)
602 : {
603 : /* SQL99 says to throw an error for E < S, i.e., negative length */
604 0 : ereport(ERROR,
605 : (errcode(ERRCODE_SUBSTRING_ERROR),
606 : errmsg("negative substring length not allowed")));
607 : L1 = -1; /* silence stupider compilers */
608 : }
609 22 : else if (pg_add_s32_overflow(S, length, &E))
610 : {
611 : /*
612 : * L could be large enough for S + L to overflow, in which case
613 : * the substring must run to end of string.
614 : */
615 0 : L1 = -1;
616 : }
617 : else
618 : {
619 : /*
620 : * A zero or negative value for the end position can happen if the
621 : * start was negative or one. SQL99 says to return a zero-length
622 : * string.
623 : */
624 22 : if (E < 1)
625 0 : return cstring_to_text("");
626 :
627 22 : L1 = E - S1;
628 : }
629 :
630 : /*
631 : * If the start position is past the end of the string, SQL99 says to
632 : * return a zero-length string -- DatumGetTextPSlice() will do that
633 : * for us. We need only convert S1 to zero-based starting position.
634 : */
635 22 : return DatumGetTextPSlice(str, S1 - 1, L1);
636 : }
637 701720 : else if (eml > 1)
638 : {
639 : /*
640 : * When encoding max length is > 1, we can't get LC without
641 : * detoasting, so we'll grab a conservatively large slice now and go
642 : * back later to do the right thing
643 : */
644 : int32 slice_start;
645 : int32 slice_size;
646 : int32 slice_strlen;
647 : text *slice;
648 : int32 E1;
649 : int32 i;
650 : char *p;
651 : char *s;
652 : text *ret;
653 :
654 : /*
655 : * We need to start at position zero because there is no way to know
656 : * in advance which byte offset corresponds to the supplied start
657 : * position.
658 : */
659 701720 : slice_start = 0;
660 :
661 701720 : if (length_not_specified) /* special case - get length to end of
662 : * string */
663 76 : slice_size = L1 = -1;
664 701644 : else if (length < 0)
665 : {
666 : /* SQL99 says to throw an error for E < S, i.e., negative length */
667 12 : ereport(ERROR,
668 : (errcode(ERRCODE_SUBSTRING_ERROR),
669 : errmsg("negative substring length not allowed")));
670 : slice_size = L1 = -1; /* silence stupider compilers */
671 : }
672 701632 : else if (pg_add_s32_overflow(S, length, &E))
673 : {
674 : /*
675 : * L could be large enough for S + L to overflow, in which case
676 : * the substring must run to end of string.
677 : */
678 6 : slice_size = L1 = -1;
679 : }
680 : else
681 : {
682 : /*
683 : * A zero or negative value for the end position can happen if the
684 : * start was negative or one. SQL99 says to return a zero-length
685 : * string.
686 : */
687 701626 : if (E < 1)
688 0 : return cstring_to_text("");
689 :
690 : /*
691 : * if E is past the end of the string, the tuple toaster will
692 : * truncate the length for us
693 : */
694 701626 : L1 = E - S1;
695 :
696 : /*
697 : * Total slice size in bytes can't be any longer than the start
698 : * position plus substring length times the encoding max length.
699 : * If that overflows, we can just use -1.
700 : */
701 701626 : if (pg_mul_s32_overflow(E, eml, &slice_size))
702 6 : slice_size = -1;
703 : }
704 :
705 : /*
706 : * If we're working with an untoasted source, no need to do an extra
707 : * copying step.
708 : */
709 1403350 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
710 701642 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
711 372 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
712 : else
713 701336 : slice = (text *) DatumGetPointer(str);
714 :
715 : /* see if we got back an empty string */
716 701708 : if (VARSIZE_ANY_EXHDR(slice) == 0)
717 : {
718 0 : if (slice != (text *) DatumGetPointer(str))
719 0 : pfree(slice);
720 0 : return cstring_to_text("");
721 : }
722 :
723 : /* Now we can get the actual length of the slice in MB characters */
724 701708 : slice_strlen = pg_mbstrlen_with_len(VARDATA_ANY(slice),
725 701708 : VARSIZE_ANY_EXHDR(slice));
726 :
727 : /*
728 : * Check that the start position wasn't > slice_strlen. If so, SQL99
729 : * says to return a zero-length string.
730 : */
731 701708 : if (S1 > slice_strlen)
732 : {
733 22 : if (slice != (text *) DatumGetPointer(str))
734 0 : pfree(slice);
735 22 : return cstring_to_text("");
736 : }
737 :
738 : /*
739 : * Adjust L1 and E1 now that we know the slice string length. Again
740 : * remember that S1 is one based, and slice_start is zero based.
741 : */
742 701686 : if (L1 > -1)
743 701626 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
744 : else
745 60 : E1 = slice_start + 1 + slice_strlen;
746 :
747 : /*
748 : * Find the start position in the slice; remember S1 is not zero based
749 : */
750 701686 : p = VARDATA_ANY(slice);
751 6716826 : for (i = 0; i < S1 - 1; i++)
752 6015140 : p += pg_mblen(p);
753 :
754 : /* hang onto a pointer to our start position */
755 701686 : s = p;
756 :
757 : /*
758 : * Count the actual bytes used by the substring of the requested
759 : * length.
760 : */
761 9961442 : for (i = S1; i < E1; i++)
762 9259756 : p += pg_mblen(p);
763 :
764 701686 : ret = (text *) palloc(VARHDRSZ + (p - s));
765 701686 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
766 701686 : memcpy(VARDATA(ret), s, (p - s));
767 :
768 701686 : if (slice != (text *) DatumGetPointer(str))
769 372 : pfree(slice);
770 :
771 701686 : return ret;
772 : }
773 : else
774 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
775 :
776 : /* not reached: suppress compiler warning */
777 : return NULL;
778 : }
779 :
780 : /*
781 : * textoverlay
782 : * Replace specified substring of first string with second
783 : *
784 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
785 : * This code is a direct implementation of what the standard says.
786 : */
787 : Datum
788 28 : textoverlay(PG_FUNCTION_ARGS)
789 : {
790 28 : text *t1 = PG_GETARG_TEXT_PP(0);
791 28 : text *t2 = PG_GETARG_TEXT_PP(1);
792 28 : int sp = PG_GETARG_INT32(2); /* substring start position */
793 28 : int sl = PG_GETARG_INT32(3); /* substring length */
794 :
795 28 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
796 : }
797 :
798 : Datum
799 12 : textoverlay_no_len(PG_FUNCTION_ARGS)
800 : {
801 12 : text *t1 = PG_GETARG_TEXT_PP(0);
802 12 : text *t2 = PG_GETARG_TEXT_PP(1);
803 12 : int sp = PG_GETARG_INT32(2); /* substring start position */
804 : int sl;
805 :
806 12 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
807 12 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
808 : }
809 :
810 : static text *
811 40 : text_overlay(text *t1, text *t2, int sp, int sl)
812 : {
813 : text *result;
814 : text *s1;
815 : text *s2;
816 : int sp_pl_sl;
817 :
818 : /*
819 : * Check for possible integer-overflow cases. For negative sp, throw a
820 : * "substring length" error because that's what should be expected
821 : * according to the spec's definition of OVERLAY().
822 : */
823 40 : if (sp <= 0)
824 0 : ereport(ERROR,
825 : (errcode(ERRCODE_SUBSTRING_ERROR),
826 : errmsg("negative substring length not allowed")));
827 40 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
828 0 : ereport(ERROR,
829 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
830 : errmsg("integer out of range")));
831 :
832 40 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
833 40 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
834 40 : result = text_catenate(s1, t2);
835 40 : result = text_catenate(result, s2);
836 :
837 40 : return result;
838 : }
839 :
840 : /*
841 : * textpos -
842 : * Return the position of the specified substring.
843 : * Implements the SQL POSITION() function.
844 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
845 : * - thomas 1997-07-27
846 : */
847 : Datum
848 130 : textpos(PG_FUNCTION_ARGS)
849 : {
850 130 : text *str = PG_GETARG_TEXT_PP(0);
851 130 : text *search_str = PG_GETARG_TEXT_PP(1);
852 :
853 130 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
854 : }
855 :
856 : /*
857 : * text_position -
858 : * Does the real work for textpos()
859 : *
860 : * Inputs:
861 : * t1 - string to be searched
862 : * t2 - pattern to match within t1
863 : * Result:
864 : * Character index of the first matched char, starting from 1,
865 : * or 0 if no match.
866 : *
867 : * This is broken out so it can be called directly by other string processing
868 : * functions.
869 : */
870 : static int
871 130 : text_position(text *t1, text *t2, Oid collid)
872 : {
873 : TextPositionState state;
874 : int result;
875 :
876 130 : check_collation_set(collid);
877 :
878 : /* Empty needle always matches at position 1 */
879 130 : if (VARSIZE_ANY_EXHDR(t2) < 1)
880 12 : return 1;
881 :
882 : /* Otherwise, can't match if haystack is shorter than needle */
883 118 : if (VARSIZE_ANY_EXHDR(t1) < VARSIZE_ANY_EXHDR(t2) &&
884 22 : pg_newlocale_from_collation(collid)->deterministic)
885 22 : return 0;
886 :
887 96 : text_position_setup(t1, t2, collid, &state);
888 : /* don't need greedy mode here */
889 96 : state.greedy = false;
890 :
891 96 : if (!text_position_next(&state))
892 24 : result = 0;
893 : else
894 72 : result = text_position_get_match_pos(&state);
895 96 : text_position_cleanup(&state);
896 96 : return result;
897 : }
898 :
899 :
900 : /*
901 : * text_position_setup, text_position_next, text_position_cleanup -
902 : * Component steps of text_position()
903 : *
904 : * These are broken out so that a string can be efficiently searched for
905 : * multiple occurrences of the same pattern. text_position_next may be
906 : * called multiple times, and it advances to the next match on each call.
907 : * text_position_get_match_ptr() and text_position_get_match_pos() return
908 : * a pointer or 1-based character position of the last match, respectively.
909 : *
910 : * The "state" variable is normally just a local variable in the caller.
911 : *
912 : * NOTE: text_position_next skips over the matched portion. For example,
913 : * searching for "xx" in "xxx" returns only one match, not two.
914 : */
915 :
916 : static void
917 1924 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
918 : {
919 1924 : int len1 = VARSIZE_ANY_EXHDR(t1);
920 1924 : int len2 = VARSIZE_ANY_EXHDR(t2);
921 :
922 1924 : check_collation_set(collid);
923 :
924 1924 : state->locale = pg_newlocale_from_collation(collid);
925 :
926 : /*
927 : * Most callers need greedy mode, but some might want to unset this to
928 : * optimize.
929 : */
930 1924 : state->greedy = true;
931 :
932 : Assert(len2 > 0);
933 :
934 : /*
935 : * Even with a multi-byte encoding, we perform the search using the raw
936 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
937 : * because in UTF-8 the byte sequence of one character cannot contain
938 : * another character. For other multi-byte encodings, we do the search
939 : * initially as a simple byte search, ignoring multibyte issues, but
940 : * verify afterwards that the match we found is at a character boundary,
941 : * and continue the search if it was a false match.
942 : */
943 1924 : if (pg_database_encoding_max_length() == 1)
944 108 : state->is_multibyte_char_in_char = false;
945 1816 : else if (GetDatabaseEncoding() == PG_UTF8)
946 1816 : state->is_multibyte_char_in_char = false;
947 : else
948 0 : state->is_multibyte_char_in_char = true;
949 :
950 1924 : state->str1 = VARDATA_ANY(t1);
951 1924 : state->str2 = VARDATA_ANY(t2);
952 1924 : state->len1 = len1;
953 1924 : state->len2 = len2;
954 1924 : state->last_match = NULL;
955 1924 : state->refpoint = state->str1;
956 1924 : state->refpos = 0;
957 :
958 : /*
959 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
960 : * notes we use the terminology that the "haystack" is the string to be
961 : * searched (t1) and the "needle" is the pattern being sought (t2).
962 : *
963 : * If the needle is empty or bigger than the haystack then there is no
964 : * point in wasting cycles initializing the table. We also choose not to
965 : * use B-M-H for needles of length 1, since the skip table can't possibly
966 : * save anything in that case.
967 : *
968 : * (With nondeterministic collations, the search is already
969 : * multibyte-aware, so we don't need this.)
970 : */
971 1924 : if (len1 >= len2 && len2 > 1 && state->locale->deterministic)
972 : {
973 1590 : int searchlength = len1 - len2;
974 : int skiptablemask;
975 : int last;
976 : int i;
977 1590 : const char *str2 = state->str2;
978 :
979 : /*
980 : * First we must determine how much of the skip table to use. The
981 : * declaration of TextPositionState allows up to 256 elements, but for
982 : * short search problems we don't really want to have to initialize so
983 : * many elements --- it would take too long in comparison to the
984 : * actual search time. So we choose a useful skip table size based on
985 : * the haystack length minus the needle length. The closer the needle
986 : * length is to the haystack length the less useful skipping becomes.
987 : *
988 : * Note: since we use bit-masking to select table elements, the skip
989 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
990 : */
991 1590 : if (searchlength < 16)
992 114 : skiptablemask = 3;
993 1476 : else if (searchlength < 64)
994 28 : skiptablemask = 7;
995 1448 : else if (searchlength < 128)
996 26 : skiptablemask = 15;
997 1422 : else if (searchlength < 512)
998 332 : skiptablemask = 31;
999 1090 : else if (searchlength < 2048)
1000 808 : skiptablemask = 63;
1001 282 : else if (searchlength < 4096)
1002 198 : skiptablemask = 127;
1003 : else
1004 84 : skiptablemask = 255;
1005 1590 : state->skiptablemask = skiptablemask;
1006 :
1007 : /*
1008 : * Initialize the skip table. We set all elements to the needle
1009 : * length, since this is the correct skip distance for any character
1010 : * not found in the needle.
1011 : */
1012 111870 : for (i = 0; i <= skiptablemask; i++)
1013 110280 : state->skiptable[i] = len2;
1014 :
1015 : /*
1016 : * Now examine the needle. For each character except the last one,
1017 : * set the corresponding table element to the appropriate skip
1018 : * distance. Note that when two characters share the same skip table
1019 : * entry, the one later in the needle must determine the skip
1020 : * distance.
1021 : */
1022 1590 : last = len2 - 1;
1023 :
1024 20326 : for (i = 0; i < last; i++)
1025 18736 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1026 : }
1027 1924 : }
1028 :
1029 : /*
1030 : * Advance to the next match, starting from the end of the previous match
1031 : * (or the beginning of the string, on first call). Returns true if a match
1032 : * is found.
1033 : *
1034 : * Note that this refuses to match an empty-string needle. Most callers
1035 : * will have handled that case specially and we'll never see it here.
1036 : */
1037 : static bool
1038 9766 : text_position_next(TextPositionState *state)
1039 : {
1040 9766 : int needle_len = state->len2;
1041 : char *start_ptr;
1042 : char *matchptr;
1043 :
1044 9766 : if (needle_len <= 0)
1045 0 : return false; /* result for empty pattern */
1046 :
1047 : /* Start from the point right after the previous match. */
1048 9766 : if (state->last_match)
1049 7830 : start_ptr = state->last_match + state->last_match_len;
1050 : else
1051 1936 : start_ptr = state->str1;
1052 :
1053 9766 : retry:
1054 9766 : matchptr = text_position_next_internal(start_ptr, state);
1055 :
1056 9766 : if (!matchptr)
1057 1840 : return false;
1058 :
1059 : /*
1060 : * Found a match for the byte sequence. If this is a multibyte encoding,
1061 : * where one character's byte sequence can appear inside a longer
1062 : * multi-byte character, we need to verify that the match was at a
1063 : * character boundary, not in the middle of a multi-byte character.
1064 : */
1065 7926 : if (state->is_multibyte_char_in_char && state->locale->deterministic)
1066 : {
1067 : /* Walk one character at a time, until we reach the match. */
1068 :
1069 : /* the search should never move backwards. */
1070 : Assert(state->refpoint <= matchptr);
1071 :
1072 0 : while (state->refpoint < matchptr)
1073 : {
1074 : /* step to next character. */
1075 0 : state->refpoint += pg_mblen(state->refpoint);
1076 0 : state->refpos++;
1077 :
1078 : /*
1079 : * If we stepped over the match's start position, then it was a
1080 : * false positive, where the byte sequence appeared in the middle
1081 : * of a multi-byte character. Skip it, and continue the search at
1082 : * the next character boundary.
1083 : */
1084 0 : if (state->refpoint > matchptr)
1085 : {
1086 0 : start_ptr = state->refpoint;
1087 0 : goto retry;
1088 : }
1089 : }
1090 : }
1091 :
1092 7926 : state->last_match = matchptr;
1093 7926 : state->last_match_len = state->last_match_len_tmp;
1094 7926 : return true;
1095 : }
1096 :
1097 : /*
1098 : * Subroutine of text_position_next(). This searches for the raw byte
1099 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1100 : * match starting at 'start_ptr', or NULL if no match is found.
1101 : */
1102 : static char *
1103 9766 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1104 : {
1105 9766 : int haystack_len = state->len1;
1106 9766 : int needle_len = state->len2;
1107 9766 : int skiptablemask = state->skiptablemask;
1108 9766 : const char *haystack = state->str1;
1109 9766 : const char *needle = state->str2;
1110 9766 : const char *haystack_end = &haystack[haystack_len];
1111 : const char *hptr;
1112 :
1113 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1114 : Assert(needle_len > 0);
1115 :
1116 9766 : state->last_match_len_tmp = needle_len;
1117 :
1118 9766 : if (!state->locale->deterministic)
1119 : {
1120 : /*
1121 : * With a nondeterministic collation, we have to use an unoptimized
1122 : * route. We walk through the haystack and see if at each position
1123 : * there is a substring of the remaining string that is equal to the
1124 : * needle under the given collation.
1125 : *
1126 : * Note, the found substring could have a different length than the
1127 : * needle. Callers that want to skip over the found string need to
1128 : * read the length of the found substring from last_match_len rather
1129 : * than just using the length of their needle.
1130 : *
1131 : * Most callers will require "greedy" semantics, meaning that we need
1132 : * to find the longest such substring, not the shortest. For callers
1133 : * that don't need greedy semantics, we can finish on the first match.
1134 : *
1135 : * This loop depends on the assumption that the needle is nonempty and
1136 : * any matching substring must also be nonempty. (Even if the
1137 : * collation would accept an empty match, returning one would send
1138 : * callers that search for successive matches into an infinite loop.)
1139 : */
1140 252 : const char *result_hptr = NULL;
1141 :
1142 252 : hptr = start_ptr;
1143 678 : while (hptr < haystack_end)
1144 : {
1145 : const char *test_end;
1146 :
1147 : /*
1148 : * First check the common case that there is a match in the
1149 : * haystack of exactly the length of the needle.
1150 : */
1151 564 : if (!state->greedy &&
1152 108 : haystack_end - hptr >= needle_len &&
1153 54 : pg_strncoll(hptr, needle_len, needle, needle_len, state->locale) == 0)
1154 12 : return (char *) hptr;
1155 :
1156 : /*
1157 : * Else check if any of the non-empty substrings starting at hptr
1158 : * compare equal to the needle.
1159 : */
1160 552 : test_end = hptr;
1161 : do
1162 : {
1163 2154 : test_end += pg_mblen(test_end);
1164 2154 : if (pg_strncoll(hptr, (test_end - hptr), needle, needle_len, state->locale) == 0)
1165 : {
1166 138 : state->last_match_len_tmp = (test_end - hptr);
1167 138 : result_hptr = hptr;
1168 138 : if (!state->greedy)
1169 0 : break;
1170 : }
1171 2154 : } while (test_end < haystack_end);
1172 :
1173 552 : if (result_hptr)
1174 126 : break;
1175 :
1176 426 : hptr += pg_mblen(hptr);
1177 : }
1178 :
1179 240 : return (char *) result_hptr;
1180 : }
1181 9514 : else if (needle_len == 1)
1182 : {
1183 : /* No point in using B-M-H for a one-character needle */
1184 760 : char nchar = *needle;
1185 :
1186 760 : hptr = start_ptr;
1187 5878 : while (hptr < haystack_end)
1188 : {
1189 5712 : if (*hptr == nchar)
1190 594 : return (char *) hptr;
1191 5118 : hptr++;
1192 : }
1193 : }
1194 : else
1195 : {
1196 8754 : const char *needle_last = &needle[needle_len - 1];
1197 :
1198 : /* Start at startpos plus the length of the needle */
1199 8754 : hptr = start_ptr + needle_len - 1;
1200 216696 : while (hptr < haystack_end)
1201 : {
1202 : /* Match the needle scanning *backward* */
1203 : const char *nptr;
1204 : const char *p;
1205 :
1206 215136 : nptr = needle_last;
1207 215136 : p = hptr;
1208 323074 : while (*nptr == *p)
1209 : {
1210 : /* Matched it all? If so, return 1-based position */
1211 115132 : if (nptr == needle)
1212 7194 : return (char *) p;
1213 107938 : nptr--, p--;
1214 : }
1215 :
1216 : /*
1217 : * No match, so use the haystack char at hptr to decide how far to
1218 : * advance. If the needle had any occurrence of that character
1219 : * (or more precisely, one sharing the same skiptable entry)
1220 : * before its last character, then we advance far enough to align
1221 : * the last such needle character with that haystack position.
1222 : * Otherwise we can advance by the whole needle length.
1223 : */
1224 207942 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1225 : }
1226 : }
1227 :
1228 1726 : return 0; /* not found */
1229 : }
1230 :
1231 : /*
1232 : * Return a pointer to the current match.
1233 : *
1234 : * The returned pointer points into the original haystack string.
1235 : */
1236 : static char *
1237 7824 : text_position_get_match_ptr(TextPositionState *state)
1238 : {
1239 7824 : return state->last_match;
1240 : }
1241 :
1242 : /*
1243 : * Return the offset of the current match.
1244 : *
1245 : * The offset is in characters, 1-based.
1246 : */
1247 : static int
1248 72 : text_position_get_match_pos(TextPositionState *state)
1249 : {
1250 : /* Convert the byte position to char position. */
1251 144 : state->refpos += pg_mbstrlen_with_len(state->refpoint,
1252 72 : state->last_match - state->refpoint);
1253 72 : state->refpoint = state->last_match;
1254 72 : return state->refpos + 1;
1255 : }
1256 :
1257 : /*
1258 : * Reset search state to the initial state installed by text_position_setup.
1259 : *
1260 : * The next call to text_position_next will search from the beginning
1261 : * of the string.
1262 : */
1263 : static void
1264 12 : text_position_reset(TextPositionState *state)
1265 : {
1266 12 : state->last_match = NULL;
1267 12 : state->refpoint = state->str1;
1268 12 : state->refpos = 0;
1269 12 : }
1270 :
1271 : static void
1272 1924 : text_position_cleanup(TextPositionState *state)
1273 : {
1274 : /* no cleanup needed */
1275 1924 : }
1276 :
1277 :
1278 : static void
1279 17579852 : check_collation_set(Oid collid)
1280 : {
1281 17579852 : if (!OidIsValid(collid))
1282 : {
1283 : /*
1284 : * This typically means that the parser could not resolve a conflict
1285 : * of implicit collations, so report it that way.
1286 : */
1287 30 : ereport(ERROR,
1288 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1289 : errmsg("could not determine which collation to use for string comparison"),
1290 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1291 : }
1292 17579822 : }
1293 :
1294 : /*
1295 : * varstr_cmp()
1296 : *
1297 : * Comparison function for text strings with given lengths, using the
1298 : * appropriate locale. Returns an integer less than, equal to, or greater than
1299 : * zero, indicating whether arg1 is less than, equal to, or greater than arg2.
1300 : *
1301 : * Note: many functions that depend on this are marked leakproof; therefore,
1302 : * avoid reporting the actual contents of the input when throwing errors.
1303 : * All errors herein should be things that can't happen except on corrupt
1304 : * data, anyway; otherwise we will have trouble with indexing strings that
1305 : * would cause them.
1306 : */
1307 : int
1308 10072758 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1309 : {
1310 : int result;
1311 : pg_locale_t mylocale;
1312 :
1313 10072758 : check_collation_set(collid);
1314 :
1315 10072740 : mylocale = pg_newlocale_from_collation(collid);
1316 :
1317 10072740 : if (mylocale->collate_is_c)
1318 : {
1319 4023854 : result = memcmp(arg1, arg2, Min(len1, len2));
1320 4023854 : if ((result == 0) && (len1 != len2))
1321 136724 : result = (len1 < len2) ? -1 : 1;
1322 : }
1323 : else
1324 : {
1325 : /*
1326 : * memcmp() can't tell us which of two unequal strings sorts first,
1327 : * but it's a cheap way to tell if they're equal. Testing shows that
1328 : * memcmp() followed by strcoll() is only trivially slower than
1329 : * strcoll() by itself, so we don't lose much if this doesn't work out
1330 : * very often, and if it does - for example, because there are many
1331 : * equal strings in the input - then we win big by avoiding expensive
1332 : * collation-aware comparisons.
1333 : */
1334 6048886 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1335 1567136 : return 0;
1336 :
1337 4481750 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1338 :
1339 : /* Break tie if necessary. */
1340 4481750 : if (result == 0 && mylocale->deterministic)
1341 : {
1342 0 : result = memcmp(arg1, arg2, Min(len1, len2));
1343 0 : if ((result == 0) && (len1 != len2))
1344 0 : result = (len1 < len2) ? -1 : 1;
1345 : }
1346 : }
1347 :
1348 8505604 : return result;
1349 : }
1350 :
1351 : /* text_cmp()
1352 : * Internal comparison function for text strings.
1353 : * Returns -1, 0 or 1
1354 : */
1355 : static int
1356 7939938 : text_cmp(text *arg1, text *arg2, Oid collid)
1357 : {
1358 : char *a1p,
1359 : *a2p;
1360 : int len1,
1361 : len2;
1362 :
1363 7939938 : a1p = VARDATA_ANY(arg1);
1364 7939938 : a2p = VARDATA_ANY(arg2);
1365 :
1366 7939938 : len1 = VARSIZE_ANY_EXHDR(arg1);
1367 7939938 : len2 = VARSIZE_ANY_EXHDR(arg2);
1368 :
1369 7939938 : return varstr_cmp(a1p, len1, a2p, len2, collid);
1370 : }
1371 :
1372 : /*
1373 : * Comparison functions for text strings.
1374 : *
1375 : * Note: btree indexes need these routines not to leak memory; therefore,
1376 : * be careful to free working copies of toasted datums. Most places don't
1377 : * need to be so careful.
1378 : */
1379 :
1380 : Datum
1381 6690656 : texteq(PG_FUNCTION_ARGS)
1382 : {
1383 6690656 : Oid collid = PG_GET_COLLATION();
1384 6690656 : pg_locale_t mylocale = 0;
1385 : bool result;
1386 :
1387 6690656 : check_collation_set(collid);
1388 :
1389 6690656 : mylocale = pg_newlocale_from_collation(collid);
1390 :
1391 6690656 : if (mylocale->deterministic)
1392 : {
1393 6686296 : Datum arg1 = PG_GETARG_DATUM(0);
1394 6686296 : Datum arg2 = PG_GETARG_DATUM(1);
1395 : Size len1,
1396 : len2;
1397 :
1398 : /*
1399 : * Since we only care about equality or not-equality, we can avoid all
1400 : * the expense of strcoll() here, and just do bitwise comparison. In
1401 : * fact, we don't even have to do a bitwise comparison if we can show
1402 : * the lengths of the strings are unequal; which might save us from
1403 : * having to detoast one or both values.
1404 : */
1405 6686296 : len1 = toast_raw_datum_size(arg1);
1406 6686296 : len2 = toast_raw_datum_size(arg2);
1407 6686296 : if (len1 != len2)
1408 3175116 : result = false;
1409 : else
1410 : {
1411 3511180 : text *targ1 = DatumGetTextPP(arg1);
1412 3511180 : text *targ2 = DatumGetTextPP(arg2);
1413 :
1414 3511180 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1415 : len1 - VARHDRSZ) == 0);
1416 :
1417 3511180 : PG_FREE_IF_COPY(targ1, 0);
1418 3511180 : PG_FREE_IF_COPY(targ2, 1);
1419 : }
1420 : }
1421 : else
1422 : {
1423 4360 : text *arg1 = PG_GETARG_TEXT_PP(0);
1424 4360 : text *arg2 = PG_GETARG_TEXT_PP(1);
1425 :
1426 4360 : result = (text_cmp(arg1, arg2, collid) == 0);
1427 :
1428 4360 : PG_FREE_IF_COPY(arg1, 0);
1429 4360 : PG_FREE_IF_COPY(arg2, 1);
1430 : }
1431 :
1432 6690656 : PG_RETURN_BOOL(result);
1433 : }
1434 :
1435 : Datum
1436 408594 : textne(PG_FUNCTION_ARGS)
1437 : {
1438 408594 : Oid collid = PG_GET_COLLATION();
1439 : pg_locale_t mylocale;
1440 : bool result;
1441 :
1442 408594 : check_collation_set(collid);
1443 :
1444 408594 : mylocale = pg_newlocale_from_collation(collid);
1445 :
1446 408594 : if (mylocale->deterministic)
1447 : {
1448 408570 : Datum arg1 = PG_GETARG_DATUM(0);
1449 408570 : Datum arg2 = PG_GETARG_DATUM(1);
1450 : Size len1,
1451 : len2;
1452 :
1453 : /* See comment in texteq() */
1454 408570 : len1 = toast_raw_datum_size(arg1);
1455 408570 : len2 = toast_raw_datum_size(arg2);
1456 408570 : if (len1 != len2)
1457 22362 : result = true;
1458 : else
1459 : {
1460 386208 : text *targ1 = DatumGetTextPP(arg1);
1461 386208 : text *targ2 = DatumGetTextPP(arg2);
1462 :
1463 386208 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1464 : len1 - VARHDRSZ) != 0);
1465 :
1466 386208 : PG_FREE_IF_COPY(targ1, 0);
1467 386208 : PG_FREE_IF_COPY(targ2, 1);
1468 : }
1469 : }
1470 : else
1471 : {
1472 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
1473 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
1474 :
1475 24 : result = (text_cmp(arg1, arg2, collid) != 0);
1476 :
1477 24 : PG_FREE_IF_COPY(arg1, 0);
1478 24 : PG_FREE_IF_COPY(arg2, 1);
1479 : }
1480 :
1481 408594 : PG_RETURN_BOOL(result);
1482 : }
1483 :
1484 : Datum
1485 209294 : text_lt(PG_FUNCTION_ARGS)
1486 : {
1487 209294 : text *arg1 = PG_GETARG_TEXT_PP(0);
1488 209294 : text *arg2 = PG_GETARG_TEXT_PP(1);
1489 : bool result;
1490 :
1491 209294 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1492 :
1493 209276 : PG_FREE_IF_COPY(arg1, 0);
1494 209276 : PG_FREE_IF_COPY(arg2, 1);
1495 :
1496 209276 : PG_RETURN_BOOL(result);
1497 : }
1498 :
1499 : Datum
1500 317724 : text_le(PG_FUNCTION_ARGS)
1501 : {
1502 317724 : text *arg1 = PG_GETARG_TEXT_PP(0);
1503 317724 : text *arg2 = PG_GETARG_TEXT_PP(1);
1504 : bool result;
1505 :
1506 317724 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1507 :
1508 317724 : PG_FREE_IF_COPY(arg1, 0);
1509 317724 : PG_FREE_IF_COPY(arg2, 1);
1510 :
1511 317724 : PG_RETURN_BOOL(result);
1512 : }
1513 :
1514 : Datum
1515 196076 : text_gt(PG_FUNCTION_ARGS)
1516 : {
1517 196076 : text *arg1 = PG_GETARG_TEXT_PP(0);
1518 196076 : text *arg2 = PG_GETARG_TEXT_PP(1);
1519 : bool result;
1520 :
1521 196076 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1522 :
1523 196076 : PG_FREE_IF_COPY(arg1, 0);
1524 196076 : PG_FREE_IF_COPY(arg2, 1);
1525 :
1526 196076 : PG_RETURN_BOOL(result);
1527 : }
1528 :
1529 : Datum
1530 175720 : text_ge(PG_FUNCTION_ARGS)
1531 : {
1532 175720 : text *arg1 = PG_GETARG_TEXT_PP(0);
1533 175720 : text *arg2 = PG_GETARG_TEXT_PP(1);
1534 : bool result;
1535 :
1536 175720 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1537 :
1538 175720 : PG_FREE_IF_COPY(arg1, 0);
1539 175720 : PG_FREE_IF_COPY(arg2, 1);
1540 :
1541 175720 : PG_RETURN_BOOL(result);
1542 : }
1543 :
1544 : Datum
1545 37914 : text_starts_with(PG_FUNCTION_ARGS)
1546 : {
1547 37914 : Datum arg1 = PG_GETARG_DATUM(0);
1548 37914 : Datum arg2 = PG_GETARG_DATUM(1);
1549 37914 : Oid collid = PG_GET_COLLATION();
1550 : pg_locale_t mylocale;
1551 : bool result;
1552 : Size len1,
1553 : len2;
1554 :
1555 37914 : check_collation_set(collid);
1556 :
1557 37914 : mylocale = pg_newlocale_from_collation(collid);
1558 :
1559 37914 : if (!mylocale->deterministic)
1560 0 : ereport(ERROR,
1561 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1562 : errmsg("nondeterministic collations are not supported for substring searches")));
1563 :
1564 37914 : len1 = toast_raw_datum_size(arg1);
1565 37914 : len2 = toast_raw_datum_size(arg2);
1566 37914 : if (len2 > len1)
1567 0 : result = false;
1568 : else
1569 : {
1570 37914 : text *targ1 = text_substring(arg1, 1, len2, false);
1571 37914 : text *targ2 = DatumGetTextPP(arg2);
1572 :
1573 37914 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1574 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1575 :
1576 37914 : PG_FREE_IF_COPY(targ1, 0);
1577 37914 : PG_FREE_IF_COPY(targ2, 1);
1578 : }
1579 :
1580 37914 : PG_RETURN_BOOL(result);
1581 : }
1582 :
1583 : Datum
1584 6721104 : bttextcmp(PG_FUNCTION_ARGS)
1585 : {
1586 6721104 : text *arg1 = PG_GETARG_TEXT_PP(0);
1587 6721104 : text *arg2 = PG_GETARG_TEXT_PP(1);
1588 : int32 result;
1589 :
1590 6721104 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1591 :
1592 6721104 : PG_FREE_IF_COPY(arg1, 0);
1593 6721104 : PG_FREE_IF_COPY(arg2, 1);
1594 :
1595 6721104 : PG_RETURN_INT32(result);
1596 : }
1597 :
1598 : Datum
1599 88992 : bttextsortsupport(PG_FUNCTION_ARGS)
1600 : {
1601 88992 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1602 88992 : Oid collid = ssup->ssup_collation;
1603 : MemoryContext oldcontext;
1604 :
1605 88992 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1606 :
1607 : /* Use generic string SortSupport */
1608 88992 : varstr_sortsupport(ssup, TEXTOID, collid);
1609 :
1610 88980 : MemoryContextSwitchTo(oldcontext);
1611 :
1612 88980 : PG_RETURN_VOID();
1613 : }
1614 :
1615 : /*
1616 : * Generic sortsupport interface for character type's operator classes.
1617 : * Includes locale support, and support for BpChar semantics (i.e. removing
1618 : * trailing spaces before comparison).
1619 : *
1620 : * Relies on the assumption that text, VarChar, and BpChar all have the
1621 : * same representation.
1622 : */
1623 : void
1624 140540 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1625 : {
1626 140540 : bool abbreviate = ssup->abbreviate;
1627 140540 : bool collate_c = false;
1628 : VarStringSortSupport *sss;
1629 : pg_locale_t locale;
1630 :
1631 140540 : check_collation_set(collid);
1632 :
1633 140528 : locale = pg_newlocale_from_collation(collid);
1634 :
1635 : /*
1636 : * If possible, set ssup->comparator to a function which can be used to
1637 : * directly compare two datums. If we can do this, we'll avoid the
1638 : * overhead of a trip through the fmgr layer for every comparison, which
1639 : * can be substantial.
1640 : *
1641 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1642 : * which uses strcoll() to perform comparisons. We use that for the
1643 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1644 : * LC_COLLATE = C, we can make things quite a bit faster with
1645 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1646 : * memcmp() rather than strcoll().
1647 : */
1648 140528 : if (locale->collate_is_c)
1649 : {
1650 93862 : if (typid == BPCHAROID)
1651 308 : ssup->comparator = bpcharfastcmp_c;
1652 93554 : else if (typid == NAMEOID)
1653 : {
1654 50498 : ssup->comparator = namefastcmp_c;
1655 : /* Not supporting abbreviation with type NAME, for now */
1656 50498 : abbreviate = false;
1657 : }
1658 : else
1659 43056 : ssup->comparator = varstrfastcmp_c;
1660 :
1661 93862 : collate_c = true;
1662 : }
1663 : else
1664 : {
1665 : /*
1666 : * We use varlenafastcmp_locale except for type NAME.
1667 : */
1668 46666 : if (typid == NAMEOID)
1669 : {
1670 0 : ssup->comparator = namefastcmp_locale;
1671 : /* Not supporting abbreviation with type NAME, for now */
1672 0 : abbreviate = false;
1673 : }
1674 : else
1675 46666 : ssup->comparator = varlenafastcmp_locale;
1676 :
1677 : /*
1678 : * Unfortunately, it seems that abbreviation for non-C collations is
1679 : * broken on many common platforms; see pg_strxfrm_enabled().
1680 : *
1681 : * Even apart from the risk of broken locales, it's possible that
1682 : * there are platforms where the use of abbreviated keys should be
1683 : * disabled at compile time. For example, macOS's strxfrm()
1684 : * implementation is known to not effectively concentrate a
1685 : * significant amount of entropy from the original string in earlier
1686 : * transformed blobs. It's possible that other supported platforms
1687 : * are similarly encumbered. So, if we ever get past disabling this
1688 : * categorically, we may still want or need to disable it for
1689 : * particular platforms.
1690 : */
1691 46666 : if (!pg_strxfrm_enabled(locale))
1692 45870 : abbreviate = false;
1693 : }
1694 :
1695 : /*
1696 : * If we're using abbreviated keys, or if we're using a locale-aware
1697 : * comparison, we need to initialize a VarStringSortSupport object. Both
1698 : * cases will make use of the temporary buffers we initialize here for
1699 : * scratch space (and to detect requirement for BpChar semantics from
1700 : * caller), and the abbreviation case requires additional state.
1701 : */
1702 140528 : if (abbreviate || !collate_c)
1703 : {
1704 70732 : sss = palloc_object(VarStringSortSupport);
1705 70732 : sss->buf1 = palloc(TEXTBUFLEN);
1706 70732 : sss->buflen1 = TEXTBUFLEN;
1707 70732 : sss->buf2 = palloc(TEXTBUFLEN);
1708 70732 : sss->buflen2 = TEXTBUFLEN;
1709 : /* Start with invalid values */
1710 70732 : sss->last_len1 = -1;
1711 70732 : sss->last_len2 = -1;
1712 : /* Initialize */
1713 70732 : sss->last_returned = 0;
1714 70732 : if (collate_c)
1715 24066 : sss->locale = NULL;
1716 : else
1717 46666 : sss->locale = locale;
1718 :
1719 : /*
1720 : * To avoid somehow confusing a strxfrm() blob and an original string,
1721 : * constantly keep track of the variety of data that buf1 and buf2
1722 : * currently contain.
1723 : *
1724 : * Comparisons may be interleaved with conversion calls. Frequently,
1725 : * conversions and comparisons are batched into two distinct phases,
1726 : * but the correctness of caching cannot hinge upon this. For
1727 : * comparison caching, buffer state is only trusted if cache_blob is
1728 : * found set to false, whereas strxfrm() caching only trusts the state
1729 : * when cache_blob is found set to true.
1730 : *
1731 : * Arbitrarily initialize cache_blob to true.
1732 : */
1733 70732 : sss->cache_blob = true;
1734 70732 : sss->collate_c = collate_c;
1735 70732 : sss->typid = typid;
1736 70732 : ssup->ssup_extra = sss;
1737 :
1738 : /*
1739 : * If possible, plan to use the abbreviated keys optimization. The
1740 : * core code may switch back to authoritative comparator should
1741 : * abbreviation be aborted.
1742 : */
1743 70732 : if (abbreviate)
1744 : {
1745 24664 : sss->prop_card = 0.20;
1746 24664 : initHyperLogLog(&sss->abbr_card, 10);
1747 24664 : initHyperLogLog(&sss->full_card, 10);
1748 24664 : ssup->abbrev_full_comparator = ssup->comparator;
1749 24664 : ssup->comparator = ssup_datum_unsigned_cmp;
1750 24664 : ssup->abbrev_converter = varstr_abbrev_convert;
1751 24664 : ssup->abbrev_abort = varstr_abbrev_abort;
1752 : }
1753 : }
1754 140528 : }
1755 :
1756 : /*
1757 : * sortsupport comparison func (for C locale case)
1758 : */
1759 : static int
1760 46091490 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
1761 : {
1762 46091490 : VarString *arg1 = DatumGetVarStringPP(x);
1763 46091490 : VarString *arg2 = DatumGetVarStringPP(y);
1764 : char *a1p,
1765 : *a2p;
1766 : int len1,
1767 : len2,
1768 : result;
1769 :
1770 46091490 : a1p = VARDATA_ANY(arg1);
1771 46091490 : a2p = VARDATA_ANY(arg2);
1772 :
1773 46091490 : len1 = VARSIZE_ANY_EXHDR(arg1);
1774 46091490 : len2 = VARSIZE_ANY_EXHDR(arg2);
1775 :
1776 46091490 : result = memcmp(a1p, a2p, Min(len1, len2));
1777 46091490 : if ((result == 0) && (len1 != len2))
1778 1218792 : result = (len1 < len2) ? -1 : 1;
1779 :
1780 : /* We can't afford to leak memory here. */
1781 46091490 : if (PointerGetDatum(arg1) != x)
1782 0 : pfree(arg1);
1783 46091490 : if (PointerGetDatum(arg2) != y)
1784 0 : pfree(arg2);
1785 :
1786 46091490 : return result;
1787 : }
1788 :
1789 : /*
1790 : * sortsupport comparison func (for BpChar C locale case)
1791 : *
1792 : * BpChar outsources its sortsupport to this module. Specialization for the
1793 : * varstr_sortsupport BpChar case, modeled on
1794 : * internal_bpchar_pattern_compare().
1795 : */
1796 : static int
1797 62412 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
1798 : {
1799 62412 : BpChar *arg1 = DatumGetBpCharPP(x);
1800 62412 : BpChar *arg2 = DatumGetBpCharPP(y);
1801 : char *a1p,
1802 : *a2p;
1803 : int len1,
1804 : len2,
1805 : result;
1806 :
1807 62412 : a1p = VARDATA_ANY(arg1);
1808 62412 : a2p = VARDATA_ANY(arg2);
1809 :
1810 62412 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
1811 62412 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
1812 :
1813 62412 : result = memcmp(a1p, a2p, Min(len1, len2));
1814 62412 : if ((result == 0) && (len1 != len2))
1815 4 : result = (len1 < len2) ? -1 : 1;
1816 :
1817 : /* We can't afford to leak memory here. */
1818 62412 : if (PointerGetDatum(arg1) != x)
1819 0 : pfree(arg1);
1820 62412 : if (PointerGetDatum(arg2) != y)
1821 0 : pfree(arg2);
1822 :
1823 62412 : return result;
1824 : }
1825 :
1826 : /*
1827 : * sortsupport comparison func (for NAME C locale case)
1828 : */
1829 : static int
1830 43803098 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
1831 : {
1832 43803098 : Name arg1 = DatumGetName(x);
1833 43803098 : Name arg2 = DatumGetName(y);
1834 :
1835 43803098 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
1836 : }
1837 :
1838 : /*
1839 : * sortsupport comparison func (for locale case with all varlena types)
1840 : */
1841 : static int
1842 37752180 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
1843 : {
1844 37752180 : VarString *arg1 = DatumGetVarStringPP(x);
1845 37752180 : VarString *arg2 = DatumGetVarStringPP(y);
1846 : char *a1p,
1847 : *a2p;
1848 : int len1,
1849 : len2,
1850 : result;
1851 :
1852 37752180 : a1p = VARDATA_ANY(arg1);
1853 37752180 : a2p = VARDATA_ANY(arg2);
1854 :
1855 37752180 : len1 = VARSIZE_ANY_EXHDR(arg1);
1856 37752180 : len2 = VARSIZE_ANY_EXHDR(arg2);
1857 :
1858 37752180 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
1859 :
1860 : /* We can't afford to leak memory here. */
1861 37752180 : if (PointerGetDatum(arg1) != x)
1862 0 : pfree(arg1);
1863 37752180 : if (PointerGetDatum(arg2) != y)
1864 0 : pfree(arg2);
1865 :
1866 37752180 : return result;
1867 : }
1868 :
1869 : /*
1870 : * sortsupport comparison func (for locale case with NAME type)
1871 : */
1872 : static int
1873 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
1874 : {
1875 0 : Name arg1 = DatumGetName(x);
1876 0 : Name arg2 = DatumGetName(y);
1877 :
1878 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
1879 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
1880 : ssup);
1881 : }
1882 :
1883 : /*
1884 : * sortsupport comparison func for locale cases
1885 : */
1886 : static int
1887 37752180 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
1888 : {
1889 37752180 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1890 : int result;
1891 : bool arg1_match;
1892 :
1893 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
1894 37752180 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
1895 : {
1896 : /*
1897 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
1898 : * last_len2. Existing contents of buffers might still be used by
1899 : * next call.
1900 : *
1901 : * It's fine to allow the comparison of BpChar padding bytes here,
1902 : * even though that implies that the memcmp() will usually be
1903 : * performed for BpChar callers (though multibyte characters could
1904 : * still prevent that from occurring). The memcmp() is still very
1905 : * cheap, and BpChar's funny semantics have us remove trailing spaces
1906 : * (not limited to padding), so we need make no distinction between
1907 : * padding space characters and "real" space characters.
1908 : */
1909 9404318 : return 0;
1910 : }
1911 :
1912 28347862 : if (sss->typid == BPCHAROID)
1913 : {
1914 : /* Get true number of bytes, ignoring trailing spaces */
1915 39184 : len1 = bpchartruelen(a1p, len1);
1916 39184 : len2 = bpchartruelen(a2p, len2);
1917 : }
1918 :
1919 28347862 : if (len1 >= sss->buflen1)
1920 : {
1921 10 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
1922 10 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
1923 : }
1924 28347862 : if (len2 >= sss->buflen2)
1925 : {
1926 6 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
1927 6 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
1928 : }
1929 :
1930 : /*
1931 : * We're likely to be asked to compare the same strings repeatedly, and
1932 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
1933 : * comparisons, even though in general there is no reason to think that
1934 : * that will work out (every string datum may be unique). Caching does
1935 : * not slow things down measurably when it doesn't work out, and can speed
1936 : * things up by rather a lot when it does. In part, this is because the
1937 : * memcmp() compares data from cachelines that are needed in L1 cache even
1938 : * when the last comparison's result cannot be reused.
1939 : */
1940 28347862 : arg1_match = true;
1941 28347862 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
1942 : {
1943 26232916 : arg1_match = false;
1944 26232916 : memcpy(sss->buf1, a1p, len1);
1945 26232916 : sss->buf1[len1] = '\0';
1946 26232916 : sss->last_len1 = len1;
1947 : }
1948 :
1949 : /*
1950 : * If we're comparing the same two strings as last time, we can return the
1951 : * same answer without calling strcoll() again. This is more likely than
1952 : * it seems (at least with moderate to low cardinality sets), because
1953 : * quicksort compares the same pivot against many values.
1954 : */
1955 28347862 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
1956 : {
1957 4363532 : memcpy(sss->buf2, a2p, len2);
1958 4363532 : sss->buf2[len2] = '\0';
1959 4363532 : sss->last_len2 = len2;
1960 : }
1961 23984330 : else if (arg1_match && !sss->cache_blob)
1962 : {
1963 : /* Use result cached following last actual strcoll() call */
1964 1660256 : return sss->last_returned;
1965 : }
1966 :
1967 26687606 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
1968 :
1969 : /* Break tie if necessary. */
1970 26687606 : if (result == 0 && sss->locale->deterministic)
1971 0 : result = strcmp(sss->buf1, sss->buf2);
1972 :
1973 : /* Cache result, perhaps saving an expensive strcoll() call next time */
1974 26687606 : sss->cache_blob = false;
1975 26687606 : sss->last_returned = result;
1976 26687606 : return result;
1977 : }
1978 :
1979 : /*
1980 : * Conversion routine for sortsupport. Converts original to abbreviated key
1981 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
1982 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
1983 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
1984 : * locale is used just memcpy() from original instead.
1985 : */
1986 : static Datum
1987 844524 : varstr_abbrev_convert(Datum original, SortSupport ssup)
1988 : {
1989 844524 : const size_t max_prefix_bytes = sizeof(Datum);
1990 844524 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1991 844524 : VarString *authoritative = DatumGetVarStringPP(original);
1992 844524 : char *authoritative_data = VARDATA_ANY(authoritative);
1993 :
1994 : /* working state */
1995 : Datum res;
1996 : char *pres;
1997 : int len;
1998 : uint32 hash;
1999 :
2000 844524 : pres = (char *) &res;
2001 : /* memset(), so any non-overwritten bytes are NUL */
2002 844524 : memset(pres, 0, max_prefix_bytes);
2003 844524 : len = VARSIZE_ANY_EXHDR(authoritative);
2004 :
2005 : /* Get number of bytes, ignoring trailing spaces */
2006 844524 : if (sss->typid == BPCHAROID)
2007 1010 : len = bpchartruelen(authoritative_data, len);
2008 :
2009 : /*
2010 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2011 : * abbreviate keys. The full comparator for the C locale is also
2012 : * memcmp(). This should be faster than strxfrm().
2013 : */
2014 844524 : if (sss->collate_c)
2015 842688 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
2016 : else
2017 : {
2018 : Size bsize;
2019 :
2020 : /*
2021 : * We're not using the C collation, so fall back on strxfrm or ICU
2022 : * analogs.
2023 : */
2024 :
2025 : /* By convention, we use buffer 1 to store and NUL-terminate */
2026 1836 : if (len >= sss->buflen1)
2027 : {
2028 0 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2029 0 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2030 : }
2031 :
2032 : /* Might be able to reuse strxfrm() blob from last call */
2033 1836 : if (sss->last_len1 == len && sss->cache_blob &&
2034 918 : memcmp(sss->buf1, authoritative_data, len) == 0)
2035 : {
2036 168 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, sss->last_len2));
2037 : /* No change affecting cardinality, so no hashing required */
2038 168 : goto done;
2039 : }
2040 :
2041 1668 : memcpy(sss->buf1, authoritative_data, len);
2042 :
2043 : /*
2044 : * pg_strxfrm() and pg_strxfrm_prefix expect NUL-terminated strings.
2045 : */
2046 1668 : sss->buf1[len] = '\0';
2047 1668 : sss->last_len1 = len;
2048 :
2049 1668 : if (pg_strxfrm_prefix_enabled(sss->locale))
2050 : {
2051 1668 : if (sss->buflen2 < max_prefix_bytes)
2052 : {
2053 0 : sss->buflen2 = Max(max_prefix_bytes,
2054 : Min(sss->buflen2 * 2, MaxAllocSize));
2055 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2056 : }
2057 :
2058 1668 : bsize = pg_strxfrm_prefix(sss->buf2, sss->buf1,
2059 : max_prefix_bytes, sss->locale);
2060 1668 : sss->last_len2 = bsize;
2061 : }
2062 : else
2063 : {
2064 : /*
2065 : * Loop: Call pg_strxfrm(), possibly enlarge buffer, and try
2066 : * again. The pg_strxfrm() function leaves the result buffer
2067 : * content undefined if the result did not fit, so we need to
2068 : * retry until everything fits, even though we only need the first
2069 : * few bytes in the end.
2070 : */
2071 : for (;;)
2072 : {
2073 0 : bsize = pg_strxfrm(sss->buf2, sss->buf1, sss->buflen2,
2074 : sss->locale);
2075 :
2076 0 : sss->last_len2 = bsize;
2077 0 : if (bsize < sss->buflen2)
2078 0 : break;
2079 :
2080 : /*
2081 : * Grow buffer and retry.
2082 : */
2083 0 : sss->buflen2 = Max(bsize + 1,
2084 : Min(sss->buflen2 * 2, MaxAllocSize));
2085 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2086 : }
2087 : }
2088 :
2089 : /*
2090 : * Every Datum byte is always compared. This is safe because the
2091 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2092 : * misinterpreting any NUL bytes not intended to be interpreted as
2093 : * logically representing termination.
2094 : */
2095 1668 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, bsize));
2096 : }
2097 :
2098 : /*
2099 : * Maintain approximate cardinality of both abbreviated keys and original,
2100 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2101 : * the worst case, where we do many string transformations for no saving
2102 : * in full strcoll()-based comparisons. These statistics are used by
2103 : * varstr_abbrev_abort().
2104 : *
2105 : * First, Hash key proper, or a significant fraction of it. Mix in length
2106 : * in order to compensate for cases where differences are past
2107 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2108 : */
2109 844356 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2110 : Min(len, PG_CACHE_LINE_SIZE)));
2111 :
2112 844356 : if (len > PG_CACHE_LINE_SIZE)
2113 192 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2114 :
2115 844356 : addHyperLogLog(&sss->full_card, hash);
2116 :
2117 : /* Hash abbreviated key */
2118 : {
2119 : uint32 tmp;
2120 :
2121 844356 : tmp = DatumGetUInt32(res) ^ (uint32) (DatumGetUInt64(res) >> 32);
2122 844356 : hash = DatumGetUInt32(hash_uint32(tmp));
2123 : }
2124 :
2125 844356 : addHyperLogLog(&sss->abbr_card, hash);
2126 :
2127 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2128 844356 : sss->cache_blob = true;
2129 844524 : done:
2130 :
2131 : /*
2132 : * Byteswap on little-endian machines.
2133 : *
2134 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2135 : * 3-way comparator) works correctly on all platforms. If we didn't do
2136 : * this, the comparator would have to call memcmp() with a pair of
2137 : * pointers to the first byte of each abbreviated key, which is slower.
2138 : */
2139 844524 : res = DatumBigEndianToNative(res);
2140 :
2141 : /* Don't leak memory here */
2142 844524 : if (PointerGetDatum(authoritative) != original)
2143 2 : pfree(authoritative);
2144 :
2145 844524 : return res;
2146 : }
2147 :
2148 : /*
2149 : * Callback for estimating effectiveness of abbreviated key optimization, using
2150 : * heuristic rules. Returns value indicating if the abbreviation optimization
2151 : * should be aborted, based on its projected effectiveness.
2152 : */
2153 : static bool
2154 2386 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2155 : {
2156 2386 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2157 : double abbrev_distinct,
2158 : key_distinct;
2159 :
2160 : Assert(ssup->abbreviate);
2161 :
2162 : /* Have a little patience */
2163 2386 : if (memtupcount < 100)
2164 1388 : return false;
2165 :
2166 998 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2167 998 : key_distinct = estimateHyperLogLog(&sss->full_card);
2168 :
2169 : /*
2170 : * Clamp cardinality estimates to at least one distinct value. While
2171 : * NULLs are generally disregarded, if only NULL values were seen so far,
2172 : * that might misrepresent costs if we failed to clamp.
2173 : */
2174 998 : if (abbrev_distinct < 1.0)
2175 0 : abbrev_distinct = 1.0;
2176 :
2177 998 : if (key_distinct < 1.0)
2178 0 : key_distinct = 1.0;
2179 :
2180 : /*
2181 : * In the worst case all abbreviated keys are identical, while at the same
2182 : * time there are differences within full key strings not captured in
2183 : * abbreviations.
2184 : */
2185 998 : if (trace_sort)
2186 : {
2187 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2188 :
2189 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2190 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2191 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2192 : sss->prop_card);
2193 : }
2194 :
2195 : /*
2196 : * If the number of distinct abbreviated keys approximately matches the
2197 : * number of distinct authoritative original keys, that's reason enough to
2198 : * proceed. We can win even with a very low cardinality set if most
2199 : * tie-breakers only memcmp(). This is by far the most important
2200 : * consideration.
2201 : *
2202 : * While comparisons that are resolved at the abbreviated key level are
2203 : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2204 : * those two outcomes are so much cheaper than a full strcoll() once
2205 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2206 : * cardinality against the overall size of the set in order to more
2207 : * accurately model costs. Assume that an abbreviated comparison, and an
2208 : * abbreviated comparison with a cheap memcmp()-based authoritative
2209 : * resolution are equivalent.
2210 : */
2211 998 : if (abbrev_distinct > key_distinct * sss->prop_card)
2212 : {
2213 : /*
2214 : * When we have exceeded 10,000 tuples, decay required cardinality
2215 : * aggressively for next call.
2216 : *
2217 : * This is useful because the number of comparisons required on
2218 : * average increases at a linearithmic rate, and at roughly 10,000
2219 : * tuples that factor will start to dominate over the linear costs of
2220 : * string transformation (this is a conservative estimate). The decay
2221 : * rate is chosen to be a little less aggressive than halving -- which
2222 : * (since we're called at points at which memtupcount has doubled)
2223 : * would never see the cost model actually abort past the first call
2224 : * following a decay. This decay rate is mostly a precaution against
2225 : * a sudden, violent swing in how well abbreviated cardinality tracks
2226 : * full key cardinality. The decay also serves to prevent a marginal
2227 : * case from being aborted too late, when too much has already been
2228 : * invested in string transformation.
2229 : *
2230 : * It's possible for sets of several million distinct strings with
2231 : * mere tens of thousands of distinct abbreviated keys to still
2232 : * benefit very significantly. This will generally occur provided
2233 : * each abbreviated key is a proxy for a roughly uniform number of the
2234 : * set's full keys. If it isn't so, we hope to catch that early and
2235 : * abort. If it isn't caught early, by the time the problem is
2236 : * apparent it's probably not worth aborting.
2237 : */
2238 998 : if (memtupcount > 10000)
2239 4 : sss->prop_card *= 0.65;
2240 :
2241 998 : return false;
2242 : }
2243 :
2244 : /*
2245 : * Abort abbreviation strategy.
2246 : *
2247 : * The worst case, where all abbreviated keys are identical while all
2248 : * original strings differ will typically only see a regression of about
2249 : * 10% in execution time for small to medium sized lists of strings.
2250 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2251 : * often expect very large improvements, particularly with sets of strings
2252 : * of moderately high to high abbreviated cardinality. There is little to
2253 : * lose but much to gain, which our strategy reflects.
2254 : */
2255 0 : if (trace_sort)
2256 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2257 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2258 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2259 :
2260 0 : return true;
2261 : }
2262 :
2263 : /*
2264 : * Generic equalimage support function for character type's operator classes.
2265 : * Disables the use of deduplication with nondeterministic collations.
2266 : */
2267 : Datum
2268 9274 : btvarstrequalimage(PG_FUNCTION_ARGS)
2269 : {
2270 : #ifdef NOT_USED
2271 : Oid opcintype = PG_GETARG_OID(0);
2272 : #endif
2273 9274 : Oid collid = PG_GET_COLLATION();
2274 : pg_locale_t locale;
2275 :
2276 9274 : check_collation_set(collid);
2277 :
2278 9274 : locale = pg_newlocale_from_collation(collid);
2279 :
2280 9274 : PG_RETURN_BOOL(locale->deterministic);
2281 : }
2282 :
2283 : Datum
2284 229560 : text_larger(PG_FUNCTION_ARGS)
2285 : {
2286 229560 : text *arg1 = PG_GETARG_TEXT_PP(0);
2287 229560 : text *arg2 = PG_GETARG_TEXT_PP(1);
2288 : text *result;
2289 :
2290 229560 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2291 :
2292 229560 : PG_RETURN_TEXT_P(result);
2293 : }
2294 :
2295 : Datum
2296 86076 : text_smaller(PG_FUNCTION_ARGS)
2297 : {
2298 86076 : text *arg1 = PG_GETARG_TEXT_PP(0);
2299 86076 : text *arg2 = PG_GETARG_TEXT_PP(1);
2300 : text *result;
2301 :
2302 86076 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2303 :
2304 86076 : PG_RETURN_TEXT_P(result);
2305 : }
2306 :
2307 :
2308 : /*
2309 : * Cross-type comparison functions for types text and name.
2310 : */
2311 :
2312 : Datum
2313 209950 : nameeqtext(PG_FUNCTION_ARGS)
2314 : {
2315 209950 : Name arg1 = PG_GETARG_NAME(0);
2316 209950 : text *arg2 = PG_GETARG_TEXT_PP(1);
2317 209950 : size_t len1 = strlen(NameStr(*arg1));
2318 209950 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2319 209950 : Oid collid = PG_GET_COLLATION();
2320 : bool result;
2321 :
2322 209950 : check_collation_set(collid);
2323 :
2324 209950 : if (collid == C_COLLATION_OID)
2325 254238 : result = (len1 == len2 &&
2326 123692 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2327 : else
2328 79404 : result = (varstr_cmp(NameStr(*arg1), len1,
2329 79404 : VARDATA_ANY(arg2), len2,
2330 : collid) == 0);
2331 :
2332 209950 : PG_FREE_IF_COPY(arg2, 1);
2333 :
2334 209950 : PG_RETURN_BOOL(result);
2335 : }
2336 :
2337 : Datum
2338 8076 : texteqname(PG_FUNCTION_ARGS)
2339 : {
2340 8076 : text *arg1 = PG_GETARG_TEXT_PP(0);
2341 8076 : Name arg2 = PG_GETARG_NAME(1);
2342 8076 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2343 8076 : size_t len2 = strlen(NameStr(*arg2));
2344 8076 : Oid collid = PG_GET_COLLATION();
2345 : bool result;
2346 :
2347 8076 : check_collation_set(collid);
2348 :
2349 8076 : if (collid == C_COLLATION_OID)
2350 568 : result = (len1 == len2 &&
2351 182 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2352 : else
2353 7690 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2354 7690 : NameStr(*arg2), len2,
2355 : collid) == 0);
2356 :
2357 8076 : PG_FREE_IF_COPY(arg1, 0);
2358 :
2359 8076 : PG_RETURN_BOOL(result);
2360 : }
2361 :
2362 : Datum
2363 18 : namenetext(PG_FUNCTION_ARGS)
2364 : {
2365 18 : Name arg1 = PG_GETARG_NAME(0);
2366 18 : text *arg2 = PG_GETARG_TEXT_PP(1);
2367 18 : size_t len1 = strlen(NameStr(*arg1));
2368 18 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2369 18 : Oid collid = PG_GET_COLLATION();
2370 : bool result;
2371 :
2372 18 : check_collation_set(collid);
2373 :
2374 18 : if (collid == C_COLLATION_OID)
2375 0 : result = !(len1 == len2 &&
2376 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2377 : else
2378 18 : result = !(varstr_cmp(NameStr(*arg1), len1,
2379 18 : VARDATA_ANY(arg2), len2,
2380 : collid) == 0);
2381 :
2382 18 : PG_FREE_IF_COPY(arg2, 1);
2383 :
2384 18 : PG_RETURN_BOOL(result);
2385 : }
2386 :
2387 : Datum
2388 18 : textnename(PG_FUNCTION_ARGS)
2389 : {
2390 18 : text *arg1 = PG_GETARG_TEXT_PP(0);
2391 18 : Name arg2 = PG_GETARG_NAME(1);
2392 18 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2393 18 : size_t len2 = strlen(NameStr(*arg2));
2394 18 : Oid collid = PG_GET_COLLATION();
2395 : bool result;
2396 :
2397 18 : check_collation_set(collid);
2398 :
2399 18 : if (collid == C_COLLATION_OID)
2400 0 : result = !(len1 == len2 &&
2401 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2402 : else
2403 18 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2404 18 : NameStr(*arg2), len2,
2405 : collid) == 0);
2406 :
2407 18 : PG_FREE_IF_COPY(arg1, 0);
2408 :
2409 18 : PG_RETURN_BOOL(result);
2410 : }
2411 :
2412 : Datum
2413 132158 : btnametextcmp(PG_FUNCTION_ARGS)
2414 : {
2415 132158 : Name arg1 = PG_GETARG_NAME(0);
2416 132158 : text *arg2 = PG_GETARG_TEXT_PP(1);
2417 : int32 result;
2418 :
2419 132158 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2420 132158 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2421 : PG_GET_COLLATION());
2422 :
2423 132158 : PG_FREE_IF_COPY(arg2, 1);
2424 :
2425 132158 : PG_RETURN_INT32(result);
2426 : }
2427 :
2428 : Datum
2429 44 : bttextnamecmp(PG_FUNCTION_ARGS)
2430 : {
2431 44 : text *arg1 = PG_GETARG_TEXT_PP(0);
2432 44 : Name arg2 = PG_GETARG_NAME(1);
2433 : int32 result;
2434 :
2435 44 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2436 44 : NameStr(*arg2), strlen(NameStr(*arg2)),
2437 : PG_GET_COLLATION());
2438 :
2439 44 : PG_FREE_IF_COPY(arg1, 0);
2440 :
2441 44 : PG_RETURN_INT32(result);
2442 : }
2443 :
2444 : #define CmpCall(cmpfunc) \
2445 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2446 : PG_GET_COLLATION(), \
2447 : PG_GETARG_DATUM(0), \
2448 : PG_GETARG_DATUM(1)))
2449 :
2450 : Datum
2451 63730 : namelttext(PG_FUNCTION_ARGS)
2452 : {
2453 63730 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2454 : }
2455 :
2456 : Datum
2457 0 : nameletext(PG_FUNCTION_ARGS)
2458 : {
2459 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2460 : }
2461 :
2462 : Datum
2463 0 : namegttext(PG_FUNCTION_ARGS)
2464 : {
2465 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2466 : }
2467 :
2468 : Datum
2469 55722 : namegetext(PG_FUNCTION_ARGS)
2470 : {
2471 55722 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2472 : }
2473 :
2474 : Datum
2475 0 : textltname(PG_FUNCTION_ARGS)
2476 : {
2477 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
2478 : }
2479 :
2480 : Datum
2481 0 : textlename(PG_FUNCTION_ARGS)
2482 : {
2483 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2484 : }
2485 :
2486 : Datum
2487 0 : textgtname(PG_FUNCTION_ARGS)
2488 : {
2489 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2490 : }
2491 :
2492 : Datum
2493 0 : textgename(PG_FUNCTION_ARGS)
2494 : {
2495 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2496 : }
2497 :
2498 : #undef CmpCall
2499 :
2500 :
2501 : /*
2502 : * The following operators support character-by-character comparison
2503 : * of text datums, to allow building indexes suitable for LIKE clauses.
2504 : * Note that the regular texteq/textne comparison operators, and regular
2505 : * support functions 1 and 2 with "C" collation are assumed to be
2506 : * compatible with these!
2507 : */
2508 :
2509 : static int
2510 160444 : internal_text_pattern_compare(text *arg1, text *arg2)
2511 : {
2512 : int result;
2513 : int len1,
2514 : len2;
2515 :
2516 160444 : len1 = VARSIZE_ANY_EXHDR(arg1);
2517 160444 : len2 = VARSIZE_ANY_EXHDR(arg2);
2518 :
2519 160444 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2520 160444 : if (result != 0)
2521 160312 : return result;
2522 132 : else if (len1 < len2)
2523 0 : return -1;
2524 132 : else if (len1 > len2)
2525 84 : return 1;
2526 : else
2527 48 : return 0;
2528 : }
2529 :
2530 :
2531 : Datum
2532 47866 : text_pattern_lt(PG_FUNCTION_ARGS)
2533 : {
2534 47866 : text *arg1 = PG_GETARG_TEXT_PP(0);
2535 47866 : text *arg2 = PG_GETARG_TEXT_PP(1);
2536 : int result;
2537 :
2538 47866 : result = internal_text_pattern_compare(arg1, arg2);
2539 :
2540 47866 : PG_FREE_IF_COPY(arg1, 0);
2541 47866 : PG_FREE_IF_COPY(arg2, 1);
2542 :
2543 47866 : PG_RETURN_BOOL(result < 0);
2544 : }
2545 :
2546 :
2547 : Datum
2548 37510 : text_pattern_le(PG_FUNCTION_ARGS)
2549 : {
2550 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2551 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2552 : int result;
2553 :
2554 37510 : result = internal_text_pattern_compare(arg1, arg2);
2555 :
2556 37510 : PG_FREE_IF_COPY(arg1, 0);
2557 37510 : PG_FREE_IF_COPY(arg2, 1);
2558 :
2559 37510 : PG_RETURN_BOOL(result <= 0);
2560 : }
2561 :
2562 :
2563 : Datum
2564 37534 : text_pattern_ge(PG_FUNCTION_ARGS)
2565 : {
2566 37534 : text *arg1 = PG_GETARG_TEXT_PP(0);
2567 37534 : text *arg2 = PG_GETARG_TEXT_PP(1);
2568 : int result;
2569 :
2570 37534 : result = internal_text_pattern_compare(arg1, arg2);
2571 :
2572 37534 : PG_FREE_IF_COPY(arg1, 0);
2573 37534 : PG_FREE_IF_COPY(arg2, 1);
2574 :
2575 37534 : PG_RETURN_BOOL(result >= 0);
2576 : }
2577 :
2578 :
2579 : Datum
2580 37510 : text_pattern_gt(PG_FUNCTION_ARGS)
2581 : {
2582 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2583 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2584 : int result;
2585 :
2586 37510 : result = internal_text_pattern_compare(arg1, arg2);
2587 :
2588 37510 : PG_FREE_IF_COPY(arg1, 0);
2589 37510 : PG_FREE_IF_COPY(arg2, 1);
2590 :
2591 37510 : PG_RETURN_BOOL(result > 0);
2592 : }
2593 :
2594 :
2595 : Datum
2596 24 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
2597 : {
2598 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
2599 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
2600 : int result;
2601 :
2602 24 : result = internal_text_pattern_compare(arg1, arg2);
2603 :
2604 24 : PG_FREE_IF_COPY(arg1, 0);
2605 24 : PG_FREE_IF_COPY(arg2, 1);
2606 :
2607 24 : PG_RETURN_INT32(result);
2608 : }
2609 :
2610 :
2611 : Datum
2612 116 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
2613 : {
2614 116 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
2615 : MemoryContext oldcontext;
2616 :
2617 116 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2618 :
2619 : /* Use generic string SortSupport, forcing "C" collation */
2620 116 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
2621 :
2622 116 : MemoryContextSwitchTo(oldcontext);
2623 :
2624 116 : PG_RETURN_VOID();
2625 : }
2626 :
2627 :
2628 : /* text_name()
2629 : * Converts a text type to a Name type.
2630 : */
2631 : Datum
2632 30890 : text_name(PG_FUNCTION_ARGS)
2633 : {
2634 30890 : text *s = PG_GETARG_TEXT_PP(0);
2635 : Name result;
2636 : int len;
2637 :
2638 30890 : len = VARSIZE_ANY_EXHDR(s);
2639 :
2640 : /* Truncate oversize input */
2641 30890 : if (len >= NAMEDATALEN)
2642 6 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
2643 :
2644 : /* We use palloc0 here to ensure result is zero-padded */
2645 30890 : result = (Name) palloc0(NAMEDATALEN);
2646 30890 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
2647 :
2648 30890 : PG_RETURN_NAME(result);
2649 : }
2650 :
2651 : /* name_text()
2652 : * Converts a Name type to a text type.
2653 : */
2654 : Datum
2655 656348 : name_text(PG_FUNCTION_ARGS)
2656 : {
2657 656348 : Name s = PG_GETARG_NAME(0);
2658 :
2659 656348 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
2660 : }
2661 :
2662 :
2663 : /*
2664 : * textToQualifiedNameList - convert a text object to list of names
2665 : *
2666 : * This implements the input parsing needed by nextval() and other
2667 : * functions that take a text parameter representing a qualified name.
2668 : * We split the name at dots, downcase if not double-quoted, and
2669 : * truncate names if they're too long.
2670 : */
2671 : List *
2672 5432 : textToQualifiedNameList(text *textval)
2673 : {
2674 : char *rawname;
2675 5432 : List *result = NIL;
2676 : List *namelist;
2677 : ListCell *l;
2678 :
2679 : /* Convert to C string (handles possible detoasting). */
2680 : /* Note we rely on being able to modify rawname below. */
2681 5432 : rawname = text_to_cstring(textval);
2682 :
2683 5432 : if (!SplitIdentifierString(rawname, '.', &namelist))
2684 0 : ereport(ERROR,
2685 : (errcode(ERRCODE_INVALID_NAME),
2686 : errmsg("invalid name syntax")));
2687 :
2688 5432 : if (namelist == NIL)
2689 0 : ereport(ERROR,
2690 : (errcode(ERRCODE_INVALID_NAME),
2691 : errmsg("invalid name syntax")));
2692 :
2693 10980 : foreach(l, namelist)
2694 : {
2695 5548 : char *curname = (char *) lfirst(l);
2696 :
2697 5548 : result = lappend(result, makeString(pstrdup(curname)));
2698 : }
2699 :
2700 5432 : pfree(rawname);
2701 5432 : list_free(namelist);
2702 :
2703 5432 : return result;
2704 : }
2705 :
2706 : /*
2707 : * SplitIdentifierString --- parse a string containing identifiers
2708 : *
2709 : * This is the guts of textToQualifiedNameList, and is exported for use in
2710 : * other situations such as parsing GUC variables. In the GUC case, it's
2711 : * important to avoid memory leaks, so the API is designed to minimize the
2712 : * amount of stuff that needs to be allocated and freed.
2713 : *
2714 : * Inputs:
2715 : * rawstring: the input string; must be overwritable! On return, it's
2716 : * been modified to contain the separated identifiers.
2717 : * separator: the separator punctuation expected between identifiers
2718 : * (typically '.' or ','). Whitespace may also appear around
2719 : * identifiers.
2720 : * Outputs:
2721 : * namelist: filled with a palloc'd list of pointers to identifiers within
2722 : * rawstring. Caller should list_free() this even on error return.
2723 : *
2724 : * Returns true if okay, false if there is a syntax error in the string.
2725 : *
2726 : * Note that an empty string is considered okay here, though not in
2727 : * textToQualifiedNameList.
2728 : */
2729 : bool
2730 341884 : SplitIdentifierString(char *rawstring, char separator,
2731 : List **namelist)
2732 : {
2733 341884 : char *nextp = rawstring;
2734 341884 : bool done = false;
2735 :
2736 341884 : *namelist = NIL;
2737 :
2738 341890 : while (scanner_isspace(*nextp))
2739 6 : nextp++; /* skip leading whitespace */
2740 :
2741 341884 : if (*nextp == '\0')
2742 30920 : return true; /* empty string represents empty list */
2743 :
2744 : /* At the top of the loop, we are at start of a new identifier. */
2745 : do
2746 : {
2747 : char *curname;
2748 : char *endp;
2749 :
2750 576676 : if (*nextp == '"')
2751 : {
2752 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
2753 42126 : curname = nextp + 1;
2754 : for (;;)
2755 : {
2756 42130 : endp = strchr(nextp + 1, '"');
2757 42128 : if (endp == NULL)
2758 0 : return false; /* mismatched quotes */
2759 42128 : if (endp[1] != '"')
2760 42126 : break; /* found end of quoted name */
2761 : /* Collapse adjacent quotes into one quote, and look again */
2762 2 : memmove(endp, endp + 1, strlen(endp));
2763 2 : nextp = endp;
2764 : }
2765 : /* endp now points at the terminating quote */
2766 42126 : nextp = endp + 1;
2767 : }
2768 : else
2769 : {
2770 : /* Unquoted name --- extends to separator or whitespace */
2771 : char *downname;
2772 : int len;
2773 :
2774 534550 : curname = nextp;
2775 4871536 : while (*nextp && *nextp != separator &&
2776 4336988 : !scanner_isspace(*nextp))
2777 4336986 : nextp++;
2778 534550 : endp = nextp;
2779 534550 : if (curname == nextp)
2780 0 : return false; /* empty unquoted name not allowed */
2781 :
2782 : /*
2783 : * Downcase the identifier, using same code as main lexer does.
2784 : *
2785 : * XXX because we want to overwrite the input in-place, we cannot
2786 : * support a downcasing transformation that increases the string
2787 : * length. This is not a problem given the current implementation
2788 : * of downcase_truncate_identifier, but we'll probably have to do
2789 : * something about this someday.
2790 : */
2791 534550 : len = endp - curname;
2792 534550 : downname = downcase_truncate_identifier(curname, len, false);
2793 : Assert(strlen(downname) <= len);
2794 534550 : strncpy(curname, downname, len); /* strncpy is required here */
2795 534550 : pfree(downname);
2796 : }
2797 :
2798 576678 : while (scanner_isspace(*nextp))
2799 2 : nextp++; /* skip trailing whitespace */
2800 :
2801 576676 : if (*nextp == separator)
2802 : {
2803 265712 : nextp++;
2804 505992 : while (scanner_isspace(*nextp))
2805 240280 : nextp++; /* skip leading whitespace for next */
2806 : /* we expect another name, so done remains false */
2807 : }
2808 310964 : else if (*nextp == '\0')
2809 310962 : done = true;
2810 : else
2811 2 : return false; /* invalid syntax */
2812 :
2813 : /* Now safe to overwrite separator with a null */
2814 576674 : *endp = '\0';
2815 :
2816 : /* Truncate name if it's overlength */
2817 576674 : truncate_identifier(curname, strlen(curname), false);
2818 :
2819 : /*
2820 : * Finished isolating current name --- add it to list
2821 : */
2822 576674 : *namelist = lappend(*namelist, curname);
2823 :
2824 : /* Loop back if we didn't reach end of string */
2825 576674 : } while (!done);
2826 :
2827 310962 : return true;
2828 : }
2829 :
2830 :
2831 : /*
2832 : * SplitDirectoriesString --- parse a string containing file/directory names
2833 : *
2834 : * This works fine on file names too; the function name is historical.
2835 : *
2836 : * This is similar to SplitIdentifierString, except that the parsing
2837 : * rules are meant to handle pathnames instead of identifiers: there is
2838 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
2839 : * and we apply canonicalize_path() to each extracted string. Because of the
2840 : * last, the returned strings are separately palloc'd rather than being
2841 : * pointers into rawstring --- but we still scribble on rawstring.
2842 : *
2843 : * Inputs:
2844 : * rawstring: the input string; must be modifiable!
2845 : * separator: the separator punctuation expected between directories
2846 : * (typically ',' or ';'). Whitespace may also appear around
2847 : * directories.
2848 : * Outputs:
2849 : * namelist: filled with a palloc'd list of directory names.
2850 : * Caller should list_free_deep() this even on error return.
2851 : *
2852 : * Returns true if okay, false if there is a syntax error in the string.
2853 : *
2854 : * Note that an empty string is considered okay here.
2855 : */
2856 : bool
2857 1908 : SplitDirectoriesString(char *rawstring, char separator,
2858 : List **namelist)
2859 : {
2860 1908 : char *nextp = rawstring;
2861 1908 : bool done = false;
2862 :
2863 1908 : *namelist = NIL;
2864 :
2865 1908 : while (scanner_isspace(*nextp))
2866 0 : nextp++; /* skip leading whitespace */
2867 :
2868 1908 : if (*nextp == '\0')
2869 2 : return true; /* empty string represents empty list */
2870 :
2871 : /* At the top of the loop, we are at start of a new directory. */
2872 : do
2873 : {
2874 : char *curname;
2875 : char *endp;
2876 :
2877 1916 : if (*nextp == '"')
2878 : {
2879 : /* Quoted name --- collapse quote-quote pairs */
2880 0 : curname = nextp + 1;
2881 : for (;;)
2882 : {
2883 0 : endp = strchr(nextp + 1, '"');
2884 0 : if (endp == NULL)
2885 0 : return false; /* mismatched quotes */
2886 0 : if (endp[1] != '"')
2887 0 : break; /* found end of quoted name */
2888 : /* Collapse adjacent quotes into one quote, and look again */
2889 0 : memmove(endp, endp + 1, strlen(endp));
2890 0 : nextp = endp;
2891 : }
2892 : /* endp now points at the terminating quote */
2893 0 : nextp = endp + 1;
2894 : }
2895 : else
2896 : {
2897 : /* Unquoted name --- extends to separator or end of string */
2898 1916 : curname = endp = nextp;
2899 32076 : while (*nextp && *nextp != separator)
2900 : {
2901 : /* trailing whitespace should not be included in name */
2902 30160 : if (!scanner_isspace(*nextp))
2903 30160 : endp = nextp + 1;
2904 30160 : nextp++;
2905 : }
2906 1916 : if (curname == endp)
2907 0 : return false; /* empty unquoted name not allowed */
2908 : }
2909 :
2910 1916 : while (scanner_isspace(*nextp))
2911 0 : nextp++; /* skip trailing whitespace */
2912 :
2913 1916 : if (*nextp == separator)
2914 : {
2915 10 : nextp++;
2916 16 : while (scanner_isspace(*nextp))
2917 6 : nextp++; /* skip leading whitespace for next */
2918 : /* we expect another name, so done remains false */
2919 : }
2920 1906 : else if (*nextp == '\0')
2921 1906 : done = true;
2922 : else
2923 0 : return false; /* invalid syntax */
2924 :
2925 : /* Now safe to overwrite separator with a null */
2926 1916 : *endp = '\0';
2927 :
2928 : /* Truncate path if it's overlength */
2929 1916 : if (strlen(curname) >= MAXPGPATH)
2930 0 : curname[MAXPGPATH - 1] = '\0';
2931 :
2932 : /*
2933 : * Finished isolating current name --- add it to list
2934 : */
2935 1916 : curname = pstrdup(curname);
2936 1916 : canonicalize_path(curname);
2937 1916 : *namelist = lappend(*namelist, curname);
2938 :
2939 : /* Loop back if we didn't reach end of string */
2940 1916 : } while (!done);
2941 :
2942 1906 : return true;
2943 : }
2944 :
2945 :
2946 : /*
2947 : * SplitGUCList --- parse a string containing identifiers or file names
2948 : *
2949 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
2950 : * presuming whether the elements will be taken as identifiers or file names.
2951 : * We assume the input has already been through flatten_set_variable_args(),
2952 : * so that we need never downcase (if appropriate, that was done already).
2953 : * Nor do we ever truncate, since we don't know the correct max length.
2954 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
2955 : * because any embedded whitespace should have led to double-quoting).
2956 : * Otherwise the API is identical to SplitIdentifierString.
2957 : *
2958 : * XXX it's annoying to have so many copies of this string-splitting logic.
2959 : * However, it's not clear that having one function with a bunch of option
2960 : * flags would be much better.
2961 : *
2962 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
2963 : * Be sure to update that if you have to change this.
2964 : *
2965 : * Inputs:
2966 : * rawstring: the input string; must be overwritable! On return, it's
2967 : * been modified to contain the separated identifiers.
2968 : * separator: the separator punctuation expected between identifiers
2969 : * (typically '.' or ','). Whitespace may also appear around
2970 : * identifiers.
2971 : * Outputs:
2972 : * namelist: filled with a palloc'd list of pointers to identifiers within
2973 : * rawstring. Caller should list_free() this even on error return.
2974 : *
2975 : * Returns true if okay, false if there is a syntax error in the string.
2976 : */
2977 : bool
2978 4200 : SplitGUCList(char *rawstring, char separator,
2979 : List **namelist)
2980 : {
2981 4200 : char *nextp = rawstring;
2982 4200 : bool done = false;
2983 :
2984 4200 : *namelist = NIL;
2985 :
2986 4200 : while (scanner_isspace(*nextp))
2987 0 : nextp++; /* skip leading whitespace */
2988 :
2989 4200 : if (*nextp == '\0')
2990 4124 : return true; /* empty string represents empty list */
2991 :
2992 : /* At the top of the loop, we are at start of a new identifier. */
2993 : do
2994 : {
2995 : char *curname;
2996 : char *endp;
2997 :
2998 102 : if (*nextp == '"')
2999 : {
3000 : /* Quoted name --- collapse quote-quote pairs */
3001 24 : curname = nextp + 1;
3002 : for (;;)
3003 : {
3004 36 : endp = strchr(nextp + 1, '"');
3005 30 : if (endp == NULL)
3006 0 : return false; /* mismatched quotes */
3007 30 : if (endp[1] != '"')
3008 24 : break; /* found end of quoted name */
3009 : /* Collapse adjacent quotes into one quote, and look again */
3010 6 : memmove(endp, endp + 1, strlen(endp));
3011 6 : nextp = endp;
3012 : }
3013 : /* endp now points at the terminating quote */
3014 24 : nextp = endp + 1;
3015 : }
3016 : else
3017 : {
3018 : /* Unquoted name --- extends to separator or whitespace */
3019 78 : curname = nextp;
3020 738 : while (*nextp && *nextp != separator &&
3021 660 : !scanner_isspace(*nextp))
3022 660 : nextp++;
3023 78 : endp = nextp;
3024 78 : if (curname == nextp)
3025 0 : return false; /* empty unquoted name not allowed */
3026 : }
3027 :
3028 102 : while (scanner_isspace(*nextp))
3029 0 : nextp++; /* skip trailing whitespace */
3030 :
3031 102 : if (*nextp == separator)
3032 : {
3033 26 : nextp++;
3034 44 : while (scanner_isspace(*nextp))
3035 18 : nextp++; /* skip leading whitespace for next */
3036 : /* we expect another name, so done remains false */
3037 : }
3038 76 : else if (*nextp == '\0')
3039 76 : done = true;
3040 : else
3041 0 : return false; /* invalid syntax */
3042 :
3043 : /* Now safe to overwrite separator with a null */
3044 102 : *endp = '\0';
3045 :
3046 : /*
3047 : * Finished isolating current name --- add it to list
3048 : */
3049 102 : *namelist = lappend(*namelist, curname);
3050 :
3051 : /* Loop back if we didn't reach end of string */
3052 102 : } while (!done);
3053 :
3054 76 : return true;
3055 : }
3056 :
3057 : /*
3058 : * appendStringInfoText
3059 : *
3060 : * Append a text to str.
3061 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3062 : */
3063 : static void
3064 2174048 : appendStringInfoText(StringInfo str, const text *t)
3065 : {
3066 2174048 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
3067 2174048 : }
3068 :
3069 : /*
3070 : * replace_text
3071 : * replace all occurrences of 'old_sub_str' in 'orig_str'
3072 : * with 'new_sub_str' to form 'new_str'
3073 : *
3074 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3075 : * otherwise returns 'new_str'
3076 : */
3077 : Datum
3078 1568 : replace_text(PG_FUNCTION_ARGS)
3079 : {
3080 1568 : text *src_text = PG_GETARG_TEXT_PP(0);
3081 1568 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
3082 1568 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
3083 : int src_text_len;
3084 : int from_sub_text_len;
3085 : TextPositionState state;
3086 : text *ret_text;
3087 : int chunk_len;
3088 : char *curr_ptr;
3089 : char *start_ptr;
3090 : StringInfoData str;
3091 : bool found;
3092 :
3093 1568 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
3094 1568 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
3095 :
3096 : /* Return unmodified source string if empty source or pattern */
3097 1568 : if (src_text_len < 1 || from_sub_text_len < 1)
3098 : {
3099 0 : PG_RETURN_TEXT_P(src_text);
3100 : }
3101 :
3102 1568 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
3103 :
3104 1568 : found = text_position_next(&state);
3105 :
3106 : /* When the from_sub_text is not found, there is nothing to do. */
3107 1568 : if (!found)
3108 : {
3109 334 : text_position_cleanup(&state);
3110 334 : PG_RETURN_TEXT_P(src_text);
3111 : }
3112 1234 : curr_ptr = text_position_get_match_ptr(&state);
3113 1234 : start_ptr = VARDATA_ANY(src_text);
3114 :
3115 1234 : initStringInfo(&str);
3116 :
3117 : do
3118 : {
3119 7216 : CHECK_FOR_INTERRUPTS();
3120 :
3121 : /* copy the data skipped over by last text_position_next() */
3122 7216 : chunk_len = curr_ptr - start_ptr;
3123 7216 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3124 :
3125 7216 : appendStringInfoText(&str, to_sub_text);
3126 :
3127 7216 : start_ptr = curr_ptr + state.last_match_len;
3128 :
3129 7216 : found = text_position_next(&state);
3130 7216 : if (found)
3131 5982 : curr_ptr = text_position_get_match_ptr(&state);
3132 : }
3133 7216 : while (found);
3134 :
3135 : /* copy trailing data */
3136 1234 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3137 1234 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3138 :
3139 1234 : text_position_cleanup(&state);
3140 :
3141 1234 : ret_text = cstring_to_text_with_len(str.data, str.len);
3142 1234 : pfree(str.data);
3143 :
3144 1234 : PG_RETURN_TEXT_P(ret_text);
3145 : }
3146 :
3147 : /*
3148 : * check_replace_text_has_escape
3149 : *
3150 : * Returns 0 if text contains no backslashes that need processing.
3151 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
3152 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
3153 : */
3154 : static int
3155 18770 : check_replace_text_has_escape(const text *replace_text)
3156 : {
3157 18770 : int result = 0;
3158 18770 : const char *p = VARDATA_ANY(replace_text);
3159 18770 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3160 :
3161 37584 : while (p < p_end)
3162 : {
3163 : /* Find next escape char, if any. */
3164 17636 : p = memchr(p, '\\', p_end - p);
3165 17636 : if (p == NULL)
3166 16812 : break;
3167 824 : p++;
3168 : /* Note: a backslash at the end doesn't require extra processing. */
3169 824 : if (p < p_end)
3170 : {
3171 824 : if (*p >= '1' && *p <= '9')
3172 780 : return 2; /* Found a submatch specifier, so done */
3173 44 : result = 1; /* Found some other sequence, keep looking */
3174 44 : p++;
3175 : }
3176 : }
3177 17990 : return result;
3178 : }
3179 :
3180 : /*
3181 : * appendStringInfoRegexpSubstr
3182 : *
3183 : * Append replace_text to str, substituting regexp back references for
3184 : * \n escapes. start_ptr is the start of the match in the source string,
3185 : * at logical character position data_pos.
3186 : */
3187 : static void
3188 236 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
3189 : regmatch_t *pmatch,
3190 : char *start_ptr, int data_pos)
3191 : {
3192 236 : const char *p = VARDATA_ANY(replace_text);
3193 236 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3194 :
3195 574 : while (p < p_end)
3196 : {
3197 518 : const char *chunk_start = p;
3198 : int so;
3199 : int eo;
3200 :
3201 : /* Find next escape char, if any. */
3202 518 : p = memchr(p, '\\', p_end - p);
3203 518 : if (p == NULL)
3204 174 : p = p_end;
3205 :
3206 : /* Copy the text we just scanned over, if any. */
3207 518 : if (p > chunk_start)
3208 318 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3209 :
3210 : /* Done if at end of string, else advance over escape char. */
3211 518 : if (p >= p_end)
3212 174 : break;
3213 344 : p++;
3214 :
3215 344 : if (p >= p_end)
3216 : {
3217 : /* Escape at very end of input. Treat same as unexpected char */
3218 6 : appendStringInfoChar(str, '\\');
3219 6 : break;
3220 : }
3221 :
3222 338 : if (*p >= '1' && *p <= '9')
3223 278 : {
3224 : /* Use the back reference of regexp. */
3225 278 : int idx = *p - '0';
3226 :
3227 278 : so = pmatch[idx].rm_so;
3228 278 : eo = pmatch[idx].rm_eo;
3229 278 : p++;
3230 : }
3231 60 : else if (*p == '&')
3232 : {
3233 : /* Use the entire matched string. */
3234 18 : so = pmatch[0].rm_so;
3235 18 : eo = pmatch[0].rm_eo;
3236 18 : p++;
3237 : }
3238 42 : else if (*p == '\\')
3239 : {
3240 : /* \\ means transfer one \ to output. */
3241 36 : appendStringInfoChar(str, '\\');
3242 36 : p++;
3243 36 : continue;
3244 : }
3245 : else
3246 : {
3247 : /*
3248 : * If escape char is not followed by any expected char, just treat
3249 : * it as ordinary data to copy. (XXX would it be better to throw
3250 : * an error?)
3251 : */
3252 6 : appendStringInfoChar(str, '\\');
3253 6 : continue;
3254 : }
3255 :
3256 296 : if (so >= 0 && eo >= 0)
3257 : {
3258 : /*
3259 : * Copy the text that is back reference of regexp. Note so and eo
3260 : * are counted in characters not bytes.
3261 : */
3262 : char *chunk_start;
3263 : int chunk_len;
3264 :
3265 : Assert(so >= data_pos);
3266 296 : chunk_start = start_ptr;
3267 296 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3268 296 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3269 296 : appendBinaryStringInfo(str, chunk_start, chunk_len);
3270 : }
3271 : }
3272 236 : }
3273 :
3274 : /*
3275 : * replace_text_regexp
3276 : *
3277 : * replace substring(s) in src_text that match pattern with replace_text.
3278 : * The replace_text can contain backslash markers to substitute
3279 : * (parts of) the matched text.
3280 : *
3281 : * cflags: regexp compile flags.
3282 : * collation: collation to use.
3283 : * search_start: the character (not byte) offset in src_text at which to
3284 : * begin searching.
3285 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
3286 : */
3287 : text *
3288 18770 : replace_text_regexp(text *src_text, text *pattern_text,
3289 : text *replace_text,
3290 : int cflags, Oid collation,
3291 : int search_start, int n)
3292 : {
3293 : text *ret_text;
3294 : regex_t *re;
3295 18770 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3296 18770 : int nmatches = 0;
3297 : StringInfoData buf;
3298 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
3299 18770 : int nmatch = lengthof(pmatch);
3300 : pg_wchar *data;
3301 : size_t data_len;
3302 : int data_pos;
3303 : char *start_ptr;
3304 : int escape_status;
3305 :
3306 18770 : initStringInfo(&buf);
3307 :
3308 : /* Convert data string to wide characters. */
3309 18770 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3310 18770 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3311 :
3312 : /* Check whether replace_text has escapes, especially regexp submatches. */
3313 18770 : escape_status = check_replace_text_has_escape(replace_text);
3314 :
3315 : /* If no regexp submatches, we can use REG_NOSUB. */
3316 18770 : if (escape_status < 2)
3317 : {
3318 17990 : cflags |= REG_NOSUB;
3319 : /* Also tell pg_regexec we only want the whole-match location. */
3320 17990 : nmatch = 1;
3321 : }
3322 :
3323 : /* Prepare the regexp. */
3324 18770 : re = RE_compile_and_cache(pattern_text, cflags, collation);
3325 :
3326 : /* start_ptr points to the data_pos'th character of src_text */
3327 18770 : start_ptr = (char *) VARDATA_ANY(src_text);
3328 18770 : data_pos = 0;
3329 :
3330 25192 : while (search_start <= data_len)
3331 : {
3332 : int regexec_result;
3333 :
3334 25186 : CHECK_FOR_INTERRUPTS();
3335 :
3336 25186 : regexec_result = pg_regexec(re,
3337 : data,
3338 : data_len,
3339 : search_start,
3340 : NULL, /* no details */
3341 : nmatch,
3342 : pmatch,
3343 : 0);
3344 :
3345 25186 : if (regexec_result == REG_NOMATCH)
3346 16698 : break;
3347 :
3348 8488 : if (regexec_result != REG_OKAY)
3349 : {
3350 : char errMsg[100];
3351 :
3352 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3353 0 : ereport(ERROR,
3354 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3355 : errmsg("regular expression failed: %s", errMsg)));
3356 : }
3357 :
3358 : /*
3359 : * Count matches, and decide whether to replace this match.
3360 : */
3361 8488 : nmatches++;
3362 8488 : if (n > 0 && nmatches != n)
3363 : {
3364 : /*
3365 : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
3366 : * we treat the matched text as if it weren't matched, and copy it
3367 : * to the output later.)
3368 : */
3369 60 : search_start = pmatch[0].rm_eo;
3370 60 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3371 0 : search_start++;
3372 60 : continue;
3373 : }
3374 :
3375 : /*
3376 : * Copy the text to the left of the match position. Note we are given
3377 : * character not byte indexes.
3378 : */
3379 8428 : if (pmatch[0].rm_so - data_pos > 0)
3380 : {
3381 : int chunk_len;
3382 :
3383 8254 : chunk_len = charlen_to_bytelen(start_ptr,
3384 8254 : pmatch[0].rm_so - data_pos);
3385 8254 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3386 :
3387 : /*
3388 : * Advance start_ptr over that text, to avoid multiple rescans of
3389 : * it if the replace_text contains multiple back-references.
3390 : */
3391 8254 : start_ptr += chunk_len;
3392 8254 : data_pos = pmatch[0].rm_so;
3393 : }
3394 :
3395 : /*
3396 : * Copy the replace_text, processing escapes if any are present.
3397 : */
3398 8428 : if (escape_status > 0)
3399 236 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3400 : start_ptr, data_pos);
3401 : else
3402 8192 : appendStringInfoText(&buf, replace_text);
3403 :
3404 : /* Advance start_ptr and data_pos over the matched text. */
3405 16856 : start_ptr += charlen_to_bytelen(start_ptr,
3406 8428 : pmatch[0].rm_eo - data_pos);
3407 8428 : data_pos = pmatch[0].rm_eo;
3408 :
3409 : /*
3410 : * If we only want to replace one occurrence, we're done.
3411 : */
3412 8428 : if (n > 0)
3413 2066 : break;
3414 :
3415 : /*
3416 : * Advance search position. Normally we start the next search at the
3417 : * end of the previous match; but if the match was of zero length, we
3418 : * have to advance by one character, or we'd just find the same match
3419 : * again.
3420 : */
3421 6362 : search_start = data_pos;
3422 6362 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3423 12 : search_start++;
3424 : }
3425 :
3426 : /*
3427 : * Copy the text to the right of the last match.
3428 : */
3429 18770 : if (data_pos < data_len)
3430 : {
3431 : int chunk_len;
3432 :
3433 17894 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3434 17894 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3435 : }
3436 :
3437 18770 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
3438 18770 : pfree(buf.data);
3439 18770 : pfree(data);
3440 :
3441 18770 : return ret_text;
3442 : }
3443 :
3444 : /*
3445 : * split_part
3446 : * parse input string based on provided field separator
3447 : * return N'th item (1 based, negative counts from end)
3448 : */
3449 : Datum
3450 150 : split_part(PG_FUNCTION_ARGS)
3451 : {
3452 150 : text *inputstring = PG_GETARG_TEXT_PP(0);
3453 150 : text *fldsep = PG_GETARG_TEXT_PP(1);
3454 150 : int fldnum = PG_GETARG_INT32(2);
3455 : int inputstring_len;
3456 : int fldsep_len;
3457 : TextPositionState state;
3458 : char *start_ptr;
3459 : char *end_ptr;
3460 : text *result_text;
3461 : bool found;
3462 :
3463 : /* field number is 1 based */
3464 150 : if (fldnum == 0)
3465 6 : ereport(ERROR,
3466 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3467 : errmsg("field position must not be zero")));
3468 :
3469 144 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3470 144 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3471 :
3472 : /* return empty string for empty input string */
3473 144 : if (inputstring_len < 1)
3474 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3475 :
3476 : /* handle empty field separator */
3477 132 : if (fldsep_len < 1)
3478 : {
3479 : /* if first or last field, return input string, else empty string */
3480 24 : if (fldnum == 1 || fldnum == -1)
3481 12 : PG_RETURN_TEXT_P(inputstring);
3482 : else
3483 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3484 : }
3485 :
3486 : /* find the first field separator */
3487 108 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
3488 :
3489 108 : found = text_position_next(&state);
3490 :
3491 : /* special case if fldsep not found at all */
3492 108 : if (!found)
3493 : {
3494 24 : text_position_cleanup(&state);
3495 : /* if first or last field, return input string, else empty string */
3496 24 : if (fldnum == 1 || fldnum == -1)
3497 12 : PG_RETURN_TEXT_P(inputstring);
3498 : else
3499 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3500 : }
3501 :
3502 : /*
3503 : * take care of a negative field number (i.e. count from the right) by
3504 : * converting to a positive field number; we need total number of fields
3505 : */
3506 84 : if (fldnum < 0)
3507 : {
3508 : /* we found a fldsep, so there are at least two fields */
3509 42 : int numfields = 2;
3510 :
3511 54 : while (text_position_next(&state))
3512 12 : numfields++;
3513 :
3514 : /* special case of last field does not require an extra pass */
3515 42 : if (fldnum == -1)
3516 : {
3517 24 : start_ptr = text_position_get_match_ptr(&state) + state.last_match_len;
3518 24 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
3519 24 : text_position_cleanup(&state);
3520 24 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
3521 : end_ptr - start_ptr));
3522 : }
3523 :
3524 : /* else, convert fldnum to positive notation */
3525 18 : fldnum += numfields + 1;
3526 :
3527 : /* if nonexistent field, return empty string */
3528 18 : if (fldnum <= 0)
3529 : {
3530 6 : text_position_cleanup(&state);
3531 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
3532 : }
3533 :
3534 : /* reset to pointing at first match, but now with positive fldnum */
3535 12 : text_position_reset(&state);
3536 12 : found = text_position_next(&state);
3537 : Assert(found);
3538 : }
3539 :
3540 : /* identify bounds of first field */
3541 54 : start_ptr = VARDATA_ANY(inputstring);
3542 54 : end_ptr = text_position_get_match_ptr(&state);
3543 :
3544 102 : while (found && --fldnum > 0)
3545 : {
3546 : /* identify bounds of next field */
3547 48 : start_ptr = end_ptr + state.last_match_len;
3548 48 : found = text_position_next(&state);
3549 48 : if (found)
3550 18 : end_ptr = text_position_get_match_ptr(&state);
3551 : }
3552 :
3553 54 : text_position_cleanup(&state);
3554 :
3555 54 : if (fldnum > 0)
3556 : {
3557 : /* N'th field separator not found */
3558 : /* if last field requested, return it, else empty string */
3559 30 : if (fldnum == 1)
3560 : {
3561 24 : int last_len = start_ptr - VARDATA_ANY(inputstring);
3562 :
3563 24 : result_text = cstring_to_text_with_len(start_ptr,
3564 : inputstring_len - last_len);
3565 : }
3566 : else
3567 6 : result_text = cstring_to_text("");
3568 : }
3569 : else
3570 : {
3571 : /* non-last field requested */
3572 24 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
3573 : }
3574 :
3575 54 : PG_RETURN_TEXT_P(result_text);
3576 : }
3577 :
3578 : /*
3579 : * Convenience function to return true when two text params are equal.
3580 : */
3581 : static bool
3582 384 : text_isequal(text *txt1, text *txt2, Oid collid)
3583 : {
3584 384 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
3585 : collid,
3586 : PointerGetDatum(txt1),
3587 : PointerGetDatum(txt2)));
3588 : }
3589 :
3590 : /*
3591 : * text_to_array
3592 : * parse input string and return text array of elements,
3593 : * based on provided field separator
3594 : */
3595 : Datum
3596 170 : text_to_array(PG_FUNCTION_ARGS)
3597 : {
3598 : SplitTextOutputData tstate;
3599 :
3600 : /* For array output, tstate should start as all zeroes */
3601 170 : memset(&tstate, 0, sizeof(tstate));
3602 :
3603 170 : if (!split_text(fcinfo, &tstate))
3604 6 : PG_RETURN_NULL();
3605 :
3606 164 : if (tstate.astate == NULL)
3607 6 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
3608 :
3609 158 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
3610 : CurrentMemoryContext));
3611 : }
3612 :
3613 : /*
3614 : * text_to_array_null
3615 : * parse input string and return text array of elements,
3616 : * based on provided field separator and null string
3617 : *
3618 : * This is a separate entry point only to prevent the regression tests from
3619 : * complaining about different argument sets for the same internal function.
3620 : */
3621 : Datum
3622 60 : text_to_array_null(PG_FUNCTION_ARGS)
3623 : {
3624 60 : return text_to_array(fcinfo);
3625 : }
3626 :
3627 : /*
3628 : * text_to_table
3629 : * parse input string and return table of elements,
3630 : * based on provided field separator
3631 : */
3632 : Datum
3633 84 : text_to_table(PG_FUNCTION_ARGS)
3634 : {
3635 84 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
3636 : SplitTextOutputData tstate;
3637 :
3638 84 : tstate.astate = NULL;
3639 84 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
3640 84 : tstate.tupstore = rsi->setResult;
3641 84 : tstate.tupdesc = rsi->setDesc;
3642 :
3643 84 : (void) split_text(fcinfo, &tstate);
3644 :
3645 84 : return (Datum) 0;
3646 : }
3647 :
3648 : /*
3649 : * text_to_table_null
3650 : * parse input string and return table of elements,
3651 : * based on provided field separator and null string
3652 : *
3653 : * This is a separate entry point only to prevent the regression tests from
3654 : * complaining about different argument sets for the same internal function.
3655 : */
3656 : Datum
3657 24 : text_to_table_null(PG_FUNCTION_ARGS)
3658 : {
3659 24 : return text_to_table(fcinfo);
3660 : }
3661 :
3662 : /*
3663 : * Common code for text_to_array, text_to_array_null, text_to_table
3664 : * and text_to_table_null functions.
3665 : *
3666 : * These are not strict so we have to test for null inputs explicitly.
3667 : * Returns false if result is to be null, else returns true.
3668 : *
3669 : * Note that if the result is valid but empty (zero elements), we return
3670 : * without changing *tstate --- caller must handle that case, too.
3671 : */
3672 : static bool
3673 254 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
3674 : {
3675 : text *inputstring;
3676 : text *fldsep;
3677 : text *null_string;
3678 254 : Oid collation = PG_GET_COLLATION();
3679 : int inputstring_len;
3680 : int fldsep_len;
3681 : char *start_ptr;
3682 : text *result_text;
3683 :
3684 : /* when input string is NULL, then result is NULL too */
3685 254 : if (PG_ARGISNULL(0))
3686 12 : return false;
3687 :
3688 242 : inputstring = PG_GETARG_TEXT_PP(0);
3689 :
3690 : /* fldsep can be NULL */
3691 242 : if (!PG_ARGISNULL(1))
3692 212 : fldsep = PG_GETARG_TEXT_PP(1);
3693 : else
3694 30 : fldsep = NULL;
3695 :
3696 : /* null_string can be NULL or omitted */
3697 242 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
3698 84 : null_string = PG_GETARG_TEXT_PP(2);
3699 : else
3700 158 : null_string = NULL;
3701 :
3702 242 : if (fldsep != NULL)
3703 : {
3704 : /*
3705 : * Normal case with non-null fldsep. Use the text_position machinery
3706 : * to search for occurrences of fldsep.
3707 : */
3708 : TextPositionState state;
3709 :
3710 212 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3711 212 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3712 :
3713 : /* return empty set for empty input string */
3714 212 : if (inputstring_len < 1)
3715 60 : return true;
3716 :
3717 : /* empty field separator: return input string as a one-element set */
3718 200 : if (fldsep_len < 1)
3719 : {
3720 48 : split_text_accum_result(tstate, inputstring,
3721 : null_string, collation);
3722 48 : return true;
3723 : }
3724 :
3725 152 : text_position_setup(inputstring, fldsep, collation, &state);
3726 :
3727 152 : start_ptr = VARDATA_ANY(inputstring);
3728 :
3729 : for (;;)
3730 512 : {
3731 : bool found;
3732 : char *end_ptr;
3733 : int chunk_len;
3734 :
3735 664 : CHECK_FOR_INTERRUPTS();
3736 :
3737 664 : found = text_position_next(&state);
3738 664 : if (!found)
3739 : {
3740 : /* fetch last field */
3741 152 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
3742 152 : end_ptr = NULL; /* not used, but some compilers complain */
3743 : }
3744 : else
3745 : {
3746 : /* fetch non-last field */
3747 512 : end_ptr = text_position_get_match_ptr(&state);
3748 512 : chunk_len = end_ptr - start_ptr;
3749 : }
3750 :
3751 : /* build a temp text datum to pass to split_text_accum_result */
3752 664 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3753 :
3754 : /* stash away this field */
3755 664 : split_text_accum_result(tstate, result_text,
3756 : null_string, collation);
3757 :
3758 664 : pfree(result_text);
3759 :
3760 664 : if (!found)
3761 152 : break;
3762 :
3763 512 : start_ptr = end_ptr + state.last_match_len;
3764 : }
3765 :
3766 152 : text_position_cleanup(&state);
3767 : }
3768 : else
3769 : {
3770 : /*
3771 : * When fldsep is NULL, each character in the input string becomes a
3772 : * separate element in the result set. The separator is effectively
3773 : * the space between characters.
3774 : */
3775 30 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3776 :
3777 30 : start_ptr = VARDATA_ANY(inputstring);
3778 :
3779 252 : while (inputstring_len > 0)
3780 : {
3781 222 : int chunk_len = pg_mblen(start_ptr);
3782 :
3783 222 : CHECK_FOR_INTERRUPTS();
3784 :
3785 : /* build a temp text datum to pass to split_text_accum_result */
3786 222 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3787 :
3788 : /* stash away this field */
3789 222 : split_text_accum_result(tstate, result_text,
3790 : null_string, collation);
3791 :
3792 222 : pfree(result_text);
3793 :
3794 222 : start_ptr += chunk_len;
3795 222 : inputstring_len -= chunk_len;
3796 : }
3797 : }
3798 :
3799 182 : return true;
3800 : }
3801 :
3802 : /*
3803 : * Add text item to result set (table or array).
3804 : *
3805 : * This is also responsible for checking to see if the item matches
3806 : * the null_string, in which case we should emit NULL instead.
3807 : */
3808 : static void
3809 934 : split_text_accum_result(SplitTextOutputData *tstate,
3810 : text *field_value,
3811 : text *null_string,
3812 : Oid collation)
3813 : {
3814 934 : bool is_null = false;
3815 :
3816 934 : if (null_string && text_isequal(field_value, null_string, collation))
3817 72 : is_null = true;
3818 :
3819 934 : if (tstate->tupstore)
3820 : {
3821 : Datum values[1];
3822 : bool nulls[1];
3823 :
3824 228 : values[0] = PointerGetDatum(field_value);
3825 228 : nulls[0] = is_null;
3826 :
3827 228 : tuplestore_putvalues(tstate->tupstore,
3828 : tstate->tupdesc,
3829 : values,
3830 : nulls);
3831 : }
3832 : else
3833 : {
3834 706 : tstate->astate = accumArrayResult(tstate->astate,
3835 : PointerGetDatum(field_value),
3836 : is_null,
3837 : TEXTOID,
3838 : CurrentMemoryContext);
3839 : }
3840 934 : }
3841 :
3842 : /*
3843 : * array_to_text
3844 : * concatenate Cstring representation of input array elements
3845 : * using provided field separator
3846 : */
3847 : Datum
3848 77108 : array_to_text(PG_FUNCTION_ARGS)
3849 : {
3850 77108 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
3851 77108 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3852 :
3853 77108 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
3854 : }
3855 :
3856 : /*
3857 : * array_to_text_null
3858 : * concatenate Cstring representation of input array elements
3859 : * using provided field separator and null string
3860 : *
3861 : * This version is not strict so we have to test for null inputs explicitly.
3862 : */
3863 : Datum
3864 12 : array_to_text_null(PG_FUNCTION_ARGS)
3865 : {
3866 : ArrayType *v;
3867 : char *fldsep;
3868 : char *null_string;
3869 :
3870 : /* returns NULL when first or second parameter is NULL */
3871 12 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
3872 0 : PG_RETURN_NULL();
3873 :
3874 12 : v = PG_GETARG_ARRAYTYPE_P(0);
3875 12 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3876 :
3877 : /* NULL null string is passed through as a null pointer */
3878 12 : if (!PG_ARGISNULL(2))
3879 6 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
3880 : else
3881 6 : null_string = NULL;
3882 :
3883 12 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
3884 : }
3885 :
3886 : /*
3887 : * common code for array_to_text and array_to_text_null functions
3888 : */
3889 : static text *
3890 77138 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
3891 : const char *fldsep, const char *null_string)
3892 : {
3893 : text *result;
3894 : int nitems,
3895 : *dims,
3896 : ndims;
3897 : Oid element_type;
3898 : int typlen;
3899 : bool typbyval;
3900 : char typalign;
3901 : uint8 typalignby;
3902 : StringInfoData buf;
3903 77138 : bool printed = false;
3904 : char *p;
3905 : bits8 *bitmap;
3906 : int bitmask;
3907 : int i;
3908 : ArrayMetaState *my_extra;
3909 :
3910 77138 : ndims = ARR_NDIM(v);
3911 77138 : dims = ARR_DIMS(v);
3912 77138 : nitems = ArrayGetNItems(ndims, dims);
3913 :
3914 : /* if there are no elements, return an empty string */
3915 77138 : if (nitems == 0)
3916 51912 : return cstring_to_text_with_len("", 0);
3917 :
3918 25226 : element_type = ARR_ELEMTYPE(v);
3919 25226 : initStringInfo(&buf);
3920 :
3921 : /*
3922 : * We arrange to look up info about element type, including its output
3923 : * conversion proc, only once per series of calls, assuming the element
3924 : * type doesn't change underneath us.
3925 : */
3926 25226 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3927 25226 : if (my_extra == NULL)
3928 : {
3929 1420 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
3930 : sizeof(ArrayMetaState));
3931 1420 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3932 1420 : my_extra->element_type = ~element_type;
3933 : }
3934 :
3935 25226 : if (my_extra->element_type != element_type)
3936 : {
3937 : /*
3938 : * Get info about element type, including its output conversion proc
3939 : */
3940 1420 : get_type_io_data(element_type, IOFunc_output,
3941 : &my_extra->typlen, &my_extra->typbyval,
3942 : &my_extra->typalign, &my_extra->typdelim,
3943 : &my_extra->typioparam, &my_extra->typiofunc);
3944 1420 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
3945 1420 : fcinfo->flinfo->fn_mcxt);
3946 1420 : my_extra->element_type = element_type;
3947 : }
3948 25226 : typlen = my_extra->typlen;
3949 25226 : typbyval = my_extra->typbyval;
3950 25226 : typalign = my_extra->typalign;
3951 25226 : typalignby = typalign_to_alignby(typalign);
3952 :
3953 25226 : p = ARR_DATA_PTR(v);
3954 25226 : bitmap = ARR_NULLBITMAP(v);
3955 25226 : bitmask = 1;
3956 :
3957 85738 : for (i = 0; i < nitems; i++)
3958 : {
3959 : Datum itemvalue;
3960 : char *value;
3961 :
3962 : /* Get source element, checking for NULL */
3963 60512 : if (bitmap && (*bitmap & bitmask) == 0)
3964 : {
3965 : /* if null_string is NULL, we just ignore null elements */
3966 18 : if (null_string != NULL)
3967 : {
3968 6 : if (printed)
3969 6 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
3970 : else
3971 0 : appendStringInfoString(&buf, null_string);
3972 6 : printed = true;
3973 : }
3974 : }
3975 : else
3976 : {
3977 60494 : itemvalue = fetch_att(p, typbyval, typlen);
3978 :
3979 60494 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
3980 :
3981 60494 : if (printed)
3982 35268 : appendStringInfo(&buf, "%s%s", fldsep, value);
3983 : else
3984 25226 : appendStringInfoString(&buf, value);
3985 60494 : printed = true;
3986 :
3987 60494 : p = att_addlength_pointer(p, typlen, p);
3988 60494 : p = (char *) att_nominal_alignby(p, typalignby);
3989 : }
3990 :
3991 : /* advance bitmap pointer if any */
3992 60512 : if (bitmap)
3993 : {
3994 108 : bitmask <<= 1;
3995 108 : if (bitmask == 0x100)
3996 : {
3997 0 : bitmap++;
3998 0 : bitmask = 1;
3999 : }
4000 : }
4001 : }
4002 :
4003 25226 : result = cstring_to_text_with_len(buf.data, buf.len);
4004 25226 : pfree(buf.data);
4005 :
4006 25226 : return result;
4007 : }
4008 :
4009 : /*
4010 : * Workhorse for to_bin, to_oct, and to_hex. Note that base must be > 1 and <=
4011 : * 16.
4012 : */
4013 : static inline text *
4014 38750 : convert_to_base(uint64 value, int base)
4015 : {
4016 38750 : const char *digits = "0123456789abcdef";
4017 :
4018 : /* We size the buffer for to_bin's longest possible return value. */
4019 : char buf[sizeof(uint64) * BITS_PER_BYTE];
4020 38750 : char *const end = buf + sizeof(buf);
4021 38750 : char *ptr = end;
4022 :
4023 : Assert(base > 1);
4024 : Assert(base <= 16);
4025 :
4026 : do
4027 : {
4028 75970 : *--ptr = digits[value % base];
4029 75970 : value /= base;
4030 75970 : } while (ptr > buf && value);
4031 :
4032 38750 : return cstring_to_text_with_len(ptr, end - ptr);
4033 : }
4034 :
4035 : /*
4036 : * Convert an integer to a string containing a base-2 (binary) representation
4037 : * of the number.
4038 : */
4039 : Datum
4040 12 : to_bin32(PG_FUNCTION_ARGS)
4041 : {
4042 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4043 :
4044 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4045 : }
4046 : Datum
4047 12 : to_bin64(PG_FUNCTION_ARGS)
4048 : {
4049 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4050 :
4051 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4052 : }
4053 :
4054 : /*
4055 : * Convert an integer to a string containing a base-8 (oct) representation of
4056 : * the number.
4057 : */
4058 : Datum
4059 12 : to_oct32(PG_FUNCTION_ARGS)
4060 : {
4061 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4062 :
4063 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4064 : }
4065 : Datum
4066 12 : to_oct64(PG_FUNCTION_ARGS)
4067 : {
4068 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4069 :
4070 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4071 : }
4072 :
4073 : /*
4074 : * Convert an integer to a string containing a base-16 (hex) representation of
4075 : * the number.
4076 : */
4077 : Datum
4078 38690 : to_hex32(PG_FUNCTION_ARGS)
4079 : {
4080 38690 : uint64 value = (uint32) PG_GETARG_INT32(0);
4081 :
4082 38690 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4083 : }
4084 : Datum
4085 12 : to_hex64(PG_FUNCTION_ARGS)
4086 : {
4087 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4088 :
4089 12 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4090 : }
4091 :
4092 : /*
4093 : * Return the size of a datum, possibly compressed
4094 : *
4095 : * Works on any data type
4096 : */
4097 : Datum
4098 122 : pg_column_size(PG_FUNCTION_ARGS)
4099 : {
4100 122 : Datum value = PG_GETARG_DATUM(0);
4101 : int32 result;
4102 : int typlen;
4103 :
4104 : /* On first call, get the input type's typlen, and save at *fn_extra */
4105 122 : if (fcinfo->flinfo->fn_extra == NULL)
4106 : {
4107 : /* Lookup the datatype of the supplied argument */
4108 122 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4109 :
4110 122 : typlen = get_typlen(argtypeid);
4111 122 : if (typlen == 0) /* should not happen */
4112 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4113 :
4114 122 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4115 : sizeof(int));
4116 122 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4117 : }
4118 : else
4119 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4120 :
4121 122 : if (typlen == -1)
4122 : {
4123 : /* varlena type, possibly toasted */
4124 122 : result = toast_datum_size(value);
4125 : }
4126 0 : else if (typlen == -2)
4127 : {
4128 : /* cstring */
4129 0 : result = strlen(DatumGetCString(value)) + 1;
4130 : }
4131 : else
4132 : {
4133 : /* ordinary fixed-width type */
4134 0 : result = typlen;
4135 : }
4136 :
4137 122 : PG_RETURN_INT32(result);
4138 : }
4139 :
4140 : /*
4141 : * Return the compression method stored in the compressed attribute. Return
4142 : * NULL for non varlena type or uncompressed data.
4143 : */
4144 : Datum
4145 192 : pg_column_compression(PG_FUNCTION_ARGS)
4146 : {
4147 : int typlen;
4148 : char *result;
4149 : ToastCompressionId cmid;
4150 :
4151 : /* On first call, get the input type's typlen, and save at *fn_extra */
4152 192 : if (fcinfo->flinfo->fn_extra == NULL)
4153 : {
4154 : /* Lookup the datatype of the supplied argument */
4155 156 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4156 :
4157 156 : typlen = get_typlen(argtypeid);
4158 156 : if (typlen == 0) /* should not happen */
4159 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4160 :
4161 156 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4162 : sizeof(int));
4163 156 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4164 : }
4165 : else
4166 36 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4167 :
4168 192 : if (typlen != -1)
4169 0 : PG_RETURN_NULL();
4170 :
4171 : /* get the compression method id stored in the compressed varlena */
4172 192 : cmid = toast_get_compression_id((struct varlena *)
4173 192 : DatumGetPointer(PG_GETARG_DATUM(0)));
4174 192 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
4175 42 : PG_RETURN_NULL();
4176 :
4177 : /* convert compression method id to compression method name */
4178 150 : switch (cmid)
4179 : {
4180 84 : case TOAST_PGLZ_COMPRESSION_ID:
4181 84 : result = "pglz";
4182 84 : break;
4183 66 : case TOAST_LZ4_COMPRESSION_ID:
4184 66 : result = "lz4";
4185 66 : break;
4186 0 : default:
4187 0 : elog(ERROR, "invalid compression method id %d", cmid);
4188 : }
4189 :
4190 150 : PG_RETURN_TEXT_P(cstring_to_text(result));
4191 : }
4192 :
4193 : /*
4194 : * Return the chunk_id of the on-disk TOASTed value. Return NULL if the value
4195 : * is un-TOASTed or not on-disk.
4196 : */
4197 : Datum
4198 52 : pg_column_toast_chunk_id(PG_FUNCTION_ARGS)
4199 : {
4200 : int typlen;
4201 : struct varlena *attr;
4202 : struct varatt_external toast_pointer;
4203 :
4204 : /* On first call, get the input type's typlen, and save at *fn_extra */
4205 52 : if (fcinfo->flinfo->fn_extra == NULL)
4206 : {
4207 : /* Lookup the datatype of the supplied argument */
4208 40 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4209 :
4210 40 : typlen = get_typlen(argtypeid);
4211 40 : if (typlen == 0) /* should not happen */
4212 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4213 :
4214 40 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4215 : sizeof(int));
4216 40 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4217 : }
4218 : else
4219 12 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4220 :
4221 52 : if (typlen != -1)
4222 0 : PG_RETURN_NULL();
4223 :
4224 52 : attr = (struct varlena *) DatumGetPointer(PG_GETARG_DATUM(0));
4225 :
4226 52 : if (!VARATT_IS_EXTERNAL_ONDISK(attr))
4227 12 : PG_RETURN_NULL();
4228 :
4229 40 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
4230 :
4231 40 : PG_RETURN_OID(toast_pointer.va_valueid);
4232 : }
4233 :
4234 : /*
4235 : * string_agg - Concatenates values and returns string.
4236 : *
4237 : * Syntax: string_agg(value text, delimiter text) RETURNS text
4238 : *
4239 : * Note: Any NULL values are ignored. The first-call delimiter isn't
4240 : * actually used at all, and on subsequent calls the delimiter precedes
4241 : * the associated value.
4242 : */
4243 :
4244 : /* subroutine to initialize state */
4245 : static StringInfo
4246 2372 : makeStringAggState(FunctionCallInfo fcinfo)
4247 : {
4248 : StringInfo state;
4249 : MemoryContext aggcontext;
4250 : MemoryContext oldcontext;
4251 :
4252 2372 : if (!AggCheckCallContext(fcinfo, &aggcontext))
4253 : {
4254 : /* cannot be called directly because of internal-type argument */
4255 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
4256 : }
4257 :
4258 : /*
4259 : * Create state in aggregate context. It'll stay there across subsequent
4260 : * calls.
4261 : */
4262 2372 : oldcontext = MemoryContextSwitchTo(aggcontext);
4263 2372 : state = makeStringInfo();
4264 2372 : MemoryContextSwitchTo(oldcontext);
4265 :
4266 2372 : return state;
4267 : }
4268 :
4269 : Datum
4270 1094368 : string_agg_transfn(PG_FUNCTION_ARGS)
4271 : {
4272 : StringInfo state;
4273 :
4274 1094368 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4275 :
4276 : /* Append the value unless null, preceding it with the delimiter. */
4277 1094368 : if (!PG_ARGISNULL(1))
4278 : {
4279 1079320 : text *value = PG_GETARG_TEXT_PP(1);
4280 1079320 : bool isfirst = false;
4281 :
4282 : /*
4283 : * You might think we can just throw away the first delimiter, however
4284 : * we must keep it as we may be a parallel worker doing partial
4285 : * aggregation building a state to send to the main process. We need
4286 : * to keep the delimiter of every aggregation so that the combine
4287 : * function can properly join up the strings of two separately
4288 : * partially aggregated results. The first delimiter is only stripped
4289 : * off in the final function. To know how much to strip off the front
4290 : * of the string, we store the length of the first delimiter in the
4291 : * StringInfo's cursor field, which we don't otherwise need here.
4292 : */
4293 1079320 : if (state == NULL)
4294 : {
4295 2052 : state = makeStringAggState(fcinfo);
4296 2052 : isfirst = true;
4297 : }
4298 :
4299 1079320 : if (!PG_ARGISNULL(2))
4300 : {
4301 1079320 : text *delim = PG_GETARG_TEXT_PP(2);
4302 :
4303 1079320 : appendStringInfoText(state, delim);
4304 1079320 : if (isfirst)
4305 2052 : state->cursor = VARSIZE_ANY_EXHDR(delim);
4306 : }
4307 :
4308 1079320 : appendStringInfoText(state, value);
4309 : }
4310 :
4311 : /*
4312 : * The transition type for string_agg() is declared to be "internal",
4313 : * which is a pass-by-value type the same size as a pointer.
4314 : */
4315 1094368 : if (state)
4316 1094282 : PG_RETURN_POINTER(state);
4317 86 : PG_RETURN_NULL();
4318 : }
4319 :
4320 : /*
4321 : * string_agg_combine
4322 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
4323 : */
4324 : Datum
4325 200 : string_agg_combine(PG_FUNCTION_ARGS)
4326 : {
4327 : StringInfo state1;
4328 : StringInfo state2;
4329 : MemoryContext agg_context;
4330 :
4331 200 : if (!AggCheckCallContext(fcinfo, &agg_context))
4332 0 : elog(ERROR, "aggregate function called in non-aggregate context");
4333 :
4334 200 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4335 200 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
4336 :
4337 200 : if (state2 == NULL)
4338 : {
4339 : /*
4340 : * NULL state2 is easy, just return state1, which we know is already
4341 : * in the agg_context
4342 : */
4343 0 : if (state1 == NULL)
4344 0 : PG_RETURN_NULL();
4345 0 : PG_RETURN_POINTER(state1);
4346 : }
4347 :
4348 200 : if (state1 == NULL)
4349 : {
4350 : /* We must copy state2's data into the agg_context */
4351 : MemoryContext old_context;
4352 :
4353 120 : old_context = MemoryContextSwitchTo(agg_context);
4354 120 : state1 = makeStringAggState(fcinfo);
4355 120 : appendBinaryStringInfo(state1, state2->data, state2->len);
4356 120 : state1->cursor = state2->cursor;
4357 120 : MemoryContextSwitchTo(old_context);
4358 : }
4359 80 : else if (state2->len > 0)
4360 : {
4361 : /* Combine ... state1->cursor does not change in this case */
4362 80 : appendBinaryStringInfo(state1, state2->data, state2->len);
4363 : }
4364 :
4365 200 : PG_RETURN_POINTER(state1);
4366 : }
4367 :
4368 : /*
4369 : * string_agg_serialize
4370 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
4371 : *
4372 : * This is strict, so we need not handle NULL input
4373 : */
4374 : Datum
4375 200 : string_agg_serialize(PG_FUNCTION_ARGS)
4376 : {
4377 : StringInfo state;
4378 : StringInfoData buf;
4379 : bytea *result;
4380 :
4381 : /* cannot be called directly because of internal-type argument */
4382 : Assert(AggCheckCallContext(fcinfo, NULL));
4383 :
4384 200 : state = (StringInfo) PG_GETARG_POINTER(0);
4385 :
4386 200 : pq_begintypsend(&buf);
4387 :
4388 : /* cursor */
4389 200 : pq_sendint(&buf, state->cursor, 4);
4390 :
4391 : /* data */
4392 200 : pq_sendbytes(&buf, state->data, state->len);
4393 :
4394 200 : result = pq_endtypsend(&buf);
4395 :
4396 200 : PG_RETURN_BYTEA_P(result);
4397 : }
4398 :
4399 : /*
4400 : * string_agg_deserialize
4401 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
4402 : *
4403 : * This is strict, so we need not handle NULL input
4404 : */
4405 : Datum
4406 200 : string_agg_deserialize(PG_FUNCTION_ARGS)
4407 : {
4408 : bytea *sstate;
4409 : StringInfo result;
4410 : StringInfoData buf;
4411 : char *data;
4412 : int datalen;
4413 :
4414 : /* cannot be called directly because of internal-type argument */
4415 : Assert(AggCheckCallContext(fcinfo, NULL));
4416 :
4417 200 : sstate = PG_GETARG_BYTEA_PP(0);
4418 :
4419 : /*
4420 : * Initialize a StringInfo so that we can "receive" it using the standard
4421 : * recv-function infrastructure.
4422 : */
4423 200 : initReadOnlyStringInfo(&buf, VARDATA_ANY(sstate),
4424 200 : VARSIZE_ANY_EXHDR(sstate));
4425 :
4426 200 : result = makeStringAggState(fcinfo);
4427 :
4428 : /* cursor */
4429 200 : result->cursor = pq_getmsgint(&buf, 4);
4430 :
4431 : /* data */
4432 200 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
4433 200 : data = (char *) pq_getmsgbytes(&buf, datalen);
4434 200 : appendBinaryStringInfo(result, data, datalen);
4435 :
4436 200 : pq_getmsgend(&buf);
4437 :
4438 200 : PG_RETURN_POINTER(result);
4439 : }
4440 :
4441 : Datum
4442 2096 : string_agg_finalfn(PG_FUNCTION_ARGS)
4443 : {
4444 : StringInfo state;
4445 :
4446 : /* cannot be called directly because of internal-type argument */
4447 : Assert(AggCheckCallContext(fcinfo, NULL));
4448 :
4449 2096 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4450 :
4451 2096 : if (state != NULL)
4452 : {
4453 : /* As per comment in transfn, strip data before the cursor position */
4454 2012 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
4455 : state->len - state->cursor));
4456 : }
4457 : else
4458 84 : PG_RETURN_NULL();
4459 : }
4460 :
4461 : /*
4462 : * Prepare cache with fmgr info for the output functions of the datatypes of
4463 : * the arguments of a concat-like function, beginning with argument "argidx".
4464 : * (Arguments before that will have corresponding slots in the resulting
4465 : * FmgrInfo array, but we don't fill those slots.)
4466 : */
4467 : static FmgrInfo *
4468 106 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
4469 : {
4470 : FmgrInfo *foutcache;
4471 : int i;
4472 :
4473 : /* We keep the info in fn_mcxt so it survives across calls */
4474 106 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4475 106 : PG_NARGS() * sizeof(FmgrInfo));
4476 :
4477 400 : for (i = argidx; i < PG_NARGS(); i++)
4478 : {
4479 : Oid valtype;
4480 : Oid typOutput;
4481 : bool typIsVarlena;
4482 :
4483 294 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4484 294 : if (!OidIsValid(valtype))
4485 0 : elog(ERROR, "could not determine data type of concat() input");
4486 :
4487 294 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4488 294 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
4489 : }
4490 :
4491 106 : fcinfo->flinfo->fn_extra = foutcache;
4492 :
4493 106 : return foutcache;
4494 : }
4495 :
4496 : /*
4497 : * Implementation of both concat() and concat_ws().
4498 : *
4499 : * sepstr is the separator string to place between values.
4500 : * argidx identifies the first argument to concatenate (counting from zero);
4501 : * note that this must be constant across any one series of calls.
4502 : *
4503 : * Returns NULL if result should be NULL, else text value.
4504 : */
4505 : static text *
4506 264 : concat_internal(const char *sepstr, int argidx,
4507 : FunctionCallInfo fcinfo)
4508 : {
4509 : text *result;
4510 : StringInfoData str;
4511 : FmgrInfo *foutcache;
4512 264 : bool first_arg = true;
4513 : int i;
4514 :
4515 : /*
4516 : * concat(VARIADIC some-array) is essentially equivalent to
4517 : * array_to_text(), ie concat the array elements with the given separator.
4518 : * So we just pass the case off to that code.
4519 : */
4520 264 : if (get_fn_expr_variadic(fcinfo->flinfo))
4521 : {
4522 : ArrayType *arr;
4523 :
4524 : /* Should have just the one argument */
4525 : Assert(argidx == PG_NARGS() - 1);
4526 :
4527 : /* concat(VARIADIC NULL) is defined as NULL */
4528 30 : if (PG_ARGISNULL(argidx))
4529 12 : return NULL;
4530 :
4531 : /*
4532 : * Non-null argument had better be an array. We assume that any call
4533 : * context that could let get_fn_expr_variadic return true will have
4534 : * checked that a VARIADIC-labeled parameter actually is an array. So
4535 : * it should be okay to just Assert that it's an array rather than
4536 : * doing a full-fledged error check.
4537 : */
4538 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
4539 :
4540 : /* OK, safe to fetch the array value */
4541 18 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
4542 :
4543 : /*
4544 : * And serialize the array. We tell array_to_text to ignore null
4545 : * elements, which matches the behavior of the loop below.
4546 : */
4547 18 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4548 : }
4549 :
4550 : /* Normal case without explicit VARIADIC marker */
4551 234 : initStringInfo(&str);
4552 :
4553 : /* Get output function info, building it if first time through */
4554 234 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
4555 234 : if (foutcache == NULL)
4556 106 : foutcache = build_concat_foutcache(fcinfo, argidx);
4557 :
4558 822 : for (i = argidx; i < PG_NARGS(); i++)
4559 : {
4560 588 : if (!PG_ARGISNULL(i))
4561 : {
4562 510 : Datum value = PG_GETARG_DATUM(i);
4563 :
4564 : /* add separator if appropriate */
4565 510 : if (first_arg)
4566 228 : first_arg = false;
4567 : else
4568 282 : appendStringInfoString(&str, sepstr);
4569 :
4570 : /* call the appropriate type output function, append the result */
4571 510 : appendStringInfoString(&str,
4572 510 : OutputFunctionCall(&foutcache[i], value));
4573 : }
4574 : }
4575 :
4576 234 : result = cstring_to_text_with_len(str.data, str.len);
4577 234 : pfree(str.data);
4578 :
4579 234 : return result;
4580 : }
4581 :
4582 : /*
4583 : * Concatenate all arguments. NULL arguments are ignored.
4584 : */
4585 : Datum
4586 186 : text_concat(PG_FUNCTION_ARGS)
4587 : {
4588 : text *result;
4589 :
4590 186 : result = concat_internal("", 0, fcinfo);
4591 186 : if (result == NULL)
4592 6 : PG_RETURN_NULL();
4593 180 : PG_RETURN_TEXT_P(result);
4594 : }
4595 :
4596 : /*
4597 : * Concatenate all but first argument value with separators. The first
4598 : * parameter is used as the separator. NULL arguments are ignored.
4599 : */
4600 : Datum
4601 84 : text_concat_ws(PG_FUNCTION_ARGS)
4602 : {
4603 : char *sep;
4604 : text *result;
4605 :
4606 : /* return NULL when separator is NULL */
4607 84 : if (PG_ARGISNULL(0))
4608 6 : PG_RETURN_NULL();
4609 78 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
4610 :
4611 78 : result = concat_internal(sep, 1, fcinfo);
4612 78 : if (result == NULL)
4613 6 : PG_RETURN_NULL();
4614 72 : PG_RETURN_TEXT_P(result);
4615 : }
4616 :
4617 : /*
4618 : * Return first n characters in the string. When n is negative,
4619 : * return all but last |n| characters.
4620 : */
4621 : Datum
4622 2148 : text_left(PG_FUNCTION_ARGS)
4623 : {
4624 2148 : int n = PG_GETARG_INT32(1);
4625 :
4626 2148 : if (n < 0)
4627 : {
4628 30 : text *str = PG_GETARG_TEXT_PP(0);
4629 30 : const char *p = VARDATA_ANY(str);
4630 30 : int len = VARSIZE_ANY_EXHDR(str);
4631 : int rlen;
4632 :
4633 30 : n = pg_mbstrlen_with_len(p, len) + n;
4634 30 : rlen = pg_mbcharcliplen(p, len, n);
4635 30 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
4636 : }
4637 : else
4638 2118 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
4639 : }
4640 :
4641 : /*
4642 : * Return last n characters in the string. When n is negative,
4643 : * return all but first |n| characters.
4644 : */
4645 : Datum
4646 66 : text_right(PG_FUNCTION_ARGS)
4647 : {
4648 66 : text *str = PG_GETARG_TEXT_PP(0);
4649 66 : const char *p = VARDATA_ANY(str);
4650 66 : int len = VARSIZE_ANY_EXHDR(str);
4651 66 : int n = PG_GETARG_INT32(1);
4652 : int off;
4653 :
4654 66 : if (n < 0)
4655 30 : n = -n;
4656 : else
4657 36 : n = pg_mbstrlen_with_len(p, len) - n;
4658 66 : off = pg_mbcharcliplen(p, len, n);
4659 :
4660 66 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4661 : }
4662 :
4663 : /*
4664 : * Return reversed string
4665 : */
4666 : Datum
4667 6 : text_reverse(PG_FUNCTION_ARGS)
4668 : {
4669 6 : text *str = PG_GETARG_TEXT_PP(0);
4670 6 : const char *p = VARDATA_ANY(str);
4671 6 : int len = VARSIZE_ANY_EXHDR(str);
4672 6 : const char *endp = p + len;
4673 : text *result;
4674 : char *dst;
4675 :
4676 6 : result = palloc(len + VARHDRSZ);
4677 6 : dst = (char *) VARDATA(result) + len;
4678 6 : SET_VARSIZE(result, len + VARHDRSZ);
4679 :
4680 6 : if (pg_database_encoding_max_length() > 1)
4681 : {
4682 : /* multibyte version */
4683 36 : while (p < endp)
4684 : {
4685 : int sz;
4686 :
4687 30 : sz = pg_mblen(p);
4688 30 : dst -= sz;
4689 30 : memcpy(dst, p, sz);
4690 30 : p += sz;
4691 : }
4692 : }
4693 : else
4694 : {
4695 : /* single byte version */
4696 0 : while (p < endp)
4697 0 : *(--dst) = *p++;
4698 : }
4699 :
4700 6 : PG_RETURN_TEXT_P(result);
4701 : }
4702 :
4703 :
4704 : /*
4705 : * Support macros for text_format()
4706 : */
4707 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4708 :
4709 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4710 : do { \
4711 : if (++(ptr) >= (end_ptr)) \
4712 : ereport(ERROR, \
4713 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4714 : errmsg("unterminated format() type specifier"), \
4715 : errhint("For a single \"%%\" use \"%%%%\"."))); \
4716 : } while (0)
4717 :
4718 : /*
4719 : * Returns a formatted string
4720 : */
4721 : Datum
4722 33232 : text_format(PG_FUNCTION_ARGS)
4723 : {
4724 : text *fmt;
4725 : StringInfoData str;
4726 : const char *cp;
4727 : const char *start_ptr;
4728 : const char *end_ptr;
4729 : text *result;
4730 : int arg;
4731 : bool funcvariadic;
4732 : int nargs;
4733 33232 : Datum *elements = NULL;
4734 33232 : bool *nulls = NULL;
4735 33232 : Oid element_type = InvalidOid;
4736 33232 : Oid prev_type = InvalidOid;
4737 33232 : Oid prev_width_type = InvalidOid;
4738 : FmgrInfo typoutputfinfo;
4739 : FmgrInfo typoutputinfo_width;
4740 :
4741 : /* When format string is null, immediately return null */
4742 33232 : if (PG_ARGISNULL(0))
4743 6 : PG_RETURN_NULL();
4744 :
4745 : /* If argument is marked VARIADIC, expand array into elements */
4746 33226 : if (get_fn_expr_variadic(fcinfo->flinfo))
4747 : {
4748 : ArrayType *arr;
4749 : int16 elmlen;
4750 : bool elmbyval;
4751 : char elmalign;
4752 : int nitems;
4753 :
4754 : /* Should have just the one argument */
4755 : Assert(PG_NARGS() == 2);
4756 :
4757 : /* If argument is NULL, we treat it as zero-length array */
4758 48 : if (PG_ARGISNULL(1))
4759 6 : nitems = 0;
4760 : else
4761 : {
4762 : /*
4763 : * Non-null argument had better be an array. We assume that any
4764 : * call context that could let get_fn_expr_variadic return true
4765 : * will have checked that a VARIADIC-labeled parameter actually is
4766 : * an array. So it should be okay to just Assert that it's an
4767 : * array rather than doing a full-fledged error check.
4768 : */
4769 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
4770 :
4771 : /* OK, safe to fetch the array value */
4772 42 : arr = PG_GETARG_ARRAYTYPE_P(1);
4773 :
4774 : /* Get info about array element type */
4775 42 : element_type = ARR_ELEMTYPE(arr);
4776 42 : get_typlenbyvalalign(element_type,
4777 : &elmlen, &elmbyval, &elmalign);
4778 :
4779 : /* Extract all array elements */
4780 42 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
4781 : &elements, &nulls, &nitems);
4782 : }
4783 :
4784 48 : nargs = nitems + 1;
4785 48 : funcvariadic = true;
4786 : }
4787 : else
4788 : {
4789 : /* Non-variadic case, we'll process the arguments individually */
4790 33178 : nargs = PG_NARGS();
4791 33178 : funcvariadic = false;
4792 : }
4793 :
4794 : /* Setup for main loop. */
4795 33226 : fmt = PG_GETARG_TEXT_PP(0);
4796 33226 : start_ptr = VARDATA_ANY(fmt);
4797 33226 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
4798 33226 : initStringInfo(&str);
4799 33226 : arg = 1; /* next argument position to print */
4800 :
4801 : /* Scan format string, looking for conversion specifiers. */
4802 1013520 : for (cp = start_ptr; cp < end_ptr; cp++)
4803 : {
4804 : int argpos;
4805 : int widthpos;
4806 : int flags;
4807 : int width;
4808 : Datum value;
4809 : bool isNull;
4810 : Oid typid;
4811 :
4812 : /*
4813 : * If it's not the start of a conversion specifier, just copy it to
4814 : * the output buffer.
4815 : */
4816 980354 : if (*cp != '%')
4817 : {
4818 914470 : appendStringInfoCharMacro(&str, *cp);
4819 914488 : continue;
4820 : }
4821 :
4822 65884 : ADVANCE_PARSE_POINTER(cp, end_ptr);
4823 :
4824 : /* Easy case: %% outputs a single % */
4825 65884 : if (*cp == '%')
4826 : {
4827 18 : appendStringInfoCharMacro(&str, *cp);
4828 18 : continue;
4829 : }
4830 :
4831 : /* Parse the optional portions of the format specifier */
4832 65866 : cp = text_format_parse_format(cp, end_ptr,
4833 : &argpos, &widthpos,
4834 : &flags, &width);
4835 :
4836 : /*
4837 : * Next we should see the main conversion specifier. Whether or not
4838 : * an argument position was present, it's known that at least one
4839 : * character remains in the string at this point. Experience suggests
4840 : * that it's worth checking that that character is one of the expected
4841 : * ones before we try to fetch arguments, so as to produce the least
4842 : * confusing response to a mis-formatted specifier.
4843 : */
4844 65842 : if (strchr("sIL", *cp) == NULL)
4845 6 : ereport(ERROR,
4846 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4847 : errmsg("unrecognized format() type specifier \"%.*s\"",
4848 : pg_mblen(cp), cp),
4849 : errhint("For a single \"%%\" use \"%%%%\".")));
4850 :
4851 : /* If indirect width was specified, get its value */
4852 65836 : if (widthpos >= 0)
4853 : {
4854 : /* Collect the specified or next argument position */
4855 42 : if (widthpos > 0)
4856 36 : arg = widthpos;
4857 42 : if (arg >= nargs)
4858 0 : ereport(ERROR,
4859 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4860 : errmsg("too few arguments for format()")));
4861 :
4862 : /* Get the value and type of the selected argument */
4863 42 : if (!funcvariadic)
4864 : {
4865 42 : value = PG_GETARG_DATUM(arg);
4866 42 : isNull = PG_ARGISNULL(arg);
4867 42 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4868 : }
4869 : else
4870 : {
4871 0 : value = elements[arg - 1];
4872 0 : isNull = nulls[arg - 1];
4873 0 : typid = element_type;
4874 : }
4875 42 : if (!OidIsValid(typid))
4876 0 : elog(ERROR, "could not determine data type of format() input");
4877 :
4878 42 : arg++;
4879 :
4880 : /* We can treat NULL width the same as zero */
4881 42 : if (isNull)
4882 6 : width = 0;
4883 36 : else if (typid == INT4OID)
4884 36 : width = DatumGetInt32(value);
4885 0 : else if (typid == INT2OID)
4886 0 : width = DatumGetInt16(value);
4887 : else
4888 : {
4889 : /* For less-usual datatypes, convert to text then to int */
4890 : char *str;
4891 :
4892 0 : if (typid != prev_width_type)
4893 : {
4894 : Oid typoutputfunc;
4895 : bool typIsVarlena;
4896 :
4897 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4898 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
4899 0 : prev_width_type = typid;
4900 : }
4901 :
4902 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
4903 :
4904 : /* pg_strtoint32 will complain about bad data or overflow */
4905 0 : width = pg_strtoint32(str);
4906 :
4907 0 : pfree(str);
4908 : }
4909 : }
4910 :
4911 : /* Collect the specified or next argument position */
4912 65836 : if (argpos > 0)
4913 132 : arg = argpos;
4914 65836 : if (arg >= nargs)
4915 24 : ereport(ERROR,
4916 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4917 : errmsg("too few arguments for format()")));
4918 :
4919 : /* Get the value and type of the selected argument */
4920 65812 : if (!funcvariadic)
4921 : {
4922 64540 : value = PG_GETARG_DATUM(arg);
4923 64540 : isNull = PG_ARGISNULL(arg);
4924 64540 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4925 : }
4926 : else
4927 : {
4928 1272 : value = elements[arg - 1];
4929 1272 : isNull = nulls[arg - 1];
4930 1272 : typid = element_type;
4931 : }
4932 65812 : if (!OidIsValid(typid))
4933 0 : elog(ERROR, "could not determine data type of format() input");
4934 :
4935 65812 : arg++;
4936 :
4937 : /*
4938 : * Get the appropriate typOutput function, reusing previous one if
4939 : * same type as previous argument. That's particularly useful in the
4940 : * variadic-array case, but often saves work even for ordinary calls.
4941 : */
4942 65812 : if (typid != prev_type)
4943 : {
4944 : Oid typoutputfunc;
4945 : bool typIsVarlena;
4946 :
4947 34294 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4948 34294 : fmgr_info(typoutputfunc, &typoutputfinfo);
4949 34294 : prev_type = typid;
4950 : }
4951 :
4952 : /*
4953 : * And now we can format the value.
4954 : */
4955 65812 : switch (*cp)
4956 : {
4957 65812 : case 's':
4958 : case 'I':
4959 : case 'L':
4960 65812 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
4961 : value, isNull,
4962 : flags, width);
4963 65806 : break;
4964 0 : default:
4965 : /* should not get here, because of previous check */
4966 0 : ereport(ERROR,
4967 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4968 : errmsg("unrecognized format() type specifier \"%.*s\"",
4969 : pg_mblen(cp), cp),
4970 : errhint("For a single \"%%\" use \"%%%%\".")));
4971 : break;
4972 : }
4973 : }
4974 :
4975 : /* Don't need deconstruct_array results anymore. */
4976 33166 : if (elements != NULL)
4977 42 : pfree(elements);
4978 33166 : if (nulls != NULL)
4979 42 : pfree(nulls);
4980 :
4981 : /* Generate results. */
4982 33166 : result = cstring_to_text_with_len(str.data, str.len);
4983 33166 : pfree(str.data);
4984 :
4985 33166 : PG_RETURN_TEXT_P(result);
4986 : }
4987 :
4988 : /*
4989 : * Parse contiguous digits as a decimal number.
4990 : *
4991 : * Returns true if some digits could be parsed.
4992 : * The value is returned into *value, and *ptr is advanced to the next
4993 : * character to be parsed.
4994 : *
4995 : * Note parsing invariant: at least one character is known available before
4996 : * string end (end_ptr) at entry, and this is still true at exit.
4997 : */
4998 : static bool
4999 131696 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5000 : {
5001 131696 : bool found = false;
5002 131696 : const char *cp = *ptr;
5003 131696 : int val = 0;
5004 :
5005 132008 : while (*cp >= '0' && *cp <= '9')
5006 : {
5007 318 : int8 digit = (*cp - '0');
5008 :
5009 318 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5010 318 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5011 0 : ereport(ERROR,
5012 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5013 : errmsg("number is out of range")));
5014 318 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5015 312 : found = true;
5016 : }
5017 :
5018 131690 : *ptr = cp;
5019 131690 : *value = val;
5020 :
5021 131690 : return found;
5022 : }
5023 :
5024 : /*
5025 : * Parse a format specifier (generally following the SUS printf spec).
5026 : *
5027 : * We have already advanced over the initial '%', and we are looking for
5028 : * [argpos][flags][width]type (but the type character is not consumed here).
5029 : *
5030 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5031 : * Output parameters:
5032 : * argpos: argument position for value to be printed. -1 means unspecified.
5033 : * widthpos: argument position for width. Zero means the argument position
5034 : * was unspecified (ie, take the next arg) and -1 means no width
5035 : * argument (width was omitted or specified as a constant).
5036 : * flags: bitmask of flags.
5037 : * width: directly-specified width value. Zero means the width was omitted
5038 : * (note it's not necessary to distinguish this case from an explicit
5039 : * zero width value).
5040 : *
5041 : * The function result is the next character position to be parsed, ie, the
5042 : * location where the type character is/should be.
5043 : *
5044 : * Note parsing invariant: at least one character is known available before
5045 : * string end (end_ptr) at entry, and this is still true at exit.
5046 : */
5047 : static const char *
5048 65866 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5049 : int *argpos, int *widthpos,
5050 : int *flags, int *width)
5051 : {
5052 65866 : const char *cp = start_ptr;
5053 : int n;
5054 :
5055 : /* set defaults for output parameters */
5056 65866 : *argpos = -1;
5057 65866 : *widthpos = -1;
5058 65866 : *flags = 0;
5059 65866 : *width = 0;
5060 :
5061 : /* try to identify first number */
5062 65866 : if (text_format_parse_digits(&cp, end_ptr, &n))
5063 : {
5064 174 : if (*cp != '$')
5065 : {
5066 : /* Must be just a width and a type, so we're done */
5067 24 : *width = n;
5068 24 : return cp;
5069 : }
5070 : /* The number was argument position */
5071 150 : *argpos = n;
5072 : /* Explicit 0 for argument index is immediately refused */
5073 150 : if (n == 0)
5074 6 : ereport(ERROR,
5075 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5076 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5077 144 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5078 : }
5079 :
5080 : /* Handle flags (only minus is supported now) */
5081 65860 : while (*cp == '-')
5082 : {
5083 30 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5084 30 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5085 : }
5086 :
5087 65830 : if (*cp == '*')
5088 : {
5089 : /* Handle indirect width */
5090 48 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5091 48 : if (text_format_parse_digits(&cp, end_ptr, &n))
5092 : {
5093 : /* number in this position must be closed by $ */
5094 42 : if (*cp != '$')
5095 0 : ereport(ERROR,
5096 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5097 : errmsg("width argument position must be ended by \"$\"")));
5098 : /* The number was width argument position */
5099 42 : *widthpos = n;
5100 : /* Explicit 0 for argument index is immediately refused */
5101 42 : if (n == 0)
5102 6 : ereport(ERROR,
5103 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5104 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5105 36 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5106 : }
5107 : else
5108 6 : *widthpos = 0; /* width's argument position is unspecified */
5109 : }
5110 : else
5111 : {
5112 : /* Check for direct width specification */
5113 65782 : if (text_format_parse_digits(&cp, end_ptr, &n))
5114 30 : *width = n;
5115 : }
5116 :
5117 : /* cp should now be pointing at type character */
5118 65818 : return cp;
5119 : }
5120 :
5121 : /*
5122 : * Format a %s, %I, or %L conversion
5123 : */
5124 : static void
5125 65812 : text_format_string_conversion(StringInfo buf, char conversion,
5126 : FmgrInfo *typOutputInfo,
5127 : Datum value, bool isNull,
5128 : int flags, int width)
5129 : {
5130 : char *str;
5131 :
5132 : /* Handle NULL arguments before trying to stringify the value. */
5133 65812 : if (isNull)
5134 : {
5135 342 : if (conversion == 's')
5136 270 : text_format_append_string(buf, "", flags, width);
5137 72 : else if (conversion == 'L')
5138 66 : text_format_append_string(buf, "NULL", flags, width);
5139 6 : else if (conversion == 'I')
5140 6 : ereport(ERROR,
5141 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5142 : errmsg("null values cannot be formatted as an SQL identifier")));
5143 336 : return;
5144 : }
5145 :
5146 : /* Stringify. */
5147 65470 : str = OutputFunctionCall(typOutputInfo, value);
5148 :
5149 : /* Escape. */
5150 65470 : if (conversion == 'I')
5151 : {
5152 : /* quote_identifier may or may not allocate a new string. */
5153 4906 : text_format_append_string(buf, quote_identifier(str), flags, width);
5154 : }
5155 60564 : else if (conversion == 'L')
5156 : {
5157 3252 : char *qstr = quote_literal_cstr(str);
5158 :
5159 3252 : text_format_append_string(buf, qstr, flags, width);
5160 : /* quote_literal_cstr() always allocates a new string */
5161 3252 : pfree(qstr);
5162 : }
5163 : else
5164 57312 : text_format_append_string(buf, str, flags, width);
5165 :
5166 : /* Cleanup. */
5167 65470 : pfree(str);
5168 : }
5169 :
5170 : /*
5171 : * Append str to buf, padding as directed by flags/width
5172 : */
5173 : static void
5174 65806 : text_format_append_string(StringInfo buf, const char *str,
5175 : int flags, int width)
5176 : {
5177 65806 : bool align_to_left = false;
5178 : int len;
5179 :
5180 : /* fast path for typical easy case */
5181 65806 : if (width == 0)
5182 : {
5183 65722 : appendStringInfoString(buf, str);
5184 65722 : return;
5185 : }
5186 :
5187 84 : if (width < 0)
5188 : {
5189 : /* Negative width: implicit '-' flag, then take absolute value */
5190 6 : align_to_left = true;
5191 : /* -INT_MIN is undefined */
5192 6 : if (width <= INT_MIN)
5193 0 : ereport(ERROR,
5194 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5195 : errmsg("number is out of range")));
5196 6 : width = -width;
5197 : }
5198 78 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
5199 24 : align_to_left = true;
5200 :
5201 84 : len = pg_mbstrlen(str);
5202 84 : if (align_to_left)
5203 : {
5204 : /* left justify */
5205 30 : appendStringInfoString(buf, str);
5206 30 : if (len < width)
5207 30 : appendStringInfoSpaces(buf, width - len);
5208 : }
5209 : else
5210 : {
5211 : /* right justify */
5212 54 : if (len < width)
5213 54 : appendStringInfoSpaces(buf, width - len);
5214 54 : appendStringInfoString(buf, str);
5215 : }
5216 : }
5217 :
5218 : /*
5219 : * text_format_nv - nonvariadic wrapper for text_format function.
5220 : *
5221 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5222 : * which checks that all built-in functions that share the implementing C
5223 : * function take the same number of arguments.
5224 : */
5225 : Datum
5226 3810 : text_format_nv(PG_FUNCTION_ARGS)
5227 : {
5228 3810 : return text_format(fcinfo);
5229 : }
5230 :
5231 : /*
5232 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
5233 : * for this use case.
5234 : */
5235 : static inline bool
5236 0 : rest_of_char_same(const char *s1, const char *s2, int len)
5237 : {
5238 0 : while (len > 0)
5239 : {
5240 0 : len--;
5241 0 : if (s1[len] != s2[len])
5242 0 : return false;
5243 : }
5244 0 : return true;
5245 : }
5246 :
5247 : /* Expand each Levenshtein distance variant */
5248 : #include "levenshtein.c"
5249 : #define LEVENSHTEIN_LESS_EQUAL
5250 : #include "levenshtein.c"
5251 :
5252 :
5253 : /*
5254 : * The following *ClosestMatch() functions can be used to determine whether a
5255 : * user-provided string resembles any known valid values, which is useful for
5256 : * providing hints in log messages, among other things. Use these functions
5257 : * like so:
5258 : *
5259 : * initClosestMatch(&state, source_string, max_distance);
5260 : *
5261 : * for (int i = 0; i < num_valid_strings; i++)
5262 : * updateClosestMatch(&state, valid_strings[i]);
5263 : *
5264 : * closestMatch = getClosestMatch(&state);
5265 : */
5266 :
5267 : /*
5268 : * Initialize the given state with the source string and maximum Levenshtein
5269 : * distance to consider.
5270 : */
5271 : void
5272 78 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
5273 : {
5274 : Assert(state);
5275 : Assert(max_d >= 0);
5276 :
5277 78 : state->source = source;
5278 78 : state->min_d = -1;
5279 78 : state->max_d = max_d;
5280 78 : state->match = NULL;
5281 78 : }
5282 :
5283 : /*
5284 : * If the candidate string is a closer match than the current one saved (or
5285 : * there is no match saved), save it as the closest match.
5286 : *
5287 : * If the source or candidate string is NULL, empty, or too long, this function
5288 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
5289 : * allowed or more than half the characters are different, no action is taken.
5290 : */
5291 : void
5292 804 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
5293 : {
5294 : int dist;
5295 :
5296 : Assert(state);
5297 :
5298 804 : if (state->source == NULL || state->source[0] == '\0' ||
5299 804 : candidate == NULL || candidate[0] == '\0')
5300 0 : return;
5301 :
5302 : /*
5303 : * To avoid ERROR-ing, we check the lengths here instead of setting
5304 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
5305 : */
5306 804 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
5307 804 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
5308 0 : return;
5309 :
5310 804 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
5311 804 : candidate, strlen(candidate), 1, 1, 1,
5312 : state->max_d, true);
5313 804 : if (dist <= state->max_d &&
5314 62 : dist <= strlen(state->source) / 2 &&
5315 14 : (state->min_d == -1 || dist < state->min_d))
5316 : {
5317 14 : state->min_d = dist;
5318 14 : state->match = candidate;
5319 : }
5320 : }
5321 :
5322 : /*
5323 : * Return the closest match. If no suitable candidates were provided via
5324 : * updateClosestMatch(), return NULL.
5325 : */
5326 : const char *
5327 78 : getClosestMatch(ClosestMatchState *state)
5328 : {
5329 : Assert(state);
5330 :
5331 78 : return state->match;
5332 : }
5333 :
5334 :
5335 : /*
5336 : * Unicode support
5337 : */
5338 :
5339 : static UnicodeNormalizationForm
5340 210 : unicode_norm_form_from_string(const char *formstr)
5341 : {
5342 210 : UnicodeNormalizationForm form = -1;
5343 :
5344 : /*
5345 : * Might as well check this while we're here.
5346 : */
5347 210 : if (GetDatabaseEncoding() != PG_UTF8)
5348 0 : ereport(ERROR,
5349 : (errcode(ERRCODE_SYNTAX_ERROR),
5350 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
5351 :
5352 210 : if (pg_strcasecmp(formstr, "NFC") == 0)
5353 66 : form = UNICODE_NFC;
5354 144 : else if (pg_strcasecmp(formstr, "NFD") == 0)
5355 60 : form = UNICODE_NFD;
5356 84 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
5357 36 : form = UNICODE_NFKC;
5358 48 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
5359 36 : form = UNICODE_NFKD;
5360 : else
5361 12 : ereport(ERROR,
5362 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5363 : errmsg("invalid normalization form: %s", formstr)));
5364 :
5365 198 : return form;
5366 : }
5367 :
5368 : /*
5369 : * Returns version of Unicode used by Postgres in "major.minor" format (the
5370 : * same format as the Unicode version reported by ICU). The third component
5371 : * ("update version") never involves additions to the character repertoire and
5372 : * is unimportant for most purposes.
5373 : *
5374 : * See: https://unicode.org/versions/
5375 : */
5376 : Datum
5377 34 : unicode_version(PG_FUNCTION_ARGS)
5378 : {
5379 34 : PG_RETURN_TEXT_P(cstring_to_text(PG_UNICODE_VERSION));
5380 : }
5381 :
5382 : /*
5383 : * Returns version of Unicode used by ICU, if enabled; otherwise NULL.
5384 : */
5385 : Datum
5386 2 : icu_unicode_version(PG_FUNCTION_ARGS)
5387 : {
5388 2 : const char *version = pg_icu_unicode_version();
5389 :
5390 2 : if (version)
5391 2 : PG_RETURN_TEXT_P(cstring_to_text(version));
5392 : else
5393 0 : PG_RETURN_NULL();
5394 : }
5395 :
5396 : /*
5397 : * Check whether the string contains only assigned Unicode code
5398 : * points. Requires that the database encoding is UTF-8.
5399 : */
5400 : Datum
5401 12 : unicode_assigned(PG_FUNCTION_ARGS)
5402 : {
5403 12 : text *input = PG_GETARG_TEXT_PP(0);
5404 : unsigned char *p;
5405 : int size;
5406 :
5407 12 : if (GetDatabaseEncoding() != PG_UTF8)
5408 0 : ereport(ERROR,
5409 : (errmsg("Unicode categorization can only be performed if server encoding is UTF8")));
5410 :
5411 : /* convert to char32_t */
5412 12 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5413 12 : p = (unsigned char *) VARDATA_ANY(input);
5414 48 : for (int i = 0; i < size; i++)
5415 : {
5416 42 : char32_t uchar = utf8_to_unicode(p);
5417 42 : int category = unicode_category(uchar);
5418 :
5419 42 : if (category == PG_U_UNASSIGNED)
5420 6 : PG_RETURN_BOOL(false);
5421 :
5422 36 : p += pg_utf_mblen(p);
5423 : }
5424 :
5425 6 : PG_RETURN_BOOL(true);
5426 : }
5427 :
5428 : Datum
5429 72 : unicode_normalize_func(PG_FUNCTION_ARGS)
5430 : {
5431 72 : text *input = PG_GETARG_TEXT_PP(0);
5432 72 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5433 : UnicodeNormalizationForm form;
5434 : int size;
5435 : char32_t *input_chars;
5436 : char32_t *output_chars;
5437 : unsigned char *p;
5438 : text *result;
5439 : int i;
5440 :
5441 72 : form = unicode_norm_form_from_string(formstr);
5442 :
5443 : /* convert to char32_t */
5444 66 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5445 66 : input_chars = palloc((size + 1) * sizeof(char32_t));
5446 66 : p = (unsigned char *) VARDATA_ANY(input);
5447 288 : for (i = 0; i < size; i++)
5448 : {
5449 222 : input_chars[i] = utf8_to_unicode(p);
5450 222 : p += pg_utf_mblen(p);
5451 : }
5452 66 : input_chars[i] = (char32_t) '\0';
5453 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5454 :
5455 : /* action */
5456 66 : output_chars = unicode_normalize(form, input_chars);
5457 :
5458 : /* convert back to UTF-8 string */
5459 66 : size = 0;
5460 306 : for (char32_t *wp = output_chars; *wp; wp++)
5461 : {
5462 : unsigned char buf[4];
5463 :
5464 240 : unicode_to_utf8(*wp, buf);
5465 240 : size += pg_utf_mblen(buf);
5466 : }
5467 :
5468 66 : result = palloc(size + VARHDRSZ);
5469 66 : SET_VARSIZE(result, size + VARHDRSZ);
5470 :
5471 66 : p = (unsigned char *) VARDATA_ANY(result);
5472 306 : for (char32_t *wp = output_chars; *wp; wp++)
5473 : {
5474 240 : unicode_to_utf8(*wp, p);
5475 240 : p += pg_utf_mblen(p);
5476 : }
5477 : Assert((char *) p == (char *) result + size + VARHDRSZ);
5478 :
5479 66 : PG_RETURN_TEXT_P(result);
5480 : }
5481 :
5482 : /*
5483 : * Check whether the string is in the specified Unicode normalization form.
5484 : *
5485 : * This is done by converting the string to the specified normal form and then
5486 : * comparing that to the original string. To speed that up, we also apply the
5487 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
5488 : * answer for many strings by just scanning the string once.
5489 : *
5490 : * This function should generally be optimized for the case where the string
5491 : * is in fact normalized. In that case, we'll end up looking at the entire
5492 : * string, so it's probably not worth doing any incremental conversion etc.
5493 : */
5494 : Datum
5495 138 : unicode_is_normalized(PG_FUNCTION_ARGS)
5496 : {
5497 138 : text *input = PG_GETARG_TEXT_PP(0);
5498 138 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5499 : UnicodeNormalizationForm form;
5500 : int size;
5501 : char32_t *input_chars;
5502 : char32_t *output_chars;
5503 : unsigned char *p;
5504 : int i;
5505 : UnicodeNormalizationQC quickcheck;
5506 : int output_size;
5507 : bool result;
5508 :
5509 138 : form = unicode_norm_form_from_string(formstr);
5510 :
5511 : /* convert to char32_t */
5512 132 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5513 132 : input_chars = palloc((size + 1) * sizeof(char32_t));
5514 132 : p = (unsigned char *) VARDATA_ANY(input);
5515 504 : for (i = 0; i < size; i++)
5516 : {
5517 372 : input_chars[i] = utf8_to_unicode(p);
5518 372 : p += pg_utf_mblen(p);
5519 : }
5520 132 : input_chars[i] = (char32_t) '\0';
5521 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5522 :
5523 : /* quick check (see UAX #15) */
5524 132 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
5525 132 : if (quickcheck == UNICODE_NORM_QC_YES)
5526 42 : PG_RETURN_BOOL(true);
5527 90 : else if (quickcheck == UNICODE_NORM_QC_NO)
5528 12 : PG_RETURN_BOOL(false);
5529 :
5530 : /* normalize and compare with original */
5531 78 : output_chars = unicode_normalize(form, input_chars);
5532 :
5533 78 : output_size = 0;
5534 324 : for (char32_t *wp = output_chars; *wp; wp++)
5535 246 : output_size++;
5536 :
5537 114 : result = (size == output_size) &&
5538 36 : (memcmp(input_chars, output_chars, size * sizeof(char32_t)) == 0);
5539 :
5540 78 : PG_RETURN_BOOL(result);
5541 : }
5542 :
5543 : /*
5544 : * Check if first n chars are hexadecimal digits
5545 : */
5546 : static bool
5547 156 : isxdigits_n(const char *instr, size_t n)
5548 : {
5549 660 : for (size_t i = 0; i < n; i++)
5550 570 : if (!isxdigit((unsigned char) instr[i]))
5551 66 : return false;
5552 :
5553 90 : return true;
5554 : }
5555 :
5556 : static unsigned int
5557 504 : hexval(unsigned char c)
5558 : {
5559 504 : if (c >= '0' && c <= '9')
5560 384 : return c - '0';
5561 120 : if (c >= 'a' && c <= 'f')
5562 60 : return c - 'a' + 0xA;
5563 60 : if (c >= 'A' && c <= 'F')
5564 60 : return c - 'A' + 0xA;
5565 0 : elog(ERROR, "invalid hexadecimal digit");
5566 : return 0; /* not reached */
5567 : }
5568 :
5569 : /*
5570 : * Translate string with hexadecimal digits to number
5571 : */
5572 : static unsigned int
5573 90 : hexval_n(const char *instr, size_t n)
5574 : {
5575 90 : unsigned int result = 0;
5576 :
5577 594 : for (size_t i = 0; i < n; i++)
5578 504 : result += hexval(instr[i]) << (4 * (n - i - 1));
5579 :
5580 90 : return result;
5581 : }
5582 :
5583 : /*
5584 : * Replaces Unicode escape sequences by Unicode characters
5585 : */
5586 : Datum
5587 66 : unistr(PG_FUNCTION_ARGS)
5588 : {
5589 66 : text *input_text = PG_GETARG_TEXT_PP(0);
5590 : char *instr;
5591 : int len;
5592 : StringInfoData str;
5593 : text *result;
5594 66 : char16_t pair_first = 0;
5595 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
5596 :
5597 66 : instr = VARDATA_ANY(input_text);
5598 66 : len = VARSIZE_ANY_EXHDR(input_text);
5599 :
5600 66 : initStringInfo(&str);
5601 :
5602 510 : while (len > 0)
5603 : {
5604 486 : if (instr[0] == '\\')
5605 : {
5606 102 : if (len >= 2 &&
5607 102 : instr[1] == '\\')
5608 : {
5609 6 : if (pair_first)
5610 0 : goto invalid_pair;
5611 6 : appendStringInfoChar(&str, '\\');
5612 6 : instr += 2;
5613 6 : len -= 2;
5614 : }
5615 96 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
5616 66 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
5617 30 : {
5618 : char32_t unicode;
5619 42 : int offset = instr[1] == 'u' ? 2 : 1;
5620 :
5621 42 : unicode = hexval_n(instr + offset, 4);
5622 :
5623 42 : if (!is_valid_unicode_codepoint(unicode))
5624 0 : ereport(ERROR,
5625 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5626 : errmsg("invalid Unicode code point: %04X", unicode));
5627 :
5628 42 : if (pair_first)
5629 : {
5630 12 : if (is_utf16_surrogate_second(unicode))
5631 : {
5632 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5633 0 : pair_first = 0;
5634 : }
5635 : else
5636 12 : goto invalid_pair;
5637 : }
5638 30 : else if (is_utf16_surrogate_second(unicode))
5639 0 : goto invalid_pair;
5640 :
5641 30 : if (is_utf16_surrogate_first(unicode))
5642 18 : pair_first = unicode;
5643 : else
5644 : {
5645 12 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5646 12 : appendStringInfoString(&str, cbuf);
5647 : }
5648 :
5649 30 : instr += 4 + offset;
5650 30 : len -= 4 + offset;
5651 : }
5652 54 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
5653 12 : {
5654 : char32_t unicode;
5655 :
5656 24 : unicode = hexval_n(instr + 2, 6);
5657 :
5658 24 : if (!is_valid_unicode_codepoint(unicode))
5659 6 : ereport(ERROR,
5660 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5661 : errmsg("invalid Unicode code point: %04X", unicode));
5662 :
5663 18 : if (pair_first)
5664 : {
5665 6 : if (is_utf16_surrogate_second(unicode))
5666 : {
5667 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5668 0 : pair_first = 0;
5669 : }
5670 : else
5671 6 : goto invalid_pair;
5672 : }
5673 12 : else if (is_utf16_surrogate_second(unicode))
5674 0 : goto invalid_pair;
5675 :
5676 12 : if (is_utf16_surrogate_first(unicode))
5677 6 : pair_first = unicode;
5678 : else
5679 : {
5680 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5681 6 : appendStringInfoString(&str, cbuf);
5682 : }
5683 :
5684 12 : instr += 8;
5685 12 : len -= 8;
5686 : }
5687 30 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
5688 12 : {
5689 : char32_t unicode;
5690 :
5691 24 : unicode = hexval_n(instr + 2, 8);
5692 :
5693 24 : if (!is_valid_unicode_codepoint(unicode))
5694 6 : ereport(ERROR,
5695 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5696 : errmsg("invalid Unicode code point: %04X", unicode));
5697 :
5698 18 : if (pair_first)
5699 : {
5700 6 : if (is_utf16_surrogate_second(unicode))
5701 : {
5702 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5703 0 : pair_first = 0;
5704 : }
5705 : else
5706 6 : goto invalid_pair;
5707 : }
5708 12 : else if (is_utf16_surrogate_second(unicode))
5709 0 : goto invalid_pair;
5710 :
5711 12 : if (is_utf16_surrogate_first(unicode))
5712 6 : pair_first = unicode;
5713 : else
5714 : {
5715 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5716 6 : appendStringInfoString(&str, cbuf);
5717 : }
5718 :
5719 12 : instr += 10;
5720 12 : len -= 10;
5721 : }
5722 : else
5723 6 : ereport(ERROR,
5724 : (errcode(ERRCODE_SYNTAX_ERROR),
5725 : errmsg("invalid Unicode escape"),
5726 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
5727 : }
5728 : else
5729 : {
5730 384 : if (pair_first)
5731 0 : goto invalid_pair;
5732 :
5733 384 : appendStringInfoChar(&str, *instr++);
5734 384 : len--;
5735 : }
5736 : }
5737 :
5738 : /* unfinished surrogate pair? */
5739 24 : if (pair_first)
5740 6 : goto invalid_pair;
5741 :
5742 18 : result = cstring_to_text_with_len(str.data, str.len);
5743 18 : pfree(str.data);
5744 :
5745 18 : PG_RETURN_TEXT_P(result);
5746 :
5747 30 : invalid_pair:
5748 30 : ereport(ERROR,
5749 : (errcode(ERRCODE_SYNTAX_ERROR),
5750 : errmsg("invalid Unicode surrogate pair")));
5751 : PG_RETURN_NULL(); /* keep compiler quiet */
5752 : }
|
