Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * varlena.c
4 : * Functions for the variable-length built-in types.
5 : *
6 : * Portions Copyright (c) 1996-2025, 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/bytea/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 25685316 : cstring_to_text(const char *s)
182 : {
183 25685316 : 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 28492326 : cstring_to_text_with_len(const char *s, int len)
194 : {
195 28492326 : text *result = (text *) palloc(len + VARHDRSZ);
196 :
197 28492326 : SET_VARSIZE(result, len + VARHDRSZ);
198 28492326 : memcpy(VARDATA(result), s, len);
199 :
200 28492326 : 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 18414216 : text_to_cstring(const text *t)
215 : {
216 : /* must cast away the const, unfortunately */
217 18414216 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
218 18414216 : int len = VARSIZE_ANY_EXHDR(tunpacked);
219 : char *result;
220 :
221 18414216 : result = (char *) palloc(len + 1);
222 18414216 : memcpy(result, VARDATA_ANY(tunpacked), len);
223 18414216 : result[len] = '\0';
224 :
225 18414216 : if (tunpacked != t)
226 45988 : pfree(tunpacked);
227 :
228 18414216 : 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
245 994 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
246 : {
247 : /* must cast away the const, unfortunately */
248 994 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
249 994 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
250 :
251 994 : if (dst_len > 0)
252 : {
253 994 : dst_len--;
254 994 : if (dst_len >= src_len)
255 994 : dst_len = src_len;
256 : else /* ensure truncation is encoding-safe */
257 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
258 994 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
259 994 : dst[dst_len] = '\0';
260 : }
261 :
262 994 : if (srcunpacked != src)
263 0 : pfree(srcunpacked);
264 994 : }
265 :
266 :
267 : /*****************************************************************************
268 : * USER I/O ROUTINES *
269 : *****************************************************************************/
270 :
271 : /*
272 : * textin - converts cstring to internal representation
273 : */
274 : Datum
275 22398462 : textin(PG_FUNCTION_ARGS)
276 : {
277 22398462 : char *inputText = PG_GETARG_CSTRING(0);
278 :
279 22398462 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
280 : }
281 :
282 : /*
283 : * textout - converts internal representation to cstring
284 : */
285 : Datum
286 9439744 : textout(PG_FUNCTION_ARGS)
287 : {
288 9439744 : Datum txt = PG_GETARG_DATUM(0);
289 :
290 9439744 : 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 4948 : textsend(PG_FUNCTION_ARGS)
316 : {
317 4948 : text *t = PG_GETARG_TEXT_PP(0);
318 : StringInfoData buf;
319 :
320 4948 : pq_begintypsend(&buf);
321 4948 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
322 4948 : 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
342 952 : unknownout(PG_FUNCTION_ARGS)
343 : {
344 : /* representation is same as cstring */
345 952 : char *str = PG_GETARG_CSTRING(0);
346 :
347 952 : 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 430800 : textlen(PG_FUNCTION_ARGS)
390 : {
391 430800 : Datum str = PG_GETARG_DATUM(0);
392 :
393 : /* try to avoid decompressing argument */
394 430800 : 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 430812 : text_length(Datum str)
408 : {
409 : /* fastpath when max encoding length is one */
410 430812 : if (pg_database_encoding_max_length() == 1)
411 20 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
412 : else
413 : {
414 430792 : text *t = DatumGetTextPP(str);
415 :
416 430792 : PG_RETURN_INT32(pg_mbstrlen_with_len(VARDATA_ANY(t),
417 : VARSIZE_ANY_EXHDR(t)));
418 : }
419 : }
420 :
421 : /*
422 : * textoctetlen -
423 : * returns the physical length of a text*
424 : * (which is less than the VARSIZE of the text*)
425 : */
426 : Datum
427 70 : textoctetlen(PG_FUNCTION_ARGS)
428 : {
429 70 : Datum str = PG_GETARG_DATUM(0);
430 :
431 : /* We need not detoast the input at all */
432 70 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
433 : }
434 :
435 : /*
436 : * textcat -
437 : * takes two text* and returns a text* that is the concatenation of
438 : * the two.
439 : *
440 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
441 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
442 : * Allocate space for output in all cases.
443 : * XXX - thomas 1997-07-10
444 : */
445 : Datum
446 1953394 : textcat(PG_FUNCTION_ARGS)
447 : {
448 1953394 : text *t1 = PG_GETARG_TEXT_PP(0);
449 1953394 : text *t2 = PG_GETARG_TEXT_PP(1);
450 :
451 1953394 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
452 : }
453 :
454 : /*
455 : * text_catenate
456 : * Guts of textcat(), broken out so it can be used by other functions
457 : *
458 : * Arguments can be in short-header form, but not compressed or out-of-line
459 : */
460 : static text *
461 1953474 : text_catenate(text *t1, text *t2)
462 : {
463 : text *result;
464 : int len1,
465 : len2,
466 : len;
467 : char *ptr;
468 :
469 1953474 : len1 = VARSIZE_ANY_EXHDR(t1);
470 1953474 : len2 = VARSIZE_ANY_EXHDR(t2);
471 :
472 : /* paranoia ... probably should throw error instead? */
473 1953474 : if (len1 < 0)
474 0 : len1 = 0;
475 1953474 : if (len2 < 0)
476 0 : len2 = 0;
477 :
478 1953474 : len = len1 + len2 + VARHDRSZ;
479 1953474 : result = (text *) palloc(len);
480 :
481 : /* Set size of result string... */
482 1953474 : SET_VARSIZE(result, len);
483 :
484 : /* Fill data field of result string... */
485 1953474 : ptr = VARDATA(result);
486 1953474 : if (len1 > 0)
487 1952650 : memcpy(ptr, VARDATA_ANY(t1), len1);
488 1953474 : if (len2 > 0)
489 1953264 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
490 :
491 1953474 : return result;
492 : }
493 :
494 : /*
495 : * charlen_to_bytelen()
496 : * Compute the number of bytes occupied by n characters starting at *p
497 : *
498 : * It is caller's responsibility that there actually are n characters;
499 : * the string need not be null-terminated.
500 : */
501 : static int
502 17126 : charlen_to_bytelen(const char *p, int n)
503 : {
504 17126 : if (pg_database_encoding_max_length() == 1)
505 : {
506 : /* Optimization for single-byte encodings */
507 180 : return n;
508 : }
509 : else
510 : {
511 : const char *s;
512 :
513 6050256 : for (s = p; n > 0; n--)
514 6033310 : s += pg_mblen(s);
515 :
516 16946 : return s - p;
517 : }
518 : }
519 :
520 : /*
521 : * text_substr()
522 : * Return a substring starting at the specified position.
523 : * - thomas 1997-12-31
524 : *
525 : * Input:
526 : * - string
527 : * - starting position (is one-based)
528 : * - string length
529 : *
530 : * If the starting position is zero or less, then return from the start of the string
531 : * adjusting the length to be consistent with the "negative start" per SQL.
532 : * If the length is less than zero, return the remaining string.
533 : *
534 : * Added multibyte support.
535 : * - Tatsuo Ishii 1998-4-21
536 : * Changed behavior if starting position is less than one to conform to SQL behavior.
537 : * Formerly returned the entire string; now returns a portion.
538 : * - Thomas Lockhart 1998-12-10
539 : * Now uses faster TOAST-slicing interface
540 : * - John Gray 2002-02-22
541 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
542 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
543 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
544 : * S > LC and < LC + 4 sometimes garbage characters are returned.
545 : * - Joe Conway 2002-08-10
546 : */
547 : Datum
548 658136 : text_substr(PG_FUNCTION_ARGS)
549 : {
550 658136 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
551 : PG_GETARG_INT32(1),
552 : PG_GETARG_INT32(2),
553 : false));
554 : }
555 :
556 : /*
557 : * text_substr_no_len -
558 : * Wrapper to avoid opr_sanity failure due to
559 : * one function accepting a different number of args.
560 : */
561 : Datum
562 36 : text_substr_no_len(PG_FUNCTION_ARGS)
563 : {
564 36 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
565 : PG_GETARG_INT32(1),
566 : -1, true));
567 : }
568 :
569 : /*
570 : * text_substring -
571 : * Does the real work for text_substr() and text_substr_no_len()
572 : *
573 : * This is broken out so it can be called directly by other string processing
574 : * functions. Note that the argument is passed as a Datum, to indicate that
575 : * it may still be in compressed/toasted form. We can avoid detoasting all
576 : * of it in some cases.
577 : *
578 : * The result is always a freshly palloc'd datum.
579 : */
580 : static text *
581 698284 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
582 : {
583 698284 : int32 eml = pg_database_encoding_max_length();
584 698284 : int32 S = start; /* start position */
585 : int32 S1; /* adjusted start position */
586 : int32 L1; /* adjusted substring length */
587 : int32 E; /* end position */
588 :
589 : /*
590 : * SQL99 says S can be zero or negative (which we don't document), but we
591 : * still must fetch from the start of the string.
592 : * https://www.postgresql.org/message-id/170905442373.643.11536838320909376197%40wrigleys.postgresql.org
593 : */
594 698284 : S1 = Max(S, 1);
595 :
596 : /* life is easy if the encoding max length is 1 */
597 698284 : if (eml == 1)
598 : {
599 22 : if (length_not_specified) /* special case - get length to end of
600 : * string */
601 0 : L1 = -1;
602 22 : else if (length < 0)
603 : {
604 : /* SQL99 says to throw an error for E < S, i.e., negative length */
605 0 : ereport(ERROR,
606 : (errcode(ERRCODE_SUBSTRING_ERROR),
607 : errmsg("negative substring length not allowed")));
608 : L1 = -1; /* silence stupider compilers */
609 : }
610 22 : else if (pg_add_s32_overflow(S, length, &E))
611 : {
612 : /*
613 : * L could be large enough for S + L to overflow, in which case
614 : * the substring must run to end of string.
615 : */
616 0 : L1 = -1;
617 : }
618 : else
619 : {
620 : /*
621 : * A zero or negative value for the end position can happen if the
622 : * start was negative or one. SQL99 says to return a zero-length
623 : * string.
624 : */
625 22 : if (E < 1)
626 0 : return cstring_to_text("");
627 :
628 22 : L1 = E - S1;
629 : }
630 :
631 : /*
632 : * If the start position is past the end of the string, SQL99 says to
633 : * return a zero-length string -- DatumGetTextPSlice() will do that
634 : * for us. We need only convert S1 to zero-based starting position.
635 : */
636 22 : return DatumGetTextPSlice(str, S1 - 1, L1);
637 : }
638 698262 : else if (eml > 1)
639 : {
640 : /*
641 : * When encoding max length is > 1, we can't get LC without
642 : * detoasting, so we'll grab a conservatively large slice now and go
643 : * back later to do the right thing
644 : */
645 : int32 slice_start;
646 : int32 slice_size;
647 : int32 slice_strlen;
648 : text *slice;
649 : int32 E1;
650 : int32 i;
651 : char *p;
652 : char *s;
653 : text *ret;
654 :
655 : /*
656 : * We need to start at position zero because there is no way to know
657 : * in advance which byte offset corresponds to the supplied start
658 : * position.
659 : */
660 698262 : slice_start = 0;
661 :
662 698262 : if (length_not_specified) /* special case - get length to end of
663 : * string */
664 76 : slice_size = L1 = -1;
665 698186 : else if (length < 0)
666 : {
667 : /* SQL99 says to throw an error for E < S, i.e., negative length */
668 12 : ereport(ERROR,
669 : (errcode(ERRCODE_SUBSTRING_ERROR),
670 : errmsg("negative substring length not allowed")));
671 : slice_size = L1 = -1; /* silence stupider compilers */
672 : }
673 698174 : else if (pg_add_s32_overflow(S, length, &E))
674 : {
675 : /*
676 : * L could be large enough for S + L to overflow, in which case
677 : * the substring must run to end of string.
678 : */
679 6 : slice_size = L1 = -1;
680 : }
681 : else
682 : {
683 : /*
684 : * A zero or negative value for the end position can happen if the
685 : * start was negative or one. SQL99 says to return a zero-length
686 : * string.
687 : */
688 698168 : if (E < 1)
689 0 : return cstring_to_text("");
690 :
691 : /*
692 : * if E is past the end of the string, the tuple toaster will
693 : * truncate the length for us
694 : */
695 698168 : L1 = E - S1;
696 :
697 : /*
698 : * Total slice size in bytes can't be any longer than the start
699 : * position plus substring length times the encoding max length.
700 : * If that overflows, we can just use -1.
701 : */
702 698168 : if (pg_mul_s32_overflow(E, eml, &slice_size))
703 6 : slice_size = -1;
704 : }
705 :
706 : /*
707 : * If we're working with an untoasted source, no need to do an extra
708 : * copying step.
709 : */
710 698250 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
711 698196 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
712 324 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
713 : else
714 697926 : slice = (text *) DatumGetPointer(str);
715 :
716 : /* see if we got back an empty string */
717 698250 : if (VARSIZE_ANY_EXHDR(slice) == 0)
718 : {
719 0 : if (slice != (text *) DatumGetPointer(str))
720 0 : pfree(slice);
721 0 : return cstring_to_text("");
722 : }
723 :
724 : /* Now we can get the actual length of the slice in MB characters */
725 698250 : slice_strlen = pg_mbstrlen_with_len(VARDATA_ANY(slice),
726 698250 : VARSIZE_ANY_EXHDR(slice));
727 :
728 : /*
729 : * Check that the start position wasn't > slice_strlen. If so, SQL99
730 : * says to return a zero-length string.
731 : */
732 698250 : if (S1 > slice_strlen)
733 : {
734 22 : if (slice != (text *) DatumGetPointer(str))
735 0 : pfree(slice);
736 22 : return cstring_to_text("");
737 : }
738 :
739 : /*
740 : * Adjust L1 and E1 now that we know the slice string length. Again
741 : * remember that S1 is one based, and slice_start is zero based.
742 : */
743 698228 : if (L1 > -1)
744 698168 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
745 : else
746 60 : E1 = slice_start + 1 + slice_strlen;
747 :
748 : /*
749 : * Find the start position in the slice; remember S1 is not zero based
750 : */
751 698228 : p = VARDATA_ANY(slice);
752 6709182 : for (i = 0; i < S1 - 1; i++)
753 6010954 : p += pg_mblen(p);
754 :
755 : /* hang onto a pointer to our start position */
756 698228 : s = p;
757 :
758 : /*
759 : * Count the actual bytes used by the substring of the requested
760 : * length.
761 : */
762 9937028 : for (i = S1; i < E1; i++)
763 9238800 : p += pg_mblen(p);
764 :
765 698228 : ret = (text *) palloc(VARHDRSZ + (p - s));
766 698228 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
767 698228 : memcpy(VARDATA(ret), s, (p - s));
768 :
769 698228 : if (slice != (text *) DatumGetPointer(str))
770 324 : pfree(slice);
771 :
772 698228 : return ret;
773 : }
774 : else
775 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
776 :
777 : /* not reached: suppress compiler warning */
778 : return NULL;
779 : }
780 :
781 : /*
782 : * textoverlay
783 : * Replace specified substring of first string with second
784 : *
785 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
786 : * This code is a direct implementation of what the standard says.
787 : */
788 : Datum
789 28 : textoverlay(PG_FUNCTION_ARGS)
790 : {
791 28 : text *t1 = PG_GETARG_TEXT_PP(0);
792 28 : text *t2 = PG_GETARG_TEXT_PP(1);
793 28 : int sp = PG_GETARG_INT32(2); /* substring start position */
794 28 : int sl = PG_GETARG_INT32(3); /* substring length */
795 :
796 28 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
797 : }
798 :
799 : Datum
800 12 : textoverlay_no_len(PG_FUNCTION_ARGS)
801 : {
802 12 : text *t1 = PG_GETARG_TEXT_PP(0);
803 12 : text *t2 = PG_GETARG_TEXT_PP(1);
804 12 : int sp = PG_GETARG_INT32(2); /* substring start position */
805 : int sl;
806 :
807 12 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
808 12 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
809 : }
810 :
811 : static text *
812 40 : text_overlay(text *t1, text *t2, int sp, int sl)
813 : {
814 : text *result;
815 : text *s1;
816 : text *s2;
817 : int sp_pl_sl;
818 :
819 : /*
820 : * Check for possible integer-overflow cases. For negative sp, throw a
821 : * "substring length" error because that's what should be expected
822 : * according to the spec's definition of OVERLAY().
823 : */
824 40 : if (sp <= 0)
825 0 : ereport(ERROR,
826 : (errcode(ERRCODE_SUBSTRING_ERROR),
827 : errmsg("negative substring length not allowed")));
828 40 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
829 0 : ereport(ERROR,
830 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
831 : errmsg("integer out of range")));
832 :
833 40 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
834 40 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
835 40 : result = text_catenate(s1, t2);
836 40 : result = text_catenate(result, s2);
837 :
838 40 : return result;
839 : }
840 :
841 : /*
842 : * textpos -
843 : * Return the position of the specified substring.
844 : * Implements the SQL POSITION() function.
845 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
846 : * - thomas 1997-07-27
847 : */
848 : Datum
849 130 : textpos(PG_FUNCTION_ARGS)
850 : {
851 130 : text *str = PG_GETARG_TEXT_PP(0);
852 130 : text *search_str = PG_GETARG_TEXT_PP(1);
853 :
854 130 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
855 : }
856 :
857 : /*
858 : * text_position -
859 : * Does the real work for textpos()
860 : *
861 : * Inputs:
862 : * t1 - string to be searched
863 : * t2 - pattern to match within t1
864 : * Result:
865 : * Character index of the first matched char, starting from 1,
866 : * or 0 if no match.
867 : *
868 : * This is broken out so it can be called directly by other string processing
869 : * functions.
870 : */
871 : static int
872 130 : text_position(text *t1, text *t2, Oid collid)
873 : {
874 : TextPositionState state;
875 : int result;
876 :
877 130 : check_collation_set(collid);
878 :
879 : /* Empty needle always matches at position 1 */
880 130 : if (VARSIZE_ANY_EXHDR(t2) < 1)
881 12 : return 1;
882 :
883 : /* Otherwise, can't match if haystack is shorter than needle */
884 118 : if (VARSIZE_ANY_EXHDR(t1) < VARSIZE_ANY_EXHDR(t2) &&
885 22 : pg_newlocale_from_collation(collid)->deterministic)
886 22 : return 0;
887 :
888 96 : text_position_setup(t1, t2, collid, &state);
889 : /* don't need greedy mode here */
890 96 : state.greedy = false;
891 :
892 96 : if (!text_position_next(&state))
893 24 : result = 0;
894 : else
895 72 : result = text_position_get_match_pos(&state);
896 96 : text_position_cleanup(&state);
897 96 : return result;
898 : }
899 :
900 :
901 : /*
902 : * text_position_setup, text_position_next, text_position_cleanup -
903 : * Component steps of text_position()
904 : *
905 : * These are broken out so that a string can be efficiently searched for
906 : * multiple occurrences of the same pattern. text_position_next may be
907 : * called multiple times, and it advances to the next match on each call.
908 : * text_position_get_match_ptr() and text_position_get_match_pos() return
909 : * a pointer or 1-based character position of the last match, respectively.
910 : *
911 : * The "state" variable is normally just a local variable in the caller.
912 : *
913 : * NOTE: text_position_next skips over the matched portion. For example,
914 : * searching for "xx" in "xxx" returns only one match, not two.
915 : */
916 :
917 : static void
918 1774 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
919 : {
920 1774 : int len1 = VARSIZE_ANY_EXHDR(t1);
921 1774 : int len2 = VARSIZE_ANY_EXHDR(t2);
922 :
923 1774 : check_collation_set(collid);
924 :
925 1774 : state->locale = pg_newlocale_from_collation(collid);
926 :
927 : /*
928 : * Most callers need greedy mode, but some might want to unset this to
929 : * optimize.
930 : */
931 1774 : state->greedy = true;
932 :
933 : Assert(len2 > 0);
934 :
935 : /*
936 : * Even with a multi-byte encoding, we perform the search using the raw
937 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
938 : * because in UTF-8 the byte sequence of one character cannot contain
939 : * another character. For other multi-byte encodings, we do the search
940 : * initially as a simple byte search, ignoring multibyte issues, but
941 : * verify afterwards that the match we found is at a character boundary,
942 : * and continue the search if it was a false match.
943 : */
944 1774 : if (pg_database_encoding_max_length() == 1)
945 108 : state->is_multibyte_char_in_char = false;
946 1666 : else if (GetDatabaseEncoding() == PG_UTF8)
947 1666 : state->is_multibyte_char_in_char = false;
948 : else
949 0 : state->is_multibyte_char_in_char = true;
950 :
951 1774 : state->str1 = VARDATA_ANY(t1);
952 1774 : state->str2 = VARDATA_ANY(t2);
953 1774 : state->len1 = len1;
954 1774 : state->len2 = len2;
955 1774 : state->last_match = NULL;
956 1774 : state->refpoint = state->str1;
957 1774 : state->refpos = 0;
958 :
959 : /*
960 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
961 : * notes we use the terminology that the "haystack" is the string to be
962 : * searched (t1) and the "needle" is the pattern being sought (t2).
963 : *
964 : * If the needle is empty or bigger than the haystack then there is no
965 : * point in wasting cycles initializing the table. We also choose not to
966 : * use B-M-H for needles of length 1, since the skip table can't possibly
967 : * save anything in that case.
968 : *
969 : * (With nondeterministic collations, the search is already
970 : * multibyte-aware, so we don't need this.)
971 : */
972 1774 : if (len1 >= len2 && len2 > 1 && state->locale->deterministic)
973 : {
974 1446 : int searchlength = len1 - len2;
975 : int skiptablemask;
976 : int last;
977 : int i;
978 1446 : const char *str2 = state->str2;
979 :
980 : /*
981 : * First we must determine how much of the skip table to use. The
982 : * declaration of TextPositionState allows up to 256 elements, but for
983 : * short search problems we don't really want to have to initialize so
984 : * many elements --- it would take too long in comparison to the
985 : * actual search time. So we choose a useful skip table size based on
986 : * the haystack length minus the needle length. The closer the needle
987 : * length is to the haystack length the less useful skipping becomes.
988 : *
989 : * Note: since we use bit-masking to select table elements, the skip
990 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
991 : */
992 1446 : if (searchlength < 16)
993 114 : skiptablemask = 3;
994 1332 : else if (searchlength < 64)
995 16 : skiptablemask = 7;
996 1316 : else if (searchlength < 128)
997 14 : skiptablemask = 15;
998 1302 : else if (searchlength < 512)
999 256 : skiptablemask = 31;
1000 1046 : else if (searchlength < 2048)
1001 794 : skiptablemask = 63;
1002 252 : else if (searchlength < 4096)
1003 176 : skiptablemask = 127;
1004 : else
1005 76 : skiptablemask = 255;
1006 1446 : state->skiptablemask = skiptablemask;
1007 :
1008 : /*
1009 : * Initialize the skip table. We set all elements to the needle
1010 : * length, since this is the correct skip distance for any character
1011 : * not found in the needle.
1012 : */
1013 103246 : for (i = 0; i <= skiptablemask; i++)
1014 101800 : state->skiptable[i] = len2;
1015 :
1016 : /*
1017 : * Now examine the needle. For each character except the last one,
1018 : * set the corresponding table element to the appropriate skip
1019 : * distance. Note that when two characters share the same skip table
1020 : * entry, the one later in the needle must determine the skip
1021 : * distance.
1022 : */
1023 1446 : last = len2 - 1;
1024 :
1025 19194 : for (i = 0; i < last; i++)
1026 17748 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1027 : }
1028 1774 : }
1029 :
1030 : /*
1031 : * Advance to the next match, starting from the end of the previous match
1032 : * (or the beginning of the string, on first call). Returns true if a match
1033 : * is found.
1034 : *
1035 : * Note that this refuses to match an empty-string needle. Most callers
1036 : * will have handled that case specially and we'll never see it here.
1037 : */
1038 : static bool
1039 8154 : text_position_next(TextPositionState *state)
1040 : {
1041 8154 : int needle_len = state->len2;
1042 : char *start_ptr;
1043 : char *matchptr;
1044 :
1045 8154 : if (needle_len <= 0)
1046 0 : return false; /* result for empty pattern */
1047 :
1048 : /* Start from the point right after the previous match. */
1049 8154 : if (state->last_match)
1050 6368 : start_ptr = state->last_match + state->last_match_len;
1051 : else
1052 1786 : start_ptr = state->str1;
1053 :
1054 8154 : retry:
1055 8154 : matchptr = text_position_next_internal(start_ptr, state);
1056 :
1057 8154 : if (!matchptr)
1058 1690 : return false;
1059 :
1060 : /*
1061 : * Found a match for the byte sequence. If this is a multibyte encoding,
1062 : * where one character's byte sequence can appear inside a longer
1063 : * multi-byte character, we need to verify that the match was at a
1064 : * character boundary, not in the middle of a multi-byte character.
1065 : */
1066 6464 : if (state->is_multibyte_char_in_char && state->locale->deterministic)
1067 : {
1068 : /* Walk one character at a time, until we reach the match. */
1069 :
1070 : /* the search should never move backwards. */
1071 : Assert(state->refpoint <= matchptr);
1072 :
1073 0 : while (state->refpoint < matchptr)
1074 : {
1075 : /* step to next character. */
1076 0 : state->refpoint += pg_mblen(state->refpoint);
1077 0 : state->refpos++;
1078 :
1079 : /*
1080 : * If we stepped over the match's start position, then it was a
1081 : * false positive, where the byte sequence appeared in the middle
1082 : * of a multi-byte character. Skip it, and continue the search at
1083 : * the next character boundary.
1084 : */
1085 0 : if (state->refpoint > matchptr)
1086 : {
1087 0 : start_ptr = state->refpoint;
1088 0 : goto retry;
1089 : }
1090 : }
1091 : }
1092 :
1093 6464 : state->last_match = matchptr;
1094 6464 : state->last_match_len = state->last_match_len_tmp;
1095 6464 : return true;
1096 : }
1097 :
1098 : /*
1099 : * Subroutine of text_position_next(). This searches for the raw byte
1100 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1101 : * match starting at 'start_ptr', or NULL if no match is found.
1102 : */
1103 : static char *
1104 8154 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1105 : {
1106 8154 : int haystack_len = state->len1;
1107 8154 : int needle_len = state->len2;
1108 8154 : int skiptablemask = state->skiptablemask;
1109 8154 : const char *haystack = state->str1;
1110 8154 : const char *needle = state->str2;
1111 8154 : const char *haystack_end = &haystack[haystack_len];
1112 : const char *hptr;
1113 :
1114 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1115 :
1116 8154 : state->last_match_len_tmp = needle_len;
1117 :
1118 8154 : 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, including being empty. Callers that want to skip over the
1128 : * found string need to read the length of the found substring from
1129 : * last_match_len rather 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 240 : const char *result_hptr = NULL;
1136 :
1137 240 : hptr = start_ptr;
1138 642 : while (hptr < haystack_end)
1139 : {
1140 : /*
1141 : * First check the common case that there is a match in the
1142 : * haystack of exactly the length of the needle.
1143 : */
1144 534 : if (!state->greedy &&
1145 108 : haystack_end - hptr >= needle_len &&
1146 54 : pg_strncoll(hptr, needle_len, needle, needle_len, state->locale) == 0)
1147 12 : return (char *) hptr;
1148 :
1149 : /*
1150 : * Else check if any of the possible substrings starting at hptr
1151 : * are equal to the needle.
1152 : */
1153 2586 : for (const char *test_end = hptr; test_end < haystack_end; test_end += pg_mblen(test_end))
1154 : {
1155 2064 : if (pg_strncoll(hptr, (test_end - hptr), needle, needle_len, state->locale) == 0)
1156 : {
1157 132 : state->last_match_len_tmp = (test_end - hptr);
1158 132 : result_hptr = hptr;
1159 132 : if (!state->greedy)
1160 0 : break;
1161 : }
1162 : }
1163 522 : if (result_hptr)
1164 120 : break;
1165 :
1166 402 : hptr += pg_mblen(hptr);
1167 : }
1168 :
1169 228 : return (char *) result_hptr;
1170 : }
1171 7914 : else if (needle_len == 1)
1172 : {
1173 : /* No point in using B-M-H for a one-character needle */
1174 760 : char nchar = *needle;
1175 :
1176 760 : hptr = start_ptr;
1177 5878 : while (hptr < haystack_end)
1178 : {
1179 5712 : if (*hptr == nchar)
1180 594 : return (char *) hptr;
1181 5118 : hptr++;
1182 : }
1183 : }
1184 : else
1185 : {
1186 7154 : const char *needle_last = &needle[needle_len - 1];
1187 :
1188 : /* Start at startpos plus the length of the needle */
1189 7154 : hptr = start_ptr + needle_len - 1;
1190 182728 : while (hptr < haystack_end)
1191 : {
1192 : /* Match the needle scanning *backward* */
1193 : const char *nptr;
1194 : const char *p;
1195 :
1196 181312 : nptr = needle_last;
1197 181312 : p = hptr;
1198 266908 : while (*nptr == *p)
1199 : {
1200 : /* Matched it all? If so, return 1-based position */
1201 91334 : if (nptr == needle)
1202 5738 : return (char *) p;
1203 85596 : nptr--, p--;
1204 : }
1205 :
1206 : /*
1207 : * No match, so use the haystack char at hptr to decide how far to
1208 : * advance. If the needle had any occurrence of that character
1209 : * (or more precisely, one sharing the same skiptable entry)
1210 : * before its last character, then we advance far enough to align
1211 : * the last such needle character with that haystack position.
1212 : * Otherwise we can advance by the whole needle length.
1213 : */
1214 175574 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1215 : }
1216 : }
1217 :
1218 1582 : return 0; /* not found */
1219 : }
1220 :
1221 : /*
1222 : * Return a pointer to the current match.
1223 : *
1224 : * The returned pointer points into the original haystack string.
1225 : */
1226 : static char *
1227 6362 : text_position_get_match_ptr(TextPositionState *state)
1228 : {
1229 6362 : return state->last_match;
1230 : }
1231 :
1232 : /*
1233 : * Return the offset of the current match.
1234 : *
1235 : * The offset is in characters, 1-based.
1236 : */
1237 : static int
1238 72 : text_position_get_match_pos(TextPositionState *state)
1239 : {
1240 : /* Convert the byte position to char position. */
1241 144 : state->refpos += pg_mbstrlen_with_len(state->refpoint,
1242 72 : state->last_match - state->refpoint);
1243 72 : state->refpoint = state->last_match;
1244 72 : return state->refpos + 1;
1245 : }
1246 :
1247 : /*
1248 : * Reset search state to the initial state installed by text_position_setup.
1249 : *
1250 : * The next call to text_position_next will search from the beginning
1251 : * of the string.
1252 : */
1253 : static void
1254 12 : text_position_reset(TextPositionState *state)
1255 : {
1256 12 : state->last_match = NULL;
1257 12 : state->refpoint = state->str1;
1258 12 : state->refpos = 0;
1259 12 : }
1260 :
1261 : static void
1262 1774 : text_position_cleanup(TextPositionState *state)
1263 : {
1264 : /* no cleanup needed */
1265 1774 : }
1266 :
1267 :
1268 : static void
1269 16819780 : check_collation_set(Oid collid)
1270 : {
1271 16819780 : if (!OidIsValid(collid))
1272 : {
1273 : /*
1274 : * This typically means that the parser could not resolve a conflict
1275 : * of implicit collations, so report it that way.
1276 : */
1277 30 : ereport(ERROR,
1278 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1279 : errmsg("could not determine which collation to use for string comparison"),
1280 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1281 : }
1282 16819750 : }
1283 :
1284 : /*
1285 : * varstr_cmp()
1286 : *
1287 : * Comparison function for text strings with given lengths, using the
1288 : * appropriate locale. Returns an integer less than, equal to, or greater than
1289 : * zero, indicating whether arg1 is less than, equal to, or greater than arg2.
1290 : *
1291 : * Note: many functions that depend on this are marked leakproof; therefore,
1292 : * avoid reporting the actual contents of the input when throwing errors.
1293 : * All errors herein should be things that can't happen except on corrupt
1294 : * data, anyway; otherwise we will have trouble with indexing strings that
1295 : * would cause them.
1296 : */
1297 : int
1298 9624384 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1299 : {
1300 : int result;
1301 : pg_locale_t mylocale;
1302 :
1303 9624384 : check_collation_set(collid);
1304 :
1305 9624366 : mylocale = pg_newlocale_from_collation(collid);
1306 :
1307 9624366 : if (mylocale->collate_is_c)
1308 : {
1309 3584740 : result = memcmp(arg1, arg2, Min(len1, len2));
1310 3584740 : if ((result == 0) && (len1 != len2))
1311 131916 : result = (len1 < len2) ? -1 : 1;
1312 : }
1313 : else
1314 : {
1315 : /*
1316 : * memcmp() can't tell us which of two unequal strings sorts first,
1317 : * but it's a cheap way to tell if they're equal. Testing shows that
1318 : * memcmp() followed by strcoll() is only trivially slower than
1319 : * strcoll() by itself, so we don't lose much if this doesn't work out
1320 : * very often, and if it does - for example, because there are many
1321 : * equal strings in the input - then we win big by avoiding expensive
1322 : * collation-aware comparisons.
1323 : */
1324 6039626 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1325 1567934 : return 0;
1326 :
1327 4471692 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1328 :
1329 : /* Break tie if necessary. */
1330 4471692 : if (result == 0 && mylocale->deterministic)
1331 : {
1332 0 : result = memcmp(arg1, arg2, Min(len1, len2));
1333 0 : if ((result == 0) && (len1 != len2))
1334 0 : result = (len1 < len2) ? -1 : 1;
1335 : }
1336 : }
1337 :
1338 8056432 : return result;
1339 : }
1340 :
1341 : /* text_cmp()
1342 : * Internal comparison function for text strings.
1343 : * Returns -1, 0 or 1
1344 : */
1345 : static int
1346 7789506 : text_cmp(text *arg1, text *arg2, Oid collid)
1347 : {
1348 : char *a1p,
1349 : *a2p;
1350 : int len1,
1351 : len2;
1352 :
1353 7789506 : a1p = VARDATA_ANY(arg1);
1354 7789506 : a2p = VARDATA_ANY(arg2);
1355 :
1356 7789506 : len1 = VARSIZE_ANY_EXHDR(arg1);
1357 7789506 : len2 = VARSIZE_ANY_EXHDR(arg2);
1358 :
1359 7789506 : return varstr_cmp(a1p, len1, a2p, len2, collid);
1360 : }
1361 :
1362 : /*
1363 : * Comparison functions for text strings.
1364 : *
1365 : * Note: btree indexes need these routines not to leak memory; therefore,
1366 : * be careful to free working copies of toasted datums. Most places don't
1367 : * need to be so careful.
1368 : */
1369 :
1370 : Datum
1371 6744936 : texteq(PG_FUNCTION_ARGS)
1372 : {
1373 6744936 : Oid collid = PG_GET_COLLATION();
1374 6744936 : pg_locale_t mylocale = 0;
1375 : bool result;
1376 :
1377 6744936 : check_collation_set(collid);
1378 :
1379 6744936 : mylocale = pg_newlocale_from_collation(collid);
1380 :
1381 6744936 : if (mylocale->deterministic)
1382 : {
1383 6736496 : Datum arg1 = PG_GETARG_DATUM(0);
1384 6736496 : Datum arg2 = PG_GETARG_DATUM(1);
1385 : Size len1,
1386 : len2;
1387 :
1388 : /*
1389 : * Since we only care about equality or not-equality, we can avoid all
1390 : * the expense of strcoll() here, and just do bitwise comparison. In
1391 : * fact, we don't even have to do a bitwise comparison if we can show
1392 : * the lengths of the strings are unequal; which might save us from
1393 : * having to detoast one or both values.
1394 : */
1395 6736496 : len1 = toast_raw_datum_size(arg1);
1396 6736496 : len2 = toast_raw_datum_size(arg2);
1397 6736496 : if (len1 != len2)
1398 3206422 : result = false;
1399 : else
1400 : {
1401 3530074 : text *targ1 = DatumGetTextPP(arg1);
1402 3530074 : text *targ2 = DatumGetTextPP(arg2);
1403 :
1404 3530074 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1405 : len1 - VARHDRSZ) == 0);
1406 :
1407 3530074 : PG_FREE_IF_COPY(targ1, 0);
1408 3530074 : PG_FREE_IF_COPY(targ2, 1);
1409 : }
1410 : }
1411 : else
1412 : {
1413 8440 : text *arg1 = PG_GETARG_TEXT_PP(0);
1414 8440 : text *arg2 = PG_GETARG_TEXT_PP(1);
1415 :
1416 8440 : result = (text_cmp(arg1, arg2, collid) == 0);
1417 :
1418 8440 : PG_FREE_IF_COPY(arg1, 0);
1419 8440 : PG_FREE_IF_COPY(arg2, 1);
1420 : }
1421 :
1422 6744936 : PG_RETURN_BOOL(result);
1423 : }
1424 :
1425 : Datum
1426 25446 : textne(PG_FUNCTION_ARGS)
1427 : {
1428 25446 : Oid collid = PG_GET_COLLATION();
1429 : pg_locale_t mylocale;
1430 : bool result;
1431 :
1432 25446 : check_collation_set(collid);
1433 :
1434 25446 : mylocale = pg_newlocale_from_collation(collid);
1435 :
1436 25446 : if (mylocale->deterministic)
1437 : {
1438 25422 : Datum arg1 = PG_GETARG_DATUM(0);
1439 25422 : Datum arg2 = PG_GETARG_DATUM(1);
1440 : Size len1,
1441 : len2;
1442 :
1443 : /* See comment in texteq() */
1444 25422 : len1 = toast_raw_datum_size(arg1);
1445 25422 : len2 = toast_raw_datum_size(arg2);
1446 25422 : if (len1 != len2)
1447 4458 : result = true;
1448 : else
1449 : {
1450 20964 : text *targ1 = DatumGetTextPP(arg1);
1451 20964 : text *targ2 = DatumGetTextPP(arg2);
1452 :
1453 20964 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1454 : len1 - VARHDRSZ) != 0);
1455 :
1456 20964 : PG_FREE_IF_COPY(targ1, 0);
1457 20964 : PG_FREE_IF_COPY(targ2, 1);
1458 : }
1459 : }
1460 : else
1461 : {
1462 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
1463 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
1464 :
1465 24 : result = (text_cmp(arg1, arg2, collid) != 0);
1466 :
1467 24 : PG_FREE_IF_COPY(arg1, 0);
1468 24 : PG_FREE_IF_COPY(arg2, 1);
1469 : }
1470 :
1471 25446 : PG_RETURN_BOOL(result);
1472 : }
1473 :
1474 : Datum
1475 212408 : text_lt(PG_FUNCTION_ARGS)
1476 : {
1477 212408 : text *arg1 = PG_GETARG_TEXT_PP(0);
1478 212408 : text *arg2 = PG_GETARG_TEXT_PP(1);
1479 : bool result;
1480 :
1481 212408 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1482 :
1483 212390 : PG_FREE_IF_COPY(arg1, 0);
1484 212390 : PG_FREE_IF_COPY(arg2, 1);
1485 :
1486 212390 : PG_RETURN_BOOL(result);
1487 : }
1488 :
1489 : Datum
1490 318600 : text_le(PG_FUNCTION_ARGS)
1491 : {
1492 318600 : text *arg1 = PG_GETARG_TEXT_PP(0);
1493 318600 : text *arg2 = PG_GETARG_TEXT_PP(1);
1494 : bool result;
1495 :
1496 318600 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1497 :
1498 318600 : PG_FREE_IF_COPY(arg1, 0);
1499 318600 : PG_FREE_IF_COPY(arg2, 1);
1500 :
1501 318600 : PG_RETURN_BOOL(result);
1502 : }
1503 :
1504 : Datum
1505 196002 : text_gt(PG_FUNCTION_ARGS)
1506 : {
1507 196002 : text *arg1 = PG_GETARG_TEXT_PP(0);
1508 196002 : text *arg2 = PG_GETARG_TEXT_PP(1);
1509 : bool result;
1510 :
1511 196002 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1512 :
1513 196002 : PG_FREE_IF_COPY(arg1, 0);
1514 196002 : PG_FREE_IF_COPY(arg2, 1);
1515 :
1516 196002 : PG_RETURN_BOOL(result);
1517 : }
1518 :
1519 : Datum
1520 176220 : text_ge(PG_FUNCTION_ARGS)
1521 : {
1522 176220 : text *arg1 = PG_GETARG_TEXT_PP(0);
1523 176220 : text *arg2 = PG_GETARG_TEXT_PP(1);
1524 : bool result;
1525 :
1526 176220 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1527 :
1528 176220 : PG_FREE_IF_COPY(arg1, 0);
1529 176220 : PG_FREE_IF_COPY(arg2, 1);
1530 :
1531 176220 : PG_RETURN_BOOL(result);
1532 : }
1533 :
1534 : Datum
1535 37914 : text_starts_with(PG_FUNCTION_ARGS)
1536 : {
1537 37914 : Datum arg1 = PG_GETARG_DATUM(0);
1538 37914 : Datum arg2 = PG_GETARG_DATUM(1);
1539 37914 : Oid collid = PG_GET_COLLATION();
1540 : pg_locale_t mylocale;
1541 : bool result;
1542 : Size len1,
1543 : len2;
1544 :
1545 37914 : check_collation_set(collid);
1546 :
1547 37914 : mylocale = pg_newlocale_from_collation(collid);
1548 :
1549 37914 : if (!mylocale->deterministic)
1550 0 : ereport(ERROR,
1551 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1552 : errmsg("nondeterministic collations are not supported for substring searches")));
1553 :
1554 37914 : len1 = toast_raw_datum_size(arg1);
1555 37914 : len2 = toast_raw_datum_size(arg2);
1556 37914 : if (len2 > len1)
1557 0 : result = false;
1558 : else
1559 : {
1560 37914 : text *targ1 = text_substring(arg1, 1, len2, false);
1561 37914 : text *targ2 = DatumGetTextPP(arg2);
1562 :
1563 37914 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1564 37914 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1565 :
1566 37914 : PG_FREE_IF_COPY(targ1, 0);
1567 37914 : PG_FREE_IF_COPY(targ2, 1);
1568 : }
1569 :
1570 37914 : PG_RETURN_BOOL(result);
1571 : }
1572 :
1573 : Datum
1574 6562176 : bttextcmp(PG_FUNCTION_ARGS)
1575 : {
1576 6562176 : text *arg1 = PG_GETARG_TEXT_PP(0);
1577 6562176 : text *arg2 = PG_GETARG_TEXT_PP(1);
1578 : int32 result;
1579 :
1580 6562176 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1581 :
1582 6562176 : PG_FREE_IF_COPY(arg1, 0);
1583 6562176 : PG_FREE_IF_COPY(arg2, 1);
1584 :
1585 6562176 : PG_RETURN_INT32(result);
1586 : }
1587 :
1588 : Datum
1589 106538 : bttextsortsupport(PG_FUNCTION_ARGS)
1590 : {
1591 106538 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1592 106538 : Oid collid = ssup->ssup_collation;
1593 : MemoryContext oldcontext;
1594 :
1595 106538 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1596 :
1597 : /* Use generic string SortSupport */
1598 106538 : varstr_sortsupport(ssup, TEXTOID, collid);
1599 :
1600 106526 : MemoryContextSwitchTo(oldcontext);
1601 :
1602 106526 : PG_RETURN_VOID();
1603 : }
1604 :
1605 : /*
1606 : * Generic sortsupport interface for character type's operator classes.
1607 : * Includes locale support, and support for BpChar semantics (i.e. removing
1608 : * trailing spaces before comparison).
1609 : *
1610 : * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
1611 : * same representation. Callers that always use the C collation (e.g.
1612 : * non-collatable type callers like bytea) may have NUL bytes in their strings;
1613 : * this will not work with any other collation, though.
1614 : */
1615 : void
1616 157586 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1617 : {
1618 157586 : bool abbreviate = ssup->abbreviate;
1619 157586 : bool collate_c = false;
1620 : VarStringSortSupport *sss;
1621 : pg_locale_t locale;
1622 :
1623 157586 : check_collation_set(collid);
1624 :
1625 157574 : locale = pg_newlocale_from_collation(collid);
1626 :
1627 : /*
1628 : * If possible, set ssup->comparator to a function which can be used to
1629 : * directly compare two datums. If we can do this, we'll avoid the
1630 : * overhead of a trip through the fmgr layer for every comparison, which
1631 : * can be substantial.
1632 : *
1633 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1634 : * which uses strcoll() to perform comparisons. We use that for the
1635 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1636 : * LC_COLLATE = C, we can make things quite a bit faster with
1637 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1638 : * memcmp() rather than strcoll().
1639 : */
1640 157574 : if (locale->collate_is_c)
1641 : {
1642 111100 : if (typid == BPCHAROID)
1643 342 : ssup->comparator = bpcharfastcmp_c;
1644 110758 : else if (typid == NAMEOID)
1645 : {
1646 49918 : ssup->comparator = namefastcmp_c;
1647 : /* Not supporting abbreviation with type NAME, for now */
1648 49918 : abbreviate = false;
1649 : }
1650 : else
1651 60840 : ssup->comparator = varstrfastcmp_c;
1652 :
1653 111100 : collate_c = true;
1654 : }
1655 : else
1656 : {
1657 : /*
1658 : * We use varlenafastcmp_locale except for type NAME.
1659 : */
1660 46474 : if (typid == NAMEOID)
1661 : {
1662 0 : ssup->comparator = namefastcmp_locale;
1663 : /* Not supporting abbreviation with type NAME, for now */
1664 0 : abbreviate = false;
1665 : }
1666 : else
1667 46474 : ssup->comparator = varlenafastcmp_locale;
1668 :
1669 : /*
1670 : * Unfortunately, it seems that abbreviation for non-C collations is
1671 : * broken on many common platforms; see pg_strxfrm_enabled().
1672 : *
1673 : * Even apart from the risk of broken locales, it's possible that
1674 : * there are platforms where the use of abbreviated keys should be
1675 : * disabled at compile time. Having only 4 byte datums could make
1676 : * worst-case performance drastically more likely, for example.
1677 : * Moreover, macOS's strxfrm() implementation is known to not
1678 : * effectively concentrate a significant amount of entropy from the
1679 : * original string in earlier transformed blobs. It's possible that
1680 : * other supported platforms are similarly encumbered. So, if we ever
1681 : * get past disabling this categorically, we may still want or need to
1682 : * disable it for particular platforms.
1683 : */
1684 46474 : if (!pg_strxfrm_enabled(locale))
1685 45678 : abbreviate = false;
1686 : }
1687 :
1688 : /*
1689 : * If we're using abbreviated keys, or if we're using a locale-aware
1690 : * comparison, we need to initialize a VarStringSortSupport object. Both
1691 : * cases will make use of the temporary buffers we initialize here for
1692 : * scratch space (and to detect requirement for BpChar semantics from
1693 : * caller), and the abbreviation case requires additional state.
1694 : */
1695 157574 : if (abbreviate || !collate_c)
1696 : {
1697 89126 : sss = palloc(sizeof(VarStringSortSupport));
1698 89126 : sss->buf1 = palloc(TEXTBUFLEN);
1699 89126 : sss->buflen1 = TEXTBUFLEN;
1700 89126 : sss->buf2 = palloc(TEXTBUFLEN);
1701 89126 : sss->buflen2 = TEXTBUFLEN;
1702 : /* Start with invalid values */
1703 89126 : sss->last_len1 = -1;
1704 89126 : sss->last_len2 = -1;
1705 : /* Initialize */
1706 89126 : sss->last_returned = 0;
1707 89126 : if (collate_c)
1708 42652 : sss->locale = NULL;
1709 : else
1710 46474 : sss->locale = locale;
1711 :
1712 : /*
1713 : * To avoid somehow confusing a strxfrm() blob and an original string,
1714 : * constantly keep track of the variety of data that buf1 and buf2
1715 : * currently contain.
1716 : *
1717 : * Comparisons may be interleaved with conversion calls. Frequently,
1718 : * conversions and comparisons are batched into two distinct phases,
1719 : * but the correctness of caching cannot hinge upon this. For
1720 : * comparison caching, buffer state is only trusted if cache_blob is
1721 : * found set to false, whereas strxfrm() caching only trusts the state
1722 : * when cache_blob is found set to true.
1723 : *
1724 : * Arbitrarily initialize cache_blob to true.
1725 : */
1726 89126 : sss->cache_blob = true;
1727 89126 : sss->collate_c = collate_c;
1728 89126 : sss->typid = typid;
1729 89126 : ssup->ssup_extra = sss;
1730 :
1731 : /*
1732 : * If possible, plan to use the abbreviated keys optimization. The
1733 : * core code may switch back to authoritative comparator should
1734 : * abbreviation be aborted.
1735 : */
1736 89126 : if (abbreviate)
1737 : {
1738 43250 : sss->prop_card = 0.20;
1739 43250 : initHyperLogLog(&sss->abbr_card, 10);
1740 43250 : initHyperLogLog(&sss->full_card, 10);
1741 43250 : ssup->abbrev_full_comparator = ssup->comparator;
1742 43250 : ssup->comparator = ssup_datum_unsigned_cmp;
1743 43250 : ssup->abbrev_converter = varstr_abbrev_convert;
1744 43250 : ssup->abbrev_abort = varstr_abbrev_abort;
1745 : }
1746 : }
1747 157574 : }
1748 :
1749 : /*
1750 : * sortsupport comparison func (for C locale case)
1751 : */
1752 : static int
1753 44985952 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
1754 : {
1755 44985952 : VarString *arg1 = DatumGetVarStringPP(x);
1756 44985952 : VarString *arg2 = DatumGetVarStringPP(y);
1757 : char *a1p,
1758 : *a2p;
1759 : int len1,
1760 : len2,
1761 : result;
1762 :
1763 44985952 : a1p = VARDATA_ANY(arg1);
1764 44985952 : a2p = VARDATA_ANY(arg2);
1765 :
1766 44985952 : len1 = VARSIZE_ANY_EXHDR(arg1);
1767 44985952 : len2 = VARSIZE_ANY_EXHDR(arg2);
1768 :
1769 44985952 : result = memcmp(a1p, a2p, Min(len1, len2));
1770 44985952 : if ((result == 0) && (len1 != len2))
1771 1174108 : result = (len1 < len2) ? -1 : 1;
1772 :
1773 : /* We can't afford to leak memory here. */
1774 44985952 : if (PointerGetDatum(arg1) != x)
1775 2 : pfree(arg1);
1776 44985952 : if (PointerGetDatum(arg2) != y)
1777 2 : pfree(arg2);
1778 :
1779 44985952 : return result;
1780 : }
1781 :
1782 : /*
1783 : * sortsupport comparison func (for BpChar C locale case)
1784 : *
1785 : * BpChar outsources its sortsupport to this module. Specialization for the
1786 : * varstr_sortsupport BpChar case, modeled on
1787 : * internal_bpchar_pattern_compare().
1788 : */
1789 : static int
1790 63536 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
1791 : {
1792 63536 : BpChar *arg1 = DatumGetBpCharPP(x);
1793 63536 : BpChar *arg2 = DatumGetBpCharPP(y);
1794 : char *a1p,
1795 : *a2p;
1796 : int len1,
1797 : len2,
1798 : result;
1799 :
1800 63536 : a1p = VARDATA_ANY(arg1);
1801 63536 : a2p = VARDATA_ANY(arg2);
1802 :
1803 63536 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
1804 63536 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
1805 :
1806 63536 : result = memcmp(a1p, a2p, Min(len1, len2));
1807 63536 : if ((result == 0) && (len1 != len2))
1808 8 : result = (len1 < len2) ? -1 : 1;
1809 :
1810 : /* We can't afford to leak memory here. */
1811 63536 : if (PointerGetDatum(arg1) != x)
1812 0 : pfree(arg1);
1813 63536 : if (PointerGetDatum(arg2) != y)
1814 0 : pfree(arg2);
1815 :
1816 63536 : return result;
1817 : }
1818 :
1819 : /*
1820 : * sortsupport comparison func (for NAME C locale case)
1821 : */
1822 : static int
1823 42936436 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
1824 : {
1825 42936436 : Name arg1 = DatumGetName(x);
1826 42936436 : Name arg2 = DatumGetName(y);
1827 :
1828 42936436 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
1829 : }
1830 :
1831 : /*
1832 : * sortsupport comparison func (for locale case with all varlena types)
1833 : */
1834 : static int
1835 37406120 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
1836 : {
1837 37406120 : VarString *arg1 = DatumGetVarStringPP(x);
1838 37406120 : VarString *arg2 = DatumGetVarStringPP(y);
1839 : char *a1p,
1840 : *a2p;
1841 : int len1,
1842 : len2,
1843 : result;
1844 :
1845 37406120 : a1p = VARDATA_ANY(arg1);
1846 37406120 : a2p = VARDATA_ANY(arg2);
1847 :
1848 37406120 : len1 = VARSIZE_ANY_EXHDR(arg1);
1849 37406120 : len2 = VARSIZE_ANY_EXHDR(arg2);
1850 :
1851 37406120 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
1852 :
1853 : /* We can't afford to leak memory here. */
1854 37406120 : if (PointerGetDatum(arg1) != x)
1855 0 : pfree(arg1);
1856 37406120 : if (PointerGetDatum(arg2) != y)
1857 0 : pfree(arg2);
1858 :
1859 37406120 : return result;
1860 : }
1861 :
1862 : /*
1863 : * sortsupport comparison func (for locale case with NAME type)
1864 : */
1865 : static int
1866 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
1867 : {
1868 0 : Name arg1 = DatumGetName(x);
1869 0 : Name arg2 = DatumGetName(y);
1870 :
1871 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
1872 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
1873 : ssup);
1874 : }
1875 :
1876 : /*
1877 : * sortsupport comparison func for locale cases
1878 : */
1879 : static int
1880 37406120 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
1881 : {
1882 37406120 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1883 : int result;
1884 : bool arg1_match;
1885 :
1886 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
1887 37406120 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
1888 : {
1889 : /*
1890 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
1891 : * last_len2. Existing contents of buffers might still be used by
1892 : * next call.
1893 : *
1894 : * It's fine to allow the comparison of BpChar padding bytes here,
1895 : * even though that implies that the memcmp() will usually be
1896 : * performed for BpChar callers (though multibyte characters could
1897 : * still prevent that from occurring). The memcmp() is still very
1898 : * cheap, and BpChar's funny semantics have us remove trailing spaces
1899 : * (not limited to padding), so we need make no distinction between
1900 : * padding space characters and "real" space characters.
1901 : */
1902 9344878 : return 0;
1903 : }
1904 :
1905 28061242 : if (sss->typid == BPCHAROID)
1906 : {
1907 : /* Get true number of bytes, ignoring trailing spaces */
1908 39182 : len1 = bpchartruelen(a1p, len1);
1909 39182 : len2 = bpchartruelen(a2p, len2);
1910 : }
1911 :
1912 28061242 : if (len1 >= sss->buflen1)
1913 : {
1914 10 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
1915 10 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
1916 : }
1917 28061242 : if (len2 >= sss->buflen2)
1918 : {
1919 6 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
1920 6 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
1921 : }
1922 :
1923 : /*
1924 : * We're likely to be asked to compare the same strings repeatedly, and
1925 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
1926 : * comparisons, even though in general there is no reason to think that
1927 : * that will work out (every string datum may be unique). Caching does
1928 : * not slow things down measurably when it doesn't work out, and can speed
1929 : * things up by rather a lot when it does. In part, this is because the
1930 : * memcmp() compares data from cachelines that are needed in L1 cache even
1931 : * when the last comparison's result cannot be reused.
1932 : */
1933 28061242 : arg1_match = true;
1934 28061242 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
1935 : {
1936 25948504 : arg1_match = false;
1937 25948504 : memcpy(sss->buf1, a1p, len1);
1938 25948504 : sss->buf1[len1] = '\0';
1939 25948504 : sss->last_len1 = len1;
1940 : }
1941 :
1942 : /*
1943 : * If we're comparing the same two strings as last time, we can return the
1944 : * same answer without calling strcoll() again. This is more likely than
1945 : * it seems (at least with moderate to low cardinality sets), because
1946 : * quicksort compares the same pivot against many values.
1947 : */
1948 28061242 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
1949 : {
1950 4298550 : memcpy(sss->buf2, a2p, len2);
1951 4298550 : sss->buf2[len2] = '\0';
1952 4298550 : sss->last_len2 = len2;
1953 : }
1954 23762692 : else if (arg1_match && !sss->cache_blob)
1955 : {
1956 : /* Use result cached following last actual strcoll() call */
1957 1662872 : return sss->last_returned;
1958 : }
1959 :
1960 26398370 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
1961 :
1962 : /* Break tie if necessary. */
1963 26398370 : if (result == 0 && sss->locale->deterministic)
1964 0 : result = strcmp(sss->buf1, sss->buf2);
1965 :
1966 : /* Cache result, perhaps saving an expensive strcoll() call next time */
1967 26398370 : sss->cache_blob = false;
1968 26398370 : sss->last_returned = result;
1969 26398370 : return result;
1970 : }
1971 :
1972 : /*
1973 : * Conversion routine for sortsupport. Converts original to abbreviated key
1974 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
1975 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
1976 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
1977 : * locale is used, or in case of bytea, just memcpy() from original instead.
1978 : */
1979 : static Datum
1980 885046 : varstr_abbrev_convert(Datum original, SortSupport ssup)
1981 : {
1982 885046 : const size_t max_prefix_bytes = sizeof(Datum);
1983 885046 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1984 885046 : VarString *authoritative = DatumGetVarStringPP(original);
1985 885046 : char *authoritative_data = VARDATA_ANY(authoritative);
1986 :
1987 : /* working state */
1988 : Datum res;
1989 : char *pres;
1990 : int len;
1991 : uint32 hash;
1992 :
1993 885046 : pres = (char *) &res;
1994 : /* memset(), so any non-overwritten bytes are NUL */
1995 885046 : memset(pres, 0, max_prefix_bytes);
1996 885046 : len = VARSIZE_ANY_EXHDR(authoritative);
1997 :
1998 : /* Get number of bytes, ignoring trailing spaces */
1999 885046 : if (sss->typid == BPCHAROID)
2000 1362 : len = bpchartruelen(authoritative_data, len);
2001 :
2002 : /*
2003 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2004 : * abbreviate keys. The full comparator for the C locale is always
2005 : * memcmp(). It would be incorrect to allow bytea callers (callers that
2006 : * always force the C collation -- bytea isn't a collatable type, but this
2007 : * approach is convenient) to use strxfrm(). This is because bytea
2008 : * strings may contain NUL bytes. Besides, this should be faster, too.
2009 : *
2010 : * More generally, it's okay that bytea callers can have NUL bytes in
2011 : * strings because abbreviated cmp need not make a distinction between
2012 : * terminating NUL bytes, and NUL bytes representing actual NULs in the
2013 : * authoritative representation. Hopefully a comparison at or past one
2014 : * abbreviated key's terminating NUL byte will resolve the comparison
2015 : * without consulting the authoritative representation; specifically, some
2016 : * later non-NUL byte in the longer string can resolve the comparison
2017 : * against a subsequent terminating NUL in the shorter string. There will
2018 : * usually be what is effectively a "length-wise" resolution there and
2019 : * then.
2020 : *
2021 : * If that doesn't work out -- if all bytes in the longer string
2022 : * positioned at or past the offset of the smaller string's (first)
2023 : * terminating NUL are actually representative of NUL bytes in the
2024 : * authoritative binary string (perhaps with some *terminating* NUL bytes
2025 : * towards the end of the longer string iff it happens to still be small)
2026 : * -- then an authoritative tie-breaker will happen, and do the right
2027 : * thing: explicitly consider string length.
2028 : */
2029 885046 : if (sss->collate_c)
2030 883210 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
2031 : else
2032 : {
2033 : Size bsize;
2034 :
2035 : /*
2036 : * We're not using the C collation, so fall back on strxfrm or ICU
2037 : * analogs.
2038 : */
2039 :
2040 : /* By convention, we use buffer 1 to store and NUL-terminate */
2041 1836 : if (len >= sss->buflen1)
2042 : {
2043 0 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2044 0 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2045 : }
2046 :
2047 : /* Might be able to reuse strxfrm() blob from last call */
2048 1836 : if (sss->last_len1 == len && sss->cache_blob &&
2049 918 : memcmp(sss->buf1, authoritative_data, len) == 0)
2050 : {
2051 168 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, sss->last_len2));
2052 : /* No change affecting cardinality, so no hashing required */
2053 168 : goto done;
2054 : }
2055 :
2056 1668 : memcpy(sss->buf1, authoritative_data, len);
2057 :
2058 : /*
2059 : * pg_strxfrm() and pg_strxfrm_prefix expect NUL-terminated strings.
2060 : */
2061 1668 : sss->buf1[len] = '\0';
2062 1668 : sss->last_len1 = len;
2063 :
2064 1668 : if (pg_strxfrm_prefix_enabled(sss->locale))
2065 : {
2066 1668 : if (sss->buflen2 < max_prefix_bytes)
2067 : {
2068 0 : sss->buflen2 = Max(max_prefix_bytes,
2069 : Min(sss->buflen2 * 2, MaxAllocSize));
2070 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2071 : }
2072 :
2073 1668 : bsize = pg_strxfrm_prefix(sss->buf2, sss->buf1,
2074 : max_prefix_bytes, sss->locale);
2075 1668 : sss->last_len2 = bsize;
2076 : }
2077 : else
2078 : {
2079 : /*
2080 : * Loop: Call pg_strxfrm(), possibly enlarge buffer, and try
2081 : * again. The pg_strxfrm() function leaves the result buffer
2082 : * content undefined if the result did not fit, so we need to
2083 : * retry until everything fits, even though we only need the first
2084 : * few bytes in the end.
2085 : */
2086 : for (;;)
2087 : {
2088 0 : bsize = pg_strxfrm(sss->buf2, sss->buf1, sss->buflen2,
2089 : sss->locale);
2090 :
2091 0 : sss->last_len2 = bsize;
2092 0 : if (bsize < sss->buflen2)
2093 0 : break;
2094 :
2095 : /*
2096 : * Grow buffer and retry.
2097 : */
2098 0 : sss->buflen2 = Max(bsize + 1,
2099 : Min(sss->buflen2 * 2, MaxAllocSize));
2100 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2101 : }
2102 : }
2103 :
2104 : /*
2105 : * Every Datum byte is always compared. This is safe because the
2106 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2107 : * misinterpreting any NUL bytes not intended to be interpreted as
2108 : * logically representing termination.
2109 : *
2110 : * (Actually, even if there were NUL bytes in the blob it would be
2111 : * okay. See remarks on bytea case above.)
2112 : */
2113 1668 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, bsize));
2114 : }
2115 :
2116 : /*
2117 : * Maintain approximate cardinality of both abbreviated keys and original,
2118 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2119 : * the worst case, where we do many string transformations for no saving
2120 : * in full strcoll()-based comparisons. These statistics are used by
2121 : * varstr_abbrev_abort().
2122 : *
2123 : * First, Hash key proper, or a significant fraction of it. Mix in length
2124 : * in order to compensate for cases where differences are past
2125 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2126 : */
2127 884878 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2128 : Min(len, PG_CACHE_LINE_SIZE)));
2129 :
2130 884878 : if (len > PG_CACHE_LINE_SIZE)
2131 458 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2132 :
2133 884878 : addHyperLogLog(&sss->full_card, hash);
2134 :
2135 : /* Hash abbreviated key */
2136 : #if SIZEOF_DATUM == 8
2137 : {
2138 : uint32 lohalf,
2139 : hihalf;
2140 :
2141 884878 : lohalf = (uint32) res;
2142 884878 : hihalf = (uint32) (res >> 32);
2143 884878 : hash = DatumGetUInt32(hash_uint32(lohalf ^ hihalf));
2144 : }
2145 : #else /* SIZEOF_DATUM != 8 */
2146 : hash = DatumGetUInt32(hash_uint32((uint32) res));
2147 : #endif
2148 :
2149 884878 : addHyperLogLog(&sss->abbr_card, hash);
2150 :
2151 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2152 884878 : sss->cache_blob = true;
2153 885046 : done:
2154 :
2155 : /*
2156 : * Byteswap on little-endian machines.
2157 : *
2158 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2159 : * 3-way comparator) works correctly on all platforms. If we didn't do
2160 : * this, the comparator would have to call memcmp() with a pair of
2161 : * pointers to the first byte of each abbreviated key, which is slower.
2162 : */
2163 885046 : res = DatumBigEndianToNative(res);
2164 :
2165 : /* Don't leak memory here */
2166 885046 : if (PointerGetDatum(authoritative) != original)
2167 8 : pfree(authoritative);
2168 :
2169 885046 : return res;
2170 : }
2171 :
2172 : /*
2173 : * Callback for estimating effectiveness of abbreviated key optimization, using
2174 : * heuristic rules. Returns value indicating if the abbreviation optimization
2175 : * should be aborted, based on its projected effectiveness.
2176 : */
2177 : static bool
2178 2378 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2179 : {
2180 2378 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2181 : double abbrev_distinct,
2182 : key_distinct;
2183 :
2184 : Assert(ssup->abbreviate);
2185 :
2186 : /* Have a little patience */
2187 2378 : if (memtupcount < 100)
2188 1358 : return false;
2189 :
2190 1020 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2191 1020 : key_distinct = estimateHyperLogLog(&sss->full_card);
2192 :
2193 : /*
2194 : * Clamp cardinality estimates to at least one distinct value. While
2195 : * NULLs are generally disregarded, if only NULL values were seen so far,
2196 : * that might misrepresent costs if we failed to clamp.
2197 : */
2198 1020 : if (abbrev_distinct <= 1.0)
2199 0 : abbrev_distinct = 1.0;
2200 :
2201 1020 : if (key_distinct <= 1.0)
2202 0 : key_distinct = 1.0;
2203 :
2204 : /*
2205 : * In the worst case all abbreviated keys are identical, while at the same
2206 : * time there are differences within full key strings not captured in
2207 : * abbreviations.
2208 : */
2209 1020 : if (trace_sort)
2210 : {
2211 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2212 :
2213 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2214 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2215 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2216 : sss->prop_card);
2217 : }
2218 :
2219 : /*
2220 : * If the number of distinct abbreviated keys approximately matches the
2221 : * number of distinct authoritative original keys, that's reason enough to
2222 : * proceed. We can win even with a very low cardinality set if most
2223 : * tie-breakers only memcmp(). This is by far the most important
2224 : * consideration.
2225 : *
2226 : * While comparisons that are resolved at the abbreviated key level are
2227 : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2228 : * those two outcomes are so much cheaper than a full strcoll() once
2229 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2230 : * cardinality against the overall size of the set in order to more
2231 : * accurately model costs. Assume that an abbreviated comparison, and an
2232 : * abbreviated comparison with a cheap memcmp()-based authoritative
2233 : * resolution are equivalent.
2234 : */
2235 1020 : if (abbrev_distinct > key_distinct * sss->prop_card)
2236 : {
2237 : /*
2238 : * When we have exceeded 10,000 tuples, decay required cardinality
2239 : * aggressively for next call.
2240 : *
2241 : * This is useful because the number of comparisons required on
2242 : * average increases at a linearithmic rate, and at roughly 10,000
2243 : * tuples that factor will start to dominate over the linear costs of
2244 : * string transformation (this is a conservative estimate). The decay
2245 : * rate is chosen to be a little less aggressive than halving -- which
2246 : * (since we're called at points at which memtupcount has doubled)
2247 : * would never see the cost model actually abort past the first call
2248 : * following a decay. This decay rate is mostly a precaution against
2249 : * a sudden, violent swing in how well abbreviated cardinality tracks
2250 : * full key cardinality. The decay also serves to prevent a marginal
2251 : * case from being aborted too late, when too much has already been
2252 : * invested in string transformation.
2253 : *
2254 : * It's possible for sets of several million distinct strings with
2255 : * mere tens of thousands of distinct abbreviated keys to still
2256 : * benefit very significantly. This will generally occur provided
2257 : * each abbreviated key is a proxy for a roughly uniform number of the
2258 : * set's full keys. If it isn't so, we hope to catch that early and
2259 : * abort. If it isn't caught early, by the time the problem is
2260 : * apparent it's probably not worth aborting.
2261 : */
2262 1020 : if (memtupcount > 10000)
2263 4 : sss->prop_card *= 0.65;
2264 :
2265 1020 : return false;
2266 : }
2267 :
2268 : /*
2269 : * Abort abbreviation strategy.
2270 : *
2271 : * The worst case, where all abbreviated keys are identical while all
2272 : * original strings differ will typically only see a regression of about
2273 : * 10% in execution time for small to medium sized lists of strings.
2274 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2275 : * often expect very large improvements, particularly with sets of strings
2276 : * of moderately high to high abbreviated cardinality. There is little to
2277 : * lose but much to gain, which our strategy reflects.
2278 : */
2279 0 : if (trace_sort)
2280 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2281 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2282 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2283 :
2284 0 : return true;
2285 : }
2286 :
2287 : /*
2288 : * Generic equalimage support function for character type's operator classes.
2289 : * Disables the use of deduplication with nondeterministic collations.
2290 : */
2291 : Datum
2292 8848 : btvarstrequalimage(PG_FUNCTION_ARGS)
2293 : {
2294 : /* Oid opcintype = PG_GETARG_OID(0); */
2295 8848 : Oid collid = PG_GET_COLLATION();
2296 : pg_locale_t locale;
2297 :
2298 8848 : check_collation_set(collid);
2299 :
2300 8848 : locale = pg_newlocale_from_collation(collid);
2301 :
2302 8848 : PG_RETURN_BOOL(locale->deterministic);
2303 : }
2304 :
2305 : Datum
2306 229560 : text_larger(PG_FUNCTION_ARGS)
2307 : {
2308 229560 : text *arg1 = PG_GETARG_TEXT_PP(0);
2309 229560 : text *arg2 = PG_GETARG_TEXT_PP(1);
2310 : text *result;
2311 :
2312 229560 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2313 :
2314 229560 : PG_RETURN_TEXT_P(result);
2315 : }
2316 :
2317 : Datum
2318 86076 : text_smaller(PG_FUNCTION_ARGS)
2319 : {
2320 86076 : text *arg1 = PG_GETARG_TEXT_PP(0);
2321 86076 : text *arg2 = PG_GETARG_TEXT_PP(1);
2322 : text *result;
2323 :
2324 86076 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2325 :
2326 86076 : PG_RETURN_TEXT_P(result);
2327 : }
2328 :
2329 :
2330 : /*
2331 : * Cross-type comparison functions for types text and name.
2332 : */
2333 :
2334 : Datum
2335 210754 : nameeqtext(PG_FUNCTION_ARGS)
2336 : {
2337 210754 : Name arg1 = PG_GETARG_NAME(0);
2338 210754 : text *arg2 = PG_GETARG_TEXT_PP(1);
2339 210754 : size_t len1 = strlen(NameStr(*arg1));
2340 210754 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2341 210754 : Oid collid = PG_GET_COLLATION();
2342 : bool result;
2343 :
2344 210754 : check_collation_set(collid);
2345 :
2346 210754 : if (collid == C_COLLATION_OID)
2347 256088 : result = (len1 == len2 &&
2348 124078 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2349 : else
2350 78744 : result = (varstr_cmp(NameStr(*arg1), len1,
2351 78744 : VARDATA_ANY(arg2), len2,
2352 : collid) == 0);
2353 :
2354 210754 : PG_FREE_IF_COPY(arg2, 1);
2355 :
2356 210754 : PG_RETURN_BOOL(result);
2357 : }
2358 :
2359 : Datum
2360 7972 : texteqname(PG_FUNCTION_ARGS)
2361 : {
2362 7972 : text *arg1 = PG_GETARG_TEXT_PP(0);
2363 7972 : Name arg2 = PG_GETARG_NAME(1);
2364 7972 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2365 7972 : size_t len2 = strlen(NameStr(*arg2));
2366 7972 : Oid collid = PG_GET_COLLATION();
2367 : bool result;
2368 :
2369 7972 : check_collation_set(collid);
2370 :
2371 7972 : if (collid == C_COLLATION_OID)
2372 568 : result = (len1 == len2 &&
2373 182 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2374 : else
2375 7586 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2376 7586 : NameStr(*arg2), len2,
2377 : collid) == 0);
2378 :
2379 7972 : PG_FREE_IF_COPY(arg1, 0);
2380 :
2381 7972 : PG_RETURN_BOOL(result);
2382 : }
2383 :
2384 : Datum
2385 18 : namenetext(PG_FUNCTION_ARGS)
2386 : {
2387 18 : Name arg1 = PG_GETARG_NAME(0);
2388 18 : text *arg2 = PG_GETARG_TEXT_PP(1);
2389 18 : size_t len1 = strlen(NameStr(*arg1));
2390 18 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2391 18 : Oid collid = PG_GET_COLLATION();
2392 : bool result;
2393 :
2394 18 : check_collation_set(collid);
2395 :
2396 18 : if (collid == C_COLLATION_OID)
2397 0 : result = !(len1 == len2 &&
2398 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2399 : else
2400 18 : result = !(varstr_cmp(NameStr(*arg1), len1,
2401 18 : VARDATA_ANY(arg2), len2,
2402 : collid) == 0);
2403 :
2404 18 : PG_FREE_IF_COPY(arg2, 1);
2405 :
2406 18 : PG_RETURN_BOOL(result);
2407 : }
2408 :
2409 : Datum
2410 18 : textnename(PG_FUNCTION_ARGS)
2411 : {
2412 18 : text *arg1 = PG_GETARG_TEXT_PP(0);
2413 18 : Name arg2 = PG_GETARG_NAME(1);
2414 18 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2415 18 : size_t len2 = strlen(NameStr(*arg2));
2416 18 : Oid collid = PG_GET_COLLATION();
2417 : bool result;
2418 :
2419 18 : check_collation_set(collid);
2420 :
2421 18 : if (collid == C_COLLATION_OID)
2422 0 : result = !(len1 == len2 &&
2423 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2424 : else
2425 18 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2426 18 : NameStr(*arg2), len2,
2427 : collid) == 0);
2428 :
2429 18 : PG_FREE_IF_COPY(arg1, 0);
2430 :
2431 18 : PG_RETURN_BOOL(result);
2432 : }
2433 :
2434 : Datum
2435 129122 : btnametextcmp(PG_FUNCTION_ARGS)
2436 : {
2437 129122 : Name arg1 = PG_GETARG_NAME(0);
2438 129122 : text *arg2 = PG_GETARG_TEXT_PP(1);
2439 : int32 result;
2440 :
2441 258244 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2442 258244 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2443 : PG_GET_COLLATION());
2444 :
2445 129122 : PG_FREE_IF_COPY(arg2, 1);
2446 :
2447 129122 : PG_RETURN_INT32(result);
2448 : }
2449 :
2450 : Datum
2451 44 : bttextnamecmp(PG_FUNCTION_ARGS)
2452 : {
2453 44 : text *arg1 = PG_GETARG_TEXT_PP(0);
2454 44 : Name arg2 = PG_GETARG_NAME(1);
2455 : int32 result;
2456 :
2457 44 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2458 44 : NameStr(*arg2), strlen(NameStr(*arg2)),
2459 : PG_GET_COLLATION());
2460 :
2461 44 : PG_FREE_IF_COPY(arg1, 0);
2462 :
2463 44 : PG_RETURN_INT32(result);
2464 : }
2465 :
2466 : #define CmpCall(cmpfunc) \
2467 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2468 : PG_GET_COLLATION(), \
2469 : PG_GETARG_DATUM(0), \
2470 : PG_GETARG_DATUM(1)))
2471 :
2472 : Datum
2473 62198 : namelttext(PG_FUNCTION_ARGS)
2474 : {
2475 62198 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2476 : }
2477 :
2478 : Datum
2479 0 : nameletext(PG_FUNCTION_ARGS)
2480 : {
2481 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2482 : }
2483 :
2484 : Datum
2485 0 : namegttext(PG_FUNCTION_ARGS)
2486 : {
2487 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2488 : }
2489 :
2490 : Datum
2491 54276 : namegetext(PG_FUNCTION_ARGS)
2492 : {
2493 54276 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2494 : }
2495 :
2496 : Datum
2497 0 : textltname(PG_FUNCTION_ARGS)
2498 : {
2499 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
2500 : }
2501 :
2502 : Datum
2503 0 : textlename(PG_FUNCTION_ARGS)
2504 : {
2505 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2506 : }
2507 :
2508 : Datum
2509 0 : textgtname(PG_FUNCTION_ARGS)
2510 : {
2511 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2512 : }
2513 :
2514 : Datum
2515 0 : textgename(PG_FUNCTION_ARGS)
2516 : {
2517 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2518 : }
2519 :
2520 : #undef CmpCall
2521 :
2522 :
2523 : /*
2524 : * The following operators support character-by-character comparison
2525 : * of text datums, to allow building indexes suitable for LIKE clauses.
2526 : * Note that the regular texteq/textne comparison operators, and regular
2527 : * support functions 1 and 2 with "C" collation are assumed to be
2528 : * compatible with these!
2529 : */
2530 :
2531 : static int
2532 160444 : internal_text_pattern_compare(text *arg1, text *arg2)
2533 : {
2534 : int result;
2535 : int len1,
2536 : len2;
2537 :
2538 160444 : len1 = VARSIZE_ANY_EXHDR(arg1);
2539 160444 : len2 = VARSIZE_ANY_EXHDR(arg2);
2540 :
2541 160444 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2542 160444 : if (result != 0)
2543 160312 : return result;
2544 132 : else if (len1 < len2)
2545 0 : return -1;
2546 132 : else if (len1 > len2)
2547 84 : return 1;
2548 : else
2549 48 : return 0;
2550 : }
2551 :
2552 :
2553 : Datum
2554 47866 : text_pattern_lt(PG_FUNCTION_ARGS)
2555 : {
2556 47866 : text *arg1 = PG_GETARG_TEXT_PP(0);
2557 47866 : text *arg2 = PG_GETARG_TEXT_PP(1);
2558 : int result;
2559 :
2560 47866 : result = internal_text_pattern_compare(arg1, arg2);
2561 :
2562 47866 : PG_FREE_IF_COPY(arg1, 0);
2563 47866 : PG_FREE_IF_COPY(arg2, 1);
2564 :
2565 47866 : PG_RETURN_BOOL(result < 0);
2566 : }
2567 :
2568 :
2569 : Datum
2570 37510 : text_pattern_le(PG_FUNCTION_ARGS)
2571 : {
2572 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2573 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2574 : int result;
2575 :
2576 37510 : result = internal_text_pattern_compare(arg1, arg2);
2577 :
2578 37510 : PG_FREE_IF_COPY(arg1, 0);
2579 37510 : PG_FREE_IF_COPY(arg2, 1);
2580 :
2581 37510 : PG_RETURN_BOOL(result <= 0);
2582 : }
2583 :
2584 :
2585 : Datum
2586 37534 : text_pattern_ge(PG_FUNCTION_ARGS)
2587 : {
2588 37534 : text *arg1 = PG_GETARG_TEXT_PP(0);
2589 37534 : text *arg2 = PG_GETARG_TEXT_PP(1);
2590 : int result;
2591 :
2592 37534 : result = internal_text_pattern_compare(arg1, arg2);
2593 :
2594 37534 : PG_FREE_IF_COPY(arg1, 0);
2595 37534 : PG_FREE_IF_COPY(arg2, 1);
2596 :
2597 37534 : PG_RETURN_BOOL(result >= 0);
2598 : }
2599 :
2600 :
2601 : Datum
2602 37510 : text_pattern_gt(PG_FUNCTION_ARGS)
2603 : {
2604 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2605 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2606 : int result;
2607 :
2608 37510 : result = internal_text_pattern_compare(arg1, arg2);
2609 :
2610 37510 : PG_FREE_IF_COPY(arg1, 0);
2611 37510 : PG_FREE_IF_COPY(arg2, 1);
2612 :
2613 37510 : PG_RETURN_BOOL(result > 0);
2614 : }
2615 :
2616 :
2617 : Datum
2618 24 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
2619 : {
2620 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
2621 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
2622 : int result;
2623 :
2624 24 : result = internal_text_pattern_compare(arg1, arg2);
2625 :
2626 24 : PG_FREE_IF_COPY(arg1, 0);
2627 24 : PG_FREE_IF_COPY(arg2, 1);
2628 :
2629 24 : PG_RETURN_INT32(result);
2630 : }
2631 :
2632 :
2633 : Datum
2634 118 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
2635 : {
2636 118 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
2637 : MemoryContext oldcontext;
2638 :
2639 118 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2640 :
2641 : /* Use generic string SortSupport, forcing "C" collation */
2642 118 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
2643 :
2644 118 : MemoryContextSwitchTo(oldcontext);
2645 :
2646 118 : PG_RETURN_VOID();
2647 : }
2648 :
2649 :
2650 : /* text_name()
2651 : * Converts a text type to a Name type.
2652 : */
2653 : Datum
2654 30782 : text_name(PG_FUNCTION_ARGS)
2655 : {
2656 30782 : text *s = PG_GETARG_TEXT_PP(0);
2657 : Name result;
2658 : int len;
2659 :
2660 30782 : len = VARSIZE_ANY_EXHDR(s);
2661 :
2662 : /* Truncate oversize input */
2663 30782 : if (len >= NAMEDATALEN)
2664 6 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
2665 :
2666 : /* We use palloc0 here to ensure result is zero-padded */
2667 30782 : result = (Name) palloc0(NAMEDATALEN);
2668 30782 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
2669 :
2670 30782 : PG_RETURN_NAME(result);
2671 : }
2672 :
2673 : /* name_text()
2674 : * Converts a Name type to a text type.
2675 : */
2676 : Datum
2677 654138 : name_text(PG_FUNCTION_ARGS)
2678 : {
2679 654138 : Name s = PG_GETARG_NAME(0);
2680 :
2681 654138 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
2682 : }
2683 :
2684 :
2685 : /*
2686 : * textToQualifiedNameList - convert a text object to list of names
2687 : *
2688 : * This implements the input parsing needed by nextval() and other
2689 : * functions that take a text parameter representing a qualified name.
2690 : * We split the name at dots, downcase if not double-quoted, and
2691 : * truncate names if they're too long.
2692 : */
2693 : List *
2694 5430 : textToQualifiedNameList(text *textval)
2695 : {
2696 : char *rawname;
2697 5430 : List *result = NIL;
2698 : List *namelist;
2699 : ListCell *l;
2700 :
2701 : /* Convert to C string (handles possible detoasting). */
2702 : /* Note we rely on being able to modify rawname below. */
2703 5430 : rawname = text_to_cstring(textval);
2704 :
2705 5430 : if (!SplitIdentifierString(rawname, '.', &namelist))
2706 0 : ereport(ERROR,
2707 : (errcode(ERRCODE_INVALID_NAME),
2708 : errmsg("invalid name syntax")));
2709 :
2710 5430 : if (namelist == NIL)
2711 0 : ereport(ERROR,
2712 : (errcode(ERRCODE_INVALID_NAME),
2713 : errmsg("invalid name syntax")));
2714 :
2715 10976 : foreach(l, namelist)
2716 : {
2717 5546 : char *curname = (char *) lfirst(l);
2718 :
2719 5546 : result = lappend(result, makeString(pstrdup(curname)));
2720 : }
2721 :
2722 5430 : pfree(rawname);
2723 5430 : list_free(namelist);
2724 :
2725 5430 : return result;
2726 : }
2727 :
2728 : /*
2729 : * SplitIdentifierString --- parse a string containing identifiers
2730 : *
2731 : * This is the guts of textToQualifiedNameList, and is exported for use in
2732 : * other situations such as parsing GUC variables. In the GUC case, it's
2733 : * important to avoid memory leaks, so the API is designed to minimize the
2734 : * amount of stuff that needs to be allocated and freed.
2735 : *
2736 : * Inputs:
2737 : * rawstring: the input string; must be overwritable! On return, it's
2738 : * been modified to contain the separated identifiers.
2739 : * separator: the separator punctuation expected between identifiers
2740 : * (typically '.' or ','). Whitespace may also appear around
2741 : * identifiers.
2742 : * Outputs:
2743 : * namelist: filled with a palloc'd list of pointers to identifiers within
2744 : * rawstring. Caller should list_free() this even on error return.
2745 : *
2746 : * Returns true if okay, false if there is a syntax error in the string.
2747 : *
2748 : * Note that an empty string is considered okay here, though not in
2749 : * textToQualifiedNameList.
2750 : */
2751 : bool
2752 369238 : SplitIdentifierString(char *rawstring, char separator,
2753 : List **namelist)
2754 : {
2755 369238 : char *nextp = rawstring;
2756 369238 : bool done = false;
2757 :
2758 369238 : *namelist = NIL;
2759 :
2760 369244 : while (scanner_isspace(*nextp))
2761 6 : nextp++; /* skip leading whitespace */
2762 :
2763 369238 : if (*nextp == '\0')
2764 30596 : return true; /* allow empty string */
2765 :
2766 : /* At the top of the loop, we are at start of a new identifier. */
2767 : do
2768 : {
2769 : char *curname;
2770 : char *endp;
2771 :
2772 631352 : if (*nextp == '"')
2773 : {
2774 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
2775 40166 : curname = nextp + 1;
2776 : for (;;)
2777 : {
2778 40170 : endp = strchr(nextp + 1, '"');
2779 40168 : if (endp == NULL)
2780 0 : return false; /* mismatched quotes */
2781 40168 : if (endp[1] != '"')
2782 40166 : break; /* found end of quoted name */
2783 : /* Collapse adjacent quotes into one quote, and look again */
2784 2 : memmove(endp, endp + 1, strlen(endp));
2785 2 : nextp = endp;
2786 : }
2787 : /* endp now points at the terminating quote */
2788 40166 : nextp = endp + 1;
2789 : }
2790 : else
2791 : {
2792 : /* Unquoted name --- extends to separator or whitespace */
2793 : char *downname;
2794 : int len;
2795 :
2796 591186 : curname = nextp;
2797 5425562 : while (*nextp && *nextp != separator &&
2798 4834378 : !scanner_isspace(*nextp))
2799 4834376 : nextp++;
2800 591186 : endp = nextp;
2801 591186 : if (curname == nextp)
2802 0 : return false; /* empty unquoted name not allowed */
2803 :
2804 : /*
2805 : * Downcase the identifier, using same code as main lexer does.
2806 : *
2807 : * XXX because we want to overwrite the input in-place, we cannot
2808 : * support a downcasing transformation that increases the string
2809 : * length. This is not a problem given the current implementation
2810 : * of downcase_truncate_identifier, but we'll probably have to do
2811 : * something about this someday.
2812 : */
2813 591186 : len = endp - curname;
2814 591186 : downname = downcase_truncate_identifier(curname, len, false);
2815 : Assert(strlen(downname) <= len);
2816 591186 : strncpy(curname, downname, len); /* strncpy is required here */
2817 591186 : pfree(downname);
2818 : }
2819 :
2820 631354 : while (scanner_isspace(*nextp))
2821 2 : nextp++; /* skip trailing whitespace */
2822 :
2823 631352 : if (*nextp == separator)
2824 : {
2825 292710 : nextp++;
2826 560010 : while (scanner_isspace(*nextp))
2827 267300 : nextp++; /* skip leading whitespace for next */
2828 : /* we expect another name, so done remains false */
2829 : }
2830 338642 : else if (*nextp == '\0')
2831 338640 : done = true;
2832 : else
2833 2 : return false; /* invalid syntax */
2834 :
2835 : /* Now safe to overwrite separator with a null */
2836 631350 : *endp = '\0';
2837 :
2838 : /* Truncate name if it's overlength */
2839 631350 : truncate_identifier(curname, strlen(curname), false);
2840 :
2841 : /*
2842 : * Finished isolating current name --- add it to list
2843 : */
2844 631350 : *namelist = lappend(*namelist, curname);
2845 :
2846 : /* Loop back if we didn't reach end of string */
2847 631350 : } while (!done);
2848 :
2849 338640 : return true;
2850 : }
2851 :
2852 :
2853 : /*
2854 : * SplitDirectoriesString --- parse a string containing file/directory names
2855 : *
2856 : * This works fine on file names too; the function name is historical.
2857 : *
2858 : * This is similar to SplitIdentifierString, except that the parsing
2859 : * rules are meant to handle pathnames instead of identifiers: there is
2860 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
2861 : * and we apply canonicalize_path() to each extracted string. Because of the
2862 : * last, the returned strings are separately palloc'd rather than being
2863 : * pointers into rawstring --- but we still scribble on rawstring.
2864 : *
2865 : * Inputs:
2866 : * rawstring: the input string; must be modifiable!
2867 : * separator: the separator punctuation expected between directories
2868 : * (typically ',' or ';'). Whitespace may also appear around
2869 : * directories.
2870 : * Outputs:
2871 : * namelist: filled with a palloc'd list of directory names.
2872 : * Caller should list_free_deep() this even on error return.
2873 : *
2874 : * Returns true if okay, false if there is a syntax error in the string.
2875 : *
2876 : * Note that an empty string is considered okay here.
2877 : */
2878 : bool
2879 1780 : SplitDirectoriesString(char *rawstring, char separator,
2880 : List **namelist)
2881 : {
2882 1780 : char *nextp = rawstring;
2883 1780 : bool done = false;
2884 :
2885 1780 : *namelist = NIL;
2886 :
2887 1780 : while (scanner_isspace(*nextp))
2888 0 : nextp++; /* skip leading whitespace */
2889 :
2890 1780 : if (*nextp == '\0')
2891 2 : return true; /* allow empty string */
2892 :
2893 : /* At the top of the loop, we are at start of a new directory. */
2894 : do
2895 : {
2896 : char *curname;
2897 : char *endp;
2898 :
2899 1780 : if (*nextp == '"')
2900 : {
2901 : /* Quoted name --- collapse quote-quote pairs */
2902 0 : curname = nextp + 1;
2903 : for (;;)
2904 : {
2905 0 : endp = strchr(nextp + 1, '"');
2906 0 : if (endp == NULL)
2907 0 : return false; /* mismatched quotes */
2908 0 : if (endp[1] != '"')
2909 0 : break; /* found end of quoted name */
2910 : /* Collapse adjacent quotes into one quote, and look again */
2911 0 : memmove(endp, endp + 1, strlen(endp));
2912 0 : nextp = endp;
2913 : }
2914 : /* endp now points at the terminating quote */
2915 0 : nextp = endp + 1;
2916 : }
2917 : else
2918 : {
2919 : /* Unquoted name --- extends to separator or end of string */
2920 1780 : curname = endp = nextp;
2921 29748 : while (*nextp && *nextp != separator)
2922 : {
2923 : /* trailing whitespace should not be included in name */
2924 27968 : if (!scanner_isspace(*nextp))
2925 27968 : endp = nextp + 1;
2926 27968 : nextp++;
2927 : }
2928 1780 : if (curname == endp)
2929 0 : return false; /* empty unquoted name not allowed */
2930 : }
2931 :
2932 1780 : while (scanner_isspace(*nextp))
2933 0 : nextp++; /* skip trailing whitespace */
2934 :
2935 1780 : if (*nextp == separator)
2936 : {
2937 2 : nextp++;
2938 2 : while (scanner_isspace(*nextp))
2939 0 : nextp++; /* skip leading whitespace for next */
2940 : /* we expect another name, so done remains false */
2941 : }
2942 1778 : else if (*nextp == '\0')
2943 1778 : done = true;
2944 : else
2945 0 : return false; /* invalid syntax */
2946 :
2947 : /* Now safe to overwrite separator with a null */
2948 1780 : *endp = '\0';
2949 :
2950 : /* Truncate path if it's overlength */
2951 1780 : if (strlen(curname) >= MAXPGPATH)
2952 0 : curname[MAXPGPATH - 1] = '\0';
2953 :
2954 : /*
2955 : * Finished isolating current name --- add it to list
2956 : */
2957 1780 : curname = pstrdup(curname);
2958 1780 : canonicalize_path(curname);
2959 1780 : *namelist = lappend(*namelist, curname);
2960 :
2961 : /* Loop back if we didn't reach end of string */
2962 1780 : } while (!done);
2963 :
2964 1778 : return true;
2965 : }
2966 :
2967 :
2968 : /*
2969 : * SplitGUCList --- parse a string containing identifiers or file names
2970 : *
2971 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
2972 : * presuming whether the elements will be taken as identifiers or file names.
2973 : * We assume the input has already been through flatten_set_variable_args(),
2974 : * so that we need never downcase (if appropriate, that was done already).
2975 : * Nor do we ever truncate, since we don't know the correct max length.
2976 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
2977 : * because any embedded whitespace should have led to double-quoting).
2978 : * Otherwise the API is identical to SplitIdentifierString.
2979 : *
2980 : * XXX it's annoying to have so many copies of this string-splitting logic.
2981 : * However, it's not clear that having one function with a bunch of option
2982 : * flags would be much better.
2983 : *
2984 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
2985 : * Be sure to update that if you have to change this.
2986 : *
2987 : * Inputs:
2988 : * rawstring: the input string; must be overwritable! On return, it's
2989 : * been modified to contain the separated identifiers.
2990 : * separator: the separator punctuation expected between identifiers
2991 : * (typically '.' or ','). Whitespace may also appear around
2992 : * identifiers.
2993 : * Outputs:
2994 : * namelist: filled with a palloc'd list of pointers to identifiers within
2995 : * rawstring. Caller should list_free() this even on error return.
2996 : *
2997 : * Returns true if okay, false if there is a syntax error in the string.
2998 : */
2999 : bool
3000 3934 : SplitGUCList(char *rawstring, char separator,
3001 : List **namelist)
3002 : {
3003 3934 : char *nextp = rawstring;
3004 3934 : bool done = false;
3005 :
3006 3934 : *namelist = NIL;
3007 :
3008 3934 : while (scanner_isspace(*nextp))
3009 0 : nextp++; /* skip leading whitespace */
3010 :
3011 3934 : if (*nextp == '\0')
3012 3860 : return true; /* allow empty string */
3013 :
3014 : /* At the top of the loop, we are at start of a new identifier. */
3015 : do
3016 : {
3017 : char *curname;
3018 : char *endp;
3019 :
3020 100 : if (*nextp == '"')
3021 : {
3022 : /* Quoted name --- collapse quote-quote pairs */
3023 24 : curname = nextp + 1;
3024 : for (;;)
3025 : {
3026 36 : endp = strchr(nextp + 1, '"');
3027 30 : if (endp == NULL)
3028 0 : return false; /* mismatched quotes */
3029 30 : if (endp[1] != '"')
3030 24 : break; /* found end of quoted name */
3031 : /* Collapse adjacent quotes into one quote, and look again */
3032 6 : memmove(endp, endp + 1, strlen(endp));
3033 6 : nextp = endp;
3034 : }
3035 : /* endp now points at the terminating quote */
3036 24 : nextp = endp + 1;
3037 : }
3038 : else
3039 : {
3040 : /* Unquoted name --- extends to separator or whitespace */
3041 76 : curname = nextp;
3042 718 : while (*nextp && *nextp != separator &&
3043 642 : !scanner_isspace(*nextp))
3044 642 : nextp++;
3045 76 : endp = nextp;
3046 76 : if (curname == nextp)
3047 0 : return false; /* empty unquoted name not allowed */
3048 : }
3049 :
3050 100 : while (scanner_isspace(*nextp))
3051 0 : nextp++; /* skip trailing whitespace */
3052 :
3053 100 : if (*nextp == separator)
3054 : {
3055 26 : nextp++;
3056 44 : while (scanner_isspace(*nextp))
3057 18 : nextp++; /* skip leading whitespace for next */
3058 : /* we expect another name, so done remains false */
3059 : }
3060 74 : else if (*nextp == '\0')
3061 74 : done = true;
3062 : else
3063 0 : return false; /* invalid syntax */
3064 :
3065 : /* Now safe to overwrite separator with a null */
3066 100 : *endp = '\0';
3067 :
3068 : /*
3069 : * Finished isolating current name --- add it to list
3070 : */
3071 100 : *namelist = lappend(*namelist, curname);
3072 :
3073 : /* Loop back if we didn't reach end of string */
3074 100 : } while (!done);
3075 :
3076 74 : return true;
3077 : }
3078 :
3079 : /*
3080 : * appendStringInfoText
3081 : *
3082 : * Append a text to str.
3083 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3084 : */
3085 : static void
3086 1926256 : appendStringInfoText(StringInfo str, const text *t)
3087 : {
3088 1926256 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
3089 1926256 : }
3090 :
3091 : /*
3092 : * replace_text
3093 : * replace all occurrences of 'old_sub_str' in 'orig_str'
3094 : * with 'new_sub_str' to form 'new_str'
3095 : *
3096 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3097 : * otherwise returns 'new_str'
3098 : */
3099 : Datum
3100 1418 : replace_text(PG_FUNCTION_ARGS)
3101 : {
3102 1418 : text *src_text = PG_GETARG_TEXT_PP(0);
3103 1418 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
3104 1418 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
3105 : int src_text_len;
3106 : int from_sub_text_len;
3107 : TextPositionState state;
3108 : text *ret_text;
3109 : int chunk_len;
3110 : char *curr_ptr;
3111 : char *start_ptr;
3112 : StringInfoData str;
3113 : bool found;
3114 :
3115 1418 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
3116 1418 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
3117 :
3118 : /* Return unmodified source string if empty source or pattern */
3119 1418 : if (src_text_len < 1 || from_sub_text_len < 1)
3120 : {
3121 0 : PG_RETURN_TEXT_P(src_text);
3122 : }
3123 :
3124 1418 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
3125 :
3126 1418 : found = text_position_next(&state);
3127 :
3128 : /* When the from_sub_text is not found, there is nothing to do. */
3129 1418 : if (!found)
3130 : {
3131 320 : text_position_cleanup(&state);
3132 320 : PG_RETURN_TEXT_P(src_text);
3133 : }
3134 1098 : curr_ptr = text_position_get_match_ptr(&state);
3135 1098 : start_ptr = VARDATA_ANY(src_text);
3136 :
3137 1098 : initStringInfo(&str);
3138 :
3139 : do
3140 : {
3141 5754 : CHECK_FOR_INTERRUPTS();
3142 :
3143 : /* copy the data skipped over by last text_position_next() */
3144 5754 : chunk_len = curr_ptr - start_ptr;
3145 5754 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3146 :
3147 5754 : appendStringInfoText(&str, to_sub_text);
3148 :
3149 5754 : start_ptr = curr_ptr + state.last_match_len;
3150 :
3151 5754 : found = text_position_next(&state);
3152 5754 : if (found)
3153 4656 : curr_ptr = text_position_get_match_ptr(&state);
3154 : }
3155 5754 : while (found);
3156 :
3157 : /* copy trailing data */
3158 1098 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3159 1098 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3160 :
3161 1098 : text_position_cleanup(&state);
3162 :
3163 1098 : ret_text = cstring_to_text_with_len(str.data, str.len);
3164 1098 : pfree(str.data);
3165 :
3166 1098 : PG_RETURN_TEXT_P(ret_text);
3167 : }
3168 :
3169 : /*
3170 : * check_replace_text_has_escape
3171 : *
3172 : * Returns 0 if text contains no backslashes that need processing.
3173 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
3174 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
3175 : */
3176 : static int
3177 18688 : check_replace_text_has_escape(const text *replace_text)
3178 : {
3179 18688 : int result = 0;
3180 18688 : const char *p = VARDATA_ANY(replace_text);
3181 18688 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3182 :
3183 37420 : while (p < p_end)
3184 : {
3185 : /* Find next escape char, if any. */
3186 17628 : p = memchr(p, '\\', p_end - p);
3187 17628 : if (p == NULL)
3188 16806 : break;
3189 822 : p++;
3190 : /* Note: a backslash at the end doesn't require extra processing. */
3191 822 : if (p < p_end)
3192 : {
3193 822 : if (*p >= '1' && *p <= '9')
3194 778 : return 2; /* Found a submatch specifier, so done */
3195 44 : result = 1; /* Found some other sequence, keep looking */
3196 44 : p++;
3197 : }
3198 : }
3199 17910 : return result;
3200 : }
3201 :
3202 : /*
3203 : * appendStringInfoRegexpSubstr
3204 : *
3205 : * Append replace_text to str, substituting regexp back references for
3206 : * \n escapes. start_ptr is the start of the match in the source string,
3207 : * at logical character position data_pos.
3208 : */
3209 : static void
3210 236 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
3211 : regmatch_t *pmatch,
3212 : char *start_ptr, int data_pos)
3213 : {
3214 236 : const char *p = VARDATA_ANY(replace_text);
3215 236 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3216 :
3217 574 : while (p < p_end)
3218 : {
3219 518 : const char *chunk_start = p;
3220 : int so;
3221 : int eo;
3222 :
3223 : /* Find next escape char, if any. */
3224 518 : p = memchr(p, '\\', p_end - p);
3225 518 : if (p == NULL)
3226 174 : p = p_end;
3227 :
3228 : /* Copy the text we just scanned over, if any. */
3229 518 : if (p > chunk_start)
3230 318 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3231 :
3232 : /* Done if at end of string, else advance over escape char. */
3233 518 : if (p >= p_end)
3234 174 : break;
3235 344 : p++;
3236 :
3237 344 : if (p >= p_end)
3238 : {
3239 : /* Escape at very end of input. Treat same as unexpected char */
3240 6 : appendStringInfoChar(str, '\\');
3241 6 : break;
3242 : }
3243 :
3244 338 : if (*p >= '1' && *p <= '9')
3245 278 : {
3246 : /* Use the back reference of regexp. */
3247 278 : int idx = *p - '0';
3248 :
3249 278 : so = pmatch[idx].rm_so;
3250 278 : eo = pmatch[idx].rm_eo;
3251 278 : p++;
3252 : }
3253 60 : else if (*p == '&')
3254 : {
3255 : /* Use the entire matched string. */
3256 18 : so = pmatch[0].rm_so;
3257 18 : eo = pmatch[0].rm_eo;
3258 18 : p++;
3259 : }
3260 42 : else if (*p == '\\')
3261 : {
3262 : /* \\ means transfer one \ to output. */
3263 36 : appendStringInfoChar(str, '\\');
3264 36 : p++;
3265 36 : continue;
3266 : }
3267 : else
3268 : {
3269 : /*
3270 : * If escape char is not followed by any expected char, just treat
3271 : * it as ordinary data to copy. (XXX would it be better to throw
3272 : * an error?)
3273 : */
3274 6 : appendStringInfoChar(str, '\\');
3275 6 : continue;
3276 : }
3277 :
3278 296 : if (so >= 0 && eo >= 0)
3279 : {
3280 : /*
3281 : * Copy the text that is back reference of regexp. Note so and eo
3282 : * are counted in characters not bytes.
3283 : */
3284 : char *chunk_start;
3285 : int chunk_len;
3286 :
3287 : Assert(so >= data_pos);
3288 296 : chunk_start = start_ptr;
3289 296 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3290 296 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3291 296 : appendBinaryStringInfo(str, chunk_start, chunk_len);
3292 : }
3293 : }
3294 236 : }
3295 :
3296 : /*
3297 : * replace_text_regexp
3298 : *
3299 : * replace substring(s) in src_text that match pattern with replace_text.
3300 : * The replace_text can contain backslash markers to substitute
3301 : * (parts of) the matched text.
3302 : *
3303 : * cflags: regexp compile flags.
3304 : * collation: collation to use.
3305 : * search_start: the character (not byte) offset in src_text at which to
3306 : * begin searching.
3307 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
3308 : */
3309 : text *
3310 18688 : replace_text_regexp(text *src_text, text *pattern_text,
3311 : text *replace_text,
3312 : int cflags, Oid collation,
3313 : int search_start, int n)
3314 : {
3315 : text *ret_text;
3316 : regex_t *re;
3317 18688 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3318 18688 : int nmatches = 0;
3319 : StringInfoData buf;
3320 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
3321 18688 : int nmatch = lengthof(pmatch);
3322 : pg_wchar *data;
3323 : size_t data_len;
3324 : int data_pos;
3325 : char *start_ptr;
3326 : int escape_status;
3327 :
3328 18688 : initStringInfo(&buf);
3329 :
3330 : /* Convert data string to wide characters. */
3331 18688 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3332 18688 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3333 :
3334 : /* Check whether replace_text has escapes, especially regexp submatches. */
3335 18688 : escape_status = check_replace_text_has_escape(replace_text);
3336 :
3337 : /* If no regexp submatches, we can use REG_NOSUB. */
3338 18688 : if (escape_status < 2)
3339 : {
3340 17910 : cflags |= REG_NOSUB;
3341 : /* Also tell pg_regexec we only want the whole-match location. */
3342 17910 : nmatch = 1;
3343 : }
3344 :
3345 : /* Prepare the regexp. */
3346 18688 : re = RE_compile_and_cache(pattern_text, cflags, collation);
3347 :
3348 : /* start_ptr points to the data_pos'th character of src_text */
3349 18688 : start_ptr = (char *) VARDATA_ANY(src_text);
3350 18688 : data_pos = 0;
3351 :
3352 25052 : while (search_start <= data_len)
3353 : {
3354 : int regexec_result;
3355 :
3356 25046 : CHECK_FOR_INTERRUPTS();
3357 :
3358 25046 : regexec_result = pg_regexec(re,
3359 : data,
3360 : data_len,
3361 : search_start,
3362 : NULL, /* no details */
3363 : nmatch,
3364 : pmatch,
3365 : 0);
3366 :
3367 25046 : if (regexec_result == REG_NOMATCH)
3368 16640 : break;
3369 :
3370 8406 : if (regexec_result != REG_OKAY)
3371 : {
3372 : char errMsg[100];
3373 :
3374 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3375 0 : ereport(ERROR,
3376 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3377 : errmsg("regular expression failed: %s", errMsg)));
3378 : }
3379 :
3380 : /*
3381 : * Count matches, and decide whether to replace this match.
3382 : */
3383 8406 : nmatches++;
3384 8406 : if (n > 0 && nmatches != n)
3385 : {
3386 : /*
3387 : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
3388 : * we treat the matched text as if it weren't matched, and copy it
3389 : * to the output later.)
3390 : */
3391 60 : search_start = pmatch[0].rm_eo;
3392 60 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3393 0 : search_start++;
3394 60 : continue;
3395 : }
3396 :
3397 : /*
3398 : * Copy the text to the left of the match position. Note we are given
3399 : * character not byte indexes.
3400 : */
3401 8346 : if (pmatch[0].rm_so - data_pos > 0)
3402 : {
3403 : int chunk_len;
3404 :
3405 8188 : chunk_len = charlen_to_bytelen(start_ptr,
3406 8188 : pmatch[0].rm_so - data_pos);
3407 8188 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3408 :
3409 : /*
3410 : * Advance start_ptr over that text, to avoid multiple rescans of
3411 : * it if the replace_text contains multiple back-references.
3412 : */
3413 8188 : start_ptr += chunk_len;
3414 8188 : data_pos = pmatch[0].rm_so;
3415 : }
3416 :
3417 : /*
3418 : * Copy the replace_text, processing escapes if any are present.
3419 : */
3420 8346 : if (escape_status > 0)
3421 236 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3422 : start_ptr, data_pos);
3423 : else
3424 8110 : appendStringInfoText(&buf, replace_text);
3425 :
3426 : /* Advance start_ptr and data_pos over the matched text. */
3427 16692 : start_ptr += charlen_to_bytelen(start_ptr,
3428 8346 : pmatch[0].rm_eo - data_pos);
3429 8346 : data_pos = pmatch[0].rm_eo;
3430 :
3431 : /*
3432 : * If we only want to replace one occurrence, we're done.
3433 : */
3434 8346 : if (n > 0)
3435 2042 : break;
3436 :
3437 : /*
3438 : * Advance search position. Normally we start the next search at the
3439 : * end of the previous match; but if the match was of zero length, we
3440 : * have to advance by one character, or we'd just find the same match
3441 : * again.
3442 : */
3443 6304 : search_start = data_pos;
3444 6304 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3445 12 : search_start++;
3446 : }
3447 :
3448 : /*
3449 : * Copy the text to the right of the last match.
3450 : */
3451 18688 : if (data_pos < data_len)
3452 : {
3453 : int chunk_len;
3454 :
3455 17814 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3456 17814 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3457 : }
3458 :
3459 18688 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
3460 18688 : pfree(buf.data);
3461 18688 : pfree(data);
3462 :
3463 18688 : return ret_text;
3464 : }
3465 :
3466 : /*
3467 : * split_part
3468 : * parse input string based on provided field separator
3469 : * return N'th item (1 based, negative counts from end)
3470 : */
3471 : Datum
3472 150 : split_part(PG_FUNCTION_ARGS)
3473 : {
3474 150 : text *inputstring = PG_GETARG_TEXT_PP(0);
3475 150 : text *fldsep = PG_GETARG_TEXT_PP(1);
3476 150 : int fldnum = PG_GETARG_INT32(2);
3477 : int inputstring_len;
3478 : int fldsep_len;
3479 : TextPositionState state;
3480 : char *start_ptr;
3481 : char *end_ptr;
3482 : text *result_text;
3483 : bool found;
3484 :
3485 : /* field number is 1 based */
3486 150 : if (fldnum == 0)
3487 6 : ereport(ERROR,
3488 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3489 : errmsg("field position must not be zero")));
3490 :
3491 144 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3492 144 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3493 :
3494 : /* return empty string for empty input string */
3495 144 : if (inputstring_len < 1)
3496 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3497 :
3498 : /* handle empty field separator */
3499 132 : if (fldsep_len < 1)
3500 : {
3501 : /* if first or last field, return input string, else empty string */
3502 24 : if (fldnum == 1 || fldnum == -1)
3503 12 : PG_RETURN_TEXT_P(inputstring);
3504 : else
3505 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3506 : }
3507 :
3508 : /* find the first field separator */
3509 108 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
3510 :
3511 108 : found = text_position_next(&state);
3512 :
3513 : /* special case if fldsep not found at all */
3514 108 : if (!found)
3515 : {
3516 24 : text_position_cleanup(&state);
3517 : /* if first or last field, return input string, else empty string */
3518 24 : if (fldnum == 1 || fldnum == -1)
3519 12 : PG_RETURN_TEXT_P(inputstring);
3520 : else
3521 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3522 : }
3523 :
3524 : /*
3525 : * take care of a negative field number (i.e. count from the right) by
3526 : * converting to a positive field number; we need total number of fields
3527 : */
3528 84 : if (fldnum < 0)
3529 : {
3530 : /* we found a fldsep, so there are at least two fields */
3531 42 : int numfields = 2;
3532 :
3533 54 : while (text_position_next(&state))
3534 12 : numfields++;
3535 :
3536 : /* special case of last field does not require an extra pass */
3537 42 : if (fldnum == -1)
3538 : {
3539 24 : start_ptr = text_position_get_match_ptr(&state) + state.last_match_len;
3540 24 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
3541 24 : text_position_cleanup(&state);
3542 24 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
3543 : end_ptr - start_ptr));
3544 : }
3545 :
3546 : /* else, convert fldnum to positive notation */
3547 18 : fldnum += numfields + 1;
3548 :
3549 : /* if nonexistent field, return empty string */
3550 18 : if (fldnum <= 0)
3551 : {
3552 6 : text_position_cleanup(&state);
3553 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
3554 : }
3555 :
3556 : /* reset to pointing at first match, but now with positive fldnum */
3557 12 : text_position_reset(&state);
3558 12 : found = text_position_next(&state);
3559 : Assert(found);
3560 : }
3561 :
3562 : /* identify bounds of first field */
3563 54 : start_ptr = VARDATA_ANY(inputstring);
3564 54 : end_ptr = text_position_get_match_ptr(&state);
3565 :
3566 102 : while (found && --fldnum > 0)
3567 : {
3568 : /* identify bounds of next field */
3569 48 : start_ptr = end_ptr + state.last_match_len;
3570 48 : found = text_position_next(&state);
3571 48 : if (found)
3572 18 : end_ptr = text_position_get_match_ptr(&state);
3573 : }
3574 :
3575 54 : text_position_cleanup(&state);
3576 :
3577 54 : if (fldnum > 0)
3578 : {
3579 : /* N'th field separator not found */
3580 : /* if last field requested, return it, else empty string */
3581 30 : if (fldnum == 1)
3582 : {
3583 24 : int last_len = start_ptr - VARDATA_ANY(inputstring);
3584 :
3585 24 : result_text = cstring_to_text_with_len(start_ptr,
3586 : inputstring_len - last_len);
3587 : }
3588 : else
3589 6 : result_text = cstring_to_text("");
3590 : }
3591 : else
3592 : {
3593 : /* non-last field requested */
3594 24 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
3595 : }
3596 :
3597 54 : PG_RETURN_TEXT_P(result_text);
3598 : }
3599 :
3600 : /*
3601 : * Convenience function to return true when two text params are equal.
3602 : */
3603 : static bool
3604 384 : text_isequal(text *txt1, text *txt2, Oid collid)
3605 : {
3606 384 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
3607 : collid,
3608 : PointerGetDatum(txt1),
3609 : PointerGetDatum(txt2)));
3610 : }
3611 :
3612 : /*
3613 : * text_to_array
3614 : * parse input string and return text array of elements,
3615 : * based on provided field separator
3616 : */
3617 : Datum
3618 170 : text_to_array(PG_FUNCTION_ARGS)
3619 : {
3620 : SplitTextOutputData tstate;
3621 :
3622 : /* For array output, tstate should start as all zeroes */
3623 170 : memset(&tstate, 0, sizeof(tstate));
3624 :
3625 170 : if (!split_text(fcinfo, &tstate))
3626 6 : PG_RETURN_NULL();
3627 :
3628 164 : if (tstate.astate == NULL)
3629 6 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
3630 :
3631 158 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
3632 : CurrentMemoryContext));
3633 : }
3634 :
3635 : /*
3636 : * text_to_array_null
3637 : * parse input string and return text array of elements,
3638 : * based on provided field separator and null string
3639 : *
3640 : * This is a separate entry point only to prevent the regression tests from
3641 : * complaining about different argument sets for the same internal function.
3642 : */
3643 : Datum
3644 60 : text_to_array_null(PG_FUNCTION_ARGS)
3645 : {
3646 60 : return text_to_array(fcinfo);
3647 : }
3648 :
3649 : /*
3650 : * text_to_table
3651 : * parse input string and return table of elements,
3652 : * based on provided field separator
3653 : */
3654 : Datum
3655 84 : text_to_table(PG_FUNCTION_ARGS)
3656 : {
3657 84 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
3658 : SplitTextOutputData tstate;
3659 :
3660 84 : tstate.astate = NULL;
3661 84 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
3662 84 : tstate.tupstore = rsi->setResult;
3663 84 : tstate.tupdesc = rsi->setDesc;
3664 :
3665 84 : (void) split_text(fcinfo, &tstate);
3666 :
3667 84 : return (Datum) 0;
3668 : }
3669 :
3670 : /*
3671 : * text_to_table_null
3672 : * parse input string and return table of elements,
3673 : * based on provided field separator and null string
3674 : *
3675 : * This is a separate entry point only to prevent the regression tests from
3676 : * complaining about different argument sets for the same internal function.
3677 : */
3678 : Datum
3679 24 : text_to_table_null(PG_FUNCTION_ARGS)
3680 : {
3681 24 : return text_to_table(fcinfo);
3682 : }
3683 :
3684 : /*
3685 : * Common code for text_to_array, text_to_array_null, text_to_table
3686 : * and text_to_table_null functions.
3687 : *
3688 : * These are not strict so we have to test for null inputs explicitly.
3689 : * Returns false if result is to be null, else returns true.
3690 : *
3691 : * Note that if the result is valid but empty (zero elements), we return
3692 : * without changing *tstate --- caller must handle that case, too.
3693 : */
3694 : static bool
3695 254 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
3696 : {
3697 : text *inputstring;
3698 : text *fldsep;
3699 : text *null_string;
3700 254 : Oid collation = PG_GET_COLLATION();
3701 : int inputstring_len;
3702 : int fldsep_len;
3703 : char *start_ptr;
3704 : text *result_text;
3705 :
3706 : /* when input string is NULL, then result is NULL too */
3707 254 : if (PG_ARGISNULL(0))
3708 12 : return false;
3709 :
3710 242 : inputstring = PG_GETARG_TEXT_PP(0);
3711 :
3712 : /* fldsep can be NULL */
3713 242 : if (!PG_ARGISNULL(1))
3714 212 : fldsep = PG_GETARG_TEXT_PP(1);
3715 : else
3716 30 : fldsep = NULL;
3717 :
3718 : /* null_string can be NULL or omitted */
3719 242 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
3720 84 : null_string = PG_GETARG_TEXT_PP(2);
3721 : else
3722 158 : null_string = NULL;
3723 :
3724 242 : if (fldsep != NULL)
3725 : {
3726 : /*
3727 : * Normal case with non-null fldsep. Use the text_position machinery
3728 : * to search for occurrences of fldsep.
3729 : */
3730 : TextPositionState state;
3731 :
3732 212 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3733 212 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3734 :
3735 : /* return empty set for empty input string */
3736 212 : if (inputstring_len < 1)
3737 60 : return true;
3738 :
3739 : /* empty field separator: return input string as a one-element set */
3740 200 : if (fldsep_len < 1)
3741 : {
3742 48 : split_text_accum_result(tstate, inputstring,
3743 : null_string, collation);
3744 48 : return true;
3745 : }
3746 :
3747 152 : text_position_setup(inputstring, fldsep, collation, &state);
3748 :
3749 152 : start_ptr = VARDATA_ANY(inputstring);
3750 :
3751 : for (;;)
3752 512 : {
3753 : bool found;
3754 : char *end_ptr;
3755 : int chunk_len;
3756 :
3757 664 : CHECK_FOR_INTERRUPTS();
3758 :
3759 664 : found = text_position_next(&state);
3760 664 : if (!found)
3761 : {
3762 : /* fetch last field */
3763 152 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
3764 152 : end_ptr = NULL; /* not used, but some compilers complain */
3765 : }
3766 : else
3767 : {
3768 : /* fetch non-last field */
3769 512 : end_ptr = text_position_get_match_ptr(&state);
3770 512 : chunk_len = end_ptr - start_ptr;
3771 : }
3772 :
3773 : /* build a temp text datum to pass to split_text_accum_result */
3774 664 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3775 :
3776 : /* stash away this field */
3777 664 : split_text_accum_result(tstate, result_text,
3778 : null_string, collation);
3779 :
3780 664 : pfree(result_text);
3781 :
3782 664 : if (!found)
3783 152 : break;
3784 :
3785 512 : start_ptr = end_ptr + state.last_match_len;
3786 : }
3787 :
3788 152 : text_position_cleanup(&state);
3789 : }
3790 : else
3791 : {
3792 : /*
3793 : * When fldsep is NULL, each character in the input string becomes a
3794 : * separate element in the result set. The separator is effectively
3795 : * the space between characters.
3796 : */
3797 30 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3798 :
3799 30 : start_ptr = VARDATA_ANY(inputstring);
3800 :
3801 252 : while (inputstring_len > 0)
3802 : {
3803 222 : int chunk_len = pg_mblen(start_ptr);
3804 :
3805 222 : CHECK_FOR_INTERRUPTS();
3806 :
3807 : /* build a temp text datum to pass to split_text_accum_result */
3808 222 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3809 :
3810 : /* stash away this field */
3811 222 : split_text_accum_result(tstate, result_text,
3812 : null_string, collation);
3813 :
3814 222 : pfree(result_text);
3815 :
3816 222 : start_ptr += chunk_len;
3817 222 : inputstring_len -= chunk_len;
3818 : }
3819 : }
3820 :
3821 182 : return true;
3822 : }
3823 :
3824 : /*
3825 : * Add text item to result set (table or array).
3826 : *
3827 : * This is also responsible for checking to see if the item matches
3828 : * the null_string, in which case we should emit NULL instead.
3829 : */
3830 : static void
3831 934 : split_text_accum_result(SplitTextOutputData *tstate,
3832 : text *field_value,
3833 : text *null_string,
3834 : Oid collation)
3835 : {
3836 934 : bool is_null = false;
3837 :
3838 934 : if (null_string && text_isequal(field_value, null_string, collation))
3839 72 : is_null = true;
3840 :
3841 934 : if (tstate->tupstore)
3842 : {
3843 : Datum values[1];
3844 : bool nulls[1];
3845 :
3846 228 : values[0] = PointerGetDatum(field_value);
3847 228 : nulls[0] = is_null;
3848 :
3849 228 : tuplestore_putvalues(tstate->tupstore,
3850 : tstate->tupdesc,
3851 : values,
3852 : nulls);
3853 : }
3854 : else
3855 : {
3856 706 : tstate->astate = accumArrayResult(tstate->astate,
3857 : PointerGetDatum(field_value),
3858 : is_null,
3859 : TEXTOID,
3860 : CurrentMemoryContext);
3861 : }
3862 934 : }
3863 :
3864 : /*
3865 : * array_to_text
3866 : * concatenate Cstring representation of input array elements
3867 : * using provided field separator
3868 : */
3869 : Datum
3870 96110 : array_to_text(PG_FUNCTION_ARGS)
3871 : {
3872 96110 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
3873 96110 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3874 :
3875 96110 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
3876 : }
3877 :
3878 : /*
3879 : * array_to_text_null
3880 : * concatenate Cstring representation of input array elements
3881 : * using provided field separator and null string
3882 : *
3883 : * This version is not strict so we have to test for null inputs explicitly.
3884 : */
3885 : Datum
3886 12 : array_to_text_null(PG_FUNCTION_ARGS)
3887 : {
3888 : ArrayType *v;
3889 : char *fldsep;
3890 : char *null_string;
3891 :
3892 : /* returns NULL when first or second parameter is NULL */
3893 12 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
3894 0 : PG_RETURN_NULL();
3895 :
3896 12 : v = PG_GETARG_ARRAYTYPE_P(0);
3897 12 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3898 :
3899 : /* NULL null string is passed through as a null pointer */
3900 12 : if (!PG_ARGISNULL(2))
3901 6 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
3902 : else
3903 6 : null_string = NULL;
3904 :
3905 12 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
3906 : }
3907 :
3908 : /*
3909 : * common code for array_to_text and array_to_text_null functions
3910 : */
3911 : static text *
3912 96140 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
3913 : const char *fldsep, const char *null_string)
3914 : {
3915 : text *result;
3916 : int nitems,
3917 : *dims,
3918 : ndims;
3919 : Oid element_type;
3920 : int typlen;
3921 : bool typbyval;
3922 : char typalign;
3923 : StringInfoData buf;
3924 96140 : bool printed = false;
3925 : char *p;
3926 : bits8 *bitmap;
3927 : int bitmask;
3928 : int i;
3929 : ArrayMetaState *my_extra;
3930 :
3931 96140 : ndims = ARR_NDIM(v);
3932 96140 : dims = ARR_DIMS(v);
3933 96140 : nitems = ArrayGetNItems(ndims, dims);
3934 :
3935 : /* if there are no elements, return an empty string */
3936 96140 : if (nitems == 0)
3937 71096 : return cstring_to_text_with_len("", 0);
3938 :
3939 25044 : element_type = ARR_ELEMTYPE(v);
3940 25044 : initStringInfo(&buf);
3941 :
3942 : /*
3943 : * We arrange to look up info about element type, including its output
3944 : * conversion proc, only once per series of calls, assuming the element
3945 : * type doesn't change underneath us.
3946 : */
3947 25044 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3948 25044 : if (my_extra == NULL)
3949 : {
3950 1474 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
3951 : sizeof(ArrayMetaState));
3952 1474 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3953 1474 : my_extra->element_type = ~element_type;
3954 : }
3955 :
3956 25044 : if (my_extra->element_type != element_type)
3957 : {
3958 : /*
3959 : * Get info about element type, including its output conversion proc
3960 : */
3961 1474 : get_type_io_data(element_type, IOFunc_output,
3962 : &my_extra->typlen, &my_extra->typbyval,
3963 : &my_extra->typalign, &my_extra->typdelim,
3964 : &my_extra->typioparam, &my_extra->typiofunc);
3965 1474 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
3966 1474 : fcinfo->flinfo->fn_mcxt);
3967 1474 : my_extra->element_type = element_type;
3968 : }
3969 25044 : typlen = my_extra->typlen;
3970 25044 : typbyval = my_extra->typbyval;
3971 25044 : typalign = my_extra->typalign;
3972 :
3973 25044 : p = ARR_DATA_PTR(v);
3974 25044 : bitmap = ARR_NULLBITMAP(v);
3975 25044 : bitmask = 1;
3976 :
3977 85100 : for (i = 0; i < nitems; i++)
3978 : {
3979 : Datum itemvalue;
3980 : char *value;
3981 :
3982 : /* Get source element, checking for NULL */
3983 60056 : if (bitmap && (*bitmap & bitmask) == 0)
3984 : {
3985 : /* if null_string is NULL, we just ignore null elements */
3986 18 : if (null_string != NULL)
3987 : {
3988 6 : if (printed)
3989 6 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
3990 : else
3991 0 : appendStringInfoString(&buf, null_string);
3992 6 : printed = true;
3993 : }
3994 : }
3995 : else
3996 : {
3997 60038 : itemvalue = fetch_att(p, typbyval, typlen);
3998 :
3999 60038 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
4000 :
4001 60038 : if (printed)
4002 34994 : appendStringInfo(&buf, "%s%s", fldsep, value);
4003 : else
4004 25044 : appendStringInfoString(&buf, value);
4005 60038 : printed = true;
4006 :
4007 60038 : p = att_addlength_pointer(p, typlen, p);
4008 60038 : p = (char *) att_align_nominal(p, typalign);
4009 : }
4010 :
4011 : /* advance bitmap pointer if any */
4012 60056 : if (bitmap)
4013 : {
4014 108 : bitmask <<= 1;
4015 108 : if (bitmask == 0x100)
4016 : {
4017 0 : bitmap++;
4018 0 : bitmask = 1;
4019 : }
4020 : }
4021 : }
4022 :
4023 25044 : result = cstring_to_text_with_len(buf.data, buf.len);
4024 25044 : pfree(buf.data);
4025 :
4026 25044 : return result;
4027 : }
4028 :
4029 : /*
4030 : * Workhorse for to_bin, to_oct, and to_hex. Note that base must be > 1 and <=
4031 : * 16.
4032 : */
4033 : static inline text *
4034 38750 : convert_to_base(uint64 value, int base)
4035 : {
4036 38750 : const char *digits = "0123456789abcdef";
4037 :
4038 : /* We size the buffer for to_bin's longest possible return value. */
4039 : char buf[sizeof(uint64) * BITS_PER_BYTE];
4040 38750 : char *const end = buf + sizeof(buf);
4041 38750 : char *ptr = end;
4042 :
4043 : Assert(base > 1);
4044 : Assert(base <= 16);
4045 :
4046 : do
4047 : {
4048 75974 : *--ptr = digits[value % base];
4049 75974 : value /= base;
4050 75974 : } while (ptr > buf && value);
4051 :
4052 38750 : return cstring_to_text_with_len(ptr, end - ptr);
4053 : }
4054 :
4055 : /*
4056 : * Convert an integer to a string containing a base-2 (binary) representation
4057 : * of the number.
4058 : */
4059 : Datum
4060 12 : to_bin32(PG_FUNCTION_ARGS)
4061 : {
4062 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4063 :
4064 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4065 : }
4066 : Datum
4067 12 : to_bin64(PG_FUNCTION_ARGS)
4068 : {
4069 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4070 :
4071 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4072 : }
4073 :
4074 : /*
4075 : * Convert an integer to a string containing a base-8 (oct) representation of
4076 : * the number.
4077 : */
4078 : Datum
4079 12 : to_oct32(PG_FUNCTION_ARGS)
4080 : {
4081 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4082 :
4083 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4084 : }
4085 : Datum
4086 12 : to_oct64(PG_FUNCTION_ARGS)
4087 : {
4088 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4089 :
4090 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4091 : }
4092 :
4093 : /*
4094 : * Convert an integer to a string containing a base-16 (hex) representation of
4095 : * the number.
4096 : */
4097 : Datum
4098 38690 : to_hex32(PG_FUNCTION_ARGS)
4099 : {
4100 38690 : uint64 value = (uint32) PG_GETARG_INT32(0);
4101 :
4102 38690 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4103 : }
4104 : Datum
4105 12 : to_hex64(PG_FUNCTION_ARGS)
4106 : {
4107 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4108 :
4109 12 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4110 : }
4111 :
4112 : /*
4113 : * Return the size of a datum, possibly compressed
4114 : *
4115 : * Works on any data type
4116 : */
4117 : Datum
4118 122 : pg_column_size(PG_FUNCTION_ARGS)
4119 : {
4120 122 : Datum value = PG_GETARG_DATUM(0);
4121 : int32 result;
4122 : int typlen;
4123 :
4124 : /* On first call, get the input type's typlen, and save at *fn_extra */
4125 122 : if (fcinfo->flinfo->fn_extra == NULL)
4126 : {
4127 : /* Lookup the datatype of the supplied argument */
4128 122 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4129 :
4130 122 : typlen = get_typlen(argtypeid);
4131 122 : if (typlen == 0) /* should not happen */
4132 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4133 :
4134 122 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4135 : sizeof(int));
4136 122 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4137 : }
4138 : else
4139 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4140 :
4141 122 : if (typlen == -1)
4142 : {
4143 : /* varlena type, possibly toasted */
4144 122 : result = toast_datum_size(value);
4145 : }
4146 0 : else if (typlen == -2)
4147 : {
4148 : /* cstring */
4149 0 : result = strlen(DatumGetCString(value)) + 1;
4150 : }
4151 : else
4152 : {
4153 : /* ordinary fixed-width type */
4154 0 : result = typlen;
4155 : }
4156 :
4157 122 : PG_RETURN_INT32(result);
4158 : }
4159 :
4160 : /*
4161 : * Return the compression method stored in the compressed attribute. Return
4162 : * NULL for non varlena type or uncompressed data.
4163 : */
4164 : Datum
4165 138 : pg_column_compression(PG_FUNCTION_ARGS)
4166 : {
4167 : int typlen;
4168 : char *result;
4169 : ToastCompressionId cmid;
4170 :
4171 : /* On first call, get the input type's typlen, and save at *fn_extra */
4172 138 : if (fcinfo->flinfo->fn_extra == NULL)
4173 : {
4174 : /* Lookup the datatype of the supplied argument */
4175 114 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4176 :
4177 114 : typlen = get_typlen(argtypeid);
4178 114 : if (typlen == 0) /* should not happen */
4179 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4180 :
4181 114 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4182 : sizeof(int));
4183 114 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4184 : }
4185 : else
4186 24 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4187 :
4188 138 : if (typlen != -1)
4189 0 : PG_RETURN_NULL();
4190 :
4191 : /* get the compression method id stored in the compressed varlena */
4192 138 : cmid = toast_get_compression_id((struct varlena *)
4193 138 : DatumGetPointer(PG_GETARG_DATUM(0)));
4194 138 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
4195 6 : PG_RETURN_NULL();
4196 :
4197 : /* convert compression method id to compression method name */
4198 132 : switch (cmid)
4199 : {
4200 66 : case TOAST_PGLZ_COMPRESSION_ID:
4201 66 : result = "pglz";
4202 66 : break;
4203 66 : case TOAST_LZ4_COMPRESSION_ID:
4204 66 : result = "lz4";
4205 66 : break;
4206 0 : default:
4207 0 : elog(ERROR, "invalid compression method id %d", cmid);
4208 : }
4209 :
4210 132 : PG_RETURN_TEXT_P(cstring_to_text(result));
4211 : }
4212 :
4213 : /*
4214 : * Return the chunk_id of the on-disk TOASTed value. Return NULL if the value
4215 : * is un-TOASTed or not on-disk.
4216 : */
4217 : Datum
4218 12 : pg_column_toast_chunk_id(PG_FUNCTION_ARGS)
4219 : {
4220 : int typlen;
4221 : struct varlena *attr;
4222 : struct varatt_external toast_pointer;
4223 :
4224 : /* On first call, get the input type's typlen, and save at *fn_extra */
4225 12 : if (fcinfo->flinfo->fn_extra == NULL)
4226 : {
4227 : /* Lookup the datatype of the supplied argument */
4228 12 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4229 :
4230 12 : typlen = get_typlen(argtypeid);
4231 12 : if (typlen == 0) /* should not happen */
4232 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4233 :
4234 12 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4235 : sizeof(int));
4236 12 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4237 : }
4238 : else
4239 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4240 :
4241 12 : if (typlen != -1)
4242 0 : PG_RETURN_NULL();
4243 :
4244 12 : attr = (struct varlena *) DatumGetPointer(PG_GETARG_DATUM(0));
4245 :
4246 12 : if (!VARATT_IS_EXTERNAL_ONDISK(attr))
4247 6 : PG_RETURN_NULL();
4248 :
4249 6 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
4250 :
4251 6 : PG_RETURN_OID(toast_pointer.va_valueid);
4252 : }
4253 :
4254 : /*
4255 : * string_agg - Concatenates values and returns string.
4256 : *
4257 : * Syntax: string_agg(value text, delimiter text) RETURNS text
4258 : *
4259 : * Note: Any NULL values are ignored. The first-call delimiter isn't
4260 : * actually used at all, and on subsequent calls the delimiter precedes
4261 : * the associated value.
4262 : */
4263 :
4264 : /* subroutine to initialize state */
4265 : static StringInfo
4266 2370 : makeStringAggState(FunctionCallInfo fcinfo)
4267 : {
4268 : StringInfo state;
4269 : MemoryContext aggcontext;
4270 : MemoryContext oldcontext;
4271 :
4272 2370 : if (!AggCheckCallContext(fcinfo, &aggcontext))
4273 : {
4274 : /* cannot be called directly because of internal-type argument */
4275 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
4276 : }
4277 :
4278 : /*
4279 : * Create state in aggregate context. It'll stay there across subsequent
4280 : * calls.
4281 : */
4282 2370 : oldcontext = MemoryContextSwitchTo(aggcontext);
4283 2370 : state = makeStringInfo();
4284 2370 : MemoryContextSwitchTo(oldcontext);
4285 :
4286 2370 : return state;
4287 : }
4288 :
4289 : Datum
4290 971244 : string_agg_transfn(PG_FUNCTION_ARGS)
4291 : {
4292 : StringInfo state;
4293 :
4294 971244 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4295 :
4296 : /* Append the value unless null, preceding it with the delimiter. */
4297 971244 : if (!PG_ARGISNULL(1))
4298 : {
4299 956196 : text *value = PG_GETARG_TEXT_PP(1);
4300 956196 : bool isfirst = false;
4301 :
4302 : /*
4303 : * You might think we can just throw away the first delimiter, however
4304 : * we must keep it as we may be a parallel worker doing partial
4305 : * aggregation building a state to send to the main process. We need
4306 : * to keep the delimiter of every aggregation so that the combine
4307 : * function can properly join up the strings of two separately
4308 : * partially aggregated results. The first delimiter is only stripped
4309 : * off in the final function. To know how much to strip off the front
4310 : * of the string, we store the length of the first delimiter in the
4311 : * StringInfo's cursor field, which we don't otherwise need here.
4312 : */
4313 956196 : if (state == NULL)
4314 : {
4315 2050 : state = makeStringAggState(fcinfo);
4316 2050 : isfirst = true;
4317 : }
4318 :
4319 956196 : if (!PG_ARGISNULL(2))
4320 : {
4321 956196 : text *delim = PG_GETARG_TEXT_PP(2);
4322 :
4323 956196 : appendStringInfoText(state, delim);
4324 956196 : if (isfirst)
4325 2050 : state->cursor = VARSIZE_ANY_EXHDR(delim);
4326 : }
4327 :
4328 956196 : appendStringInfoText(state, value);
4329 : }
4330 :
4331 : /*
4332 : * The transition type for string_agg() is declared to be "internal",
4333 : * which is a pass-by-value type the same size as a pointer.
4334 : */
4335 971244 : if (state)
4336 971156 : PG_RETURN_POINTER(state);
4337 88 : PG_RETURN_NULL();
4338 : }
4339 :
4340 : /*
4341 : * string_agg_combine
4342 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
4343 : */
4344 : Datum
4345 200 : string_agg_combine(PG_FUNCTION_ARGS)
4346 : {
4347 : StringInfo state1;
4348 : StringInfo state2;
4349 : MemoryContext agg_context;
4350 :
4351 200 : if (!AggCheckCallContext(fcinfo, &agg_context))
4352 0 : elog(ERROR, "aggregate function called in non-aggregate context");
4353 :
4354 200 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4355 200 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
4356 :
4357 200 : if (state2 == NULL)
4358 : {
4359 : /*
4360 : * NULL state2 is easy, just return state1, which we know is already
4361 : * in the agg_context
4362 : */
4363 0 : if (state1 == NULL)
4364 0 : PG_RETURN_NULL();
4365 0 : PG_RETURN_POINTER(state1);
4366 : }
4367 :
4368 200 : if (state1 == NULL)
4369 : {
4370 : /* We must copy state2's data into the agg_context */
4371 : MemoryContext old_context;
4372 :
4373 120 : old_context = MemoryContextSwitchTo(agg_context);
4374 120 : state1 = makeStringAggState(fcinfo);
4375 120 : appendBinaryStringInfo(state1, state2->data, state2->len);
4376 120 : state1->cursor = state2->cursor;
4377 120 : MemoryContextSwitchTo(old_context);
4378 : }
4379 80 : else if (state2->len > 0)
4380 : {
4381 : /* Combine ... state1->cursor does not change in this case */
4382 80 : appendBinaryStringInfo(state1, state2->data, state2->len);
4383 : }
4384 :
4385 200 : PG_RETURN_POINTER(state1);
4386 : }
4387 :
4388 : /*
4389 : * string_agg_serialize
4390 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
4391 : *
4392 : * This is strict, so we need not handle NULL input
4393 : */
4394 : Datum
4395 200 : string_agg_serialize(PG_FUNCTION_ARGS)
4396 : {
4397 : StringInfo state;
4398 : StringInfoData buf;
4399 : bytea *result;
4400 :
4401 : /* cannot be called directly because of internal-type argument */
4402 : Assert(AggCheckCallContext(fcinfo, NULL));
4403 :
4404 200 : state = (StringInfo) PG_GETARG_POINTER(0);
4405 :
4406 200 : pq_begintypsend(&buf);
4407 :
4408 : /* cursor */
4409 200 : pq_sendint(&buf, state->cursor, 4);
4410 :
4411 : /* data */
4412 200 : pq_sendbytes(&buf, state->data, state->len);
4413 :
4414 200 : result = pq_endtypsend(&buf);
4415 :
4416 200 : PG_RETURN_BYTEA_P(result);
4417 : }
4418 :
4419 : /*
4420 : * string_agg_deserialize
4421 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
4422 : *
4423 : * This is strict, so we need not handle NULL input
4424 : */
4425 : Datum
4426 200 : string_agg_deserialize(PG_FUNCTION_ARGS)
4427 : {
4428 : bytea *sstate;
4429 : StringInfo result;
4430 : StringInfoData buf;
4431 : char *data;
4432 : int datalen;
4433 :
4434 : /* cannot be called directly because of internal-type argument */
4435 : Assert(AggCheckCallContext(fcinfo, NULL));
4436 :
4437 200 : sstate = PG_GETARG_BYTEA_PP(0);
4438 :
4439 : /*
4440 : * Initialize a StringInfo so that we can "receive" it using the standard
4441 : * recv-function infrastructure.
4442 : */
4443 200 : initReadOnlyStringInfo(&buf, VARDATA_ANY(sstate),
4444 200 : VARSIZE_ANY_EXHDR(sstate));
4445 :
4446 200 : result = makeStringAggState(fcinfo);
4447 :
4448 : /* cursor */
4449 200 : result->cursor = pq_getmsgint(&buf, 4);
4450 :
4451 : /* data */
4452 200 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
4453 200 : data = (char *) pq_getmsgbytes(&buf, datalen);
4454 200 : appendBinaryStringInfo(result, data, datalen);
4455 :
4456 200 : pq_getmsgend(&buf);
4457 :
4458 200 : PG_RETURN_POINTER(result);
4459 : }
4460 :
4461 : Datum
4462 2094 : string_agg_finalfn(PG_FUNCTION_ARGS)
4463 : {
4464 : StringInfo state;
4465 :
4466 : /* cannot be called directly because of internal-type argument */
4467 : Assert(AggCheckCallContext(fcinfo, NULL));
4468 :
4469 2094 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4470 :
4471 2094 : if (state != NULL)
4472 : {
4473 : /* As per comment in transfn, strip data before the cursor position */
4474 2010 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
4475 : state->len - state->cursor));
4476 : }
4477 : else
4478 84 : PG_RETURN_NULL();
4479 : }
4480 :
4481 : /*
4482 : * Prepare cache with fmgr info for the output functions of the datatypes of
4483 : * the arguments of a concat-like function, beginning with argument "argidx".
4484 : * (Arguments before that will have corresponding slots in the resulting
4485 : * FmgrInfo array, but we don't fill those slots.)
4486 : */
4487 : static FmgrInfo *
4488 106 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
4489 : {
4490 : FmgrInfo *foutcache;
4491 : int i;
4492 :
4493 : /* We keep the info in fn_mcxt so it survives across calls */
4494 106 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4495 106 : PG_NARGS() * sizeof(FmgrInfo));
4496 :
4497 400 : for (i = argidx; i < PG_NARGS(); i++)
4498 : {
4499 : Oid valtype;
4500 : Oid typOutput;
4501 : bool typIsVarlena;
4502 :
4503 294 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4504 294 : if (!OidIsValid(valtype))
4505 0 : elog(ERROR, "could not determine data type of concat() input");
4506 :
4507 294 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4508 294 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
4509 : }
4510 :
4511 106 : fcinfo->flinfo->fn_extra = foutcache;
4512 :
4513 106 : return foutcache;
4514 : }
4515 :
4516 : /*
4517 : * Implementation of both concat() and concat_ws().
4518 : *
4519 : * sepstr is the separator string to place between values.
4520 : * argidx identifies the first argument to concatenate (counting from zero);
4521 : * note that this must be constant across any one series of calls.
4522 : *
4523 : * Returns NULL if result should be NULL, else text value.
4524 : */
4525 : static text *
4526 264 : concat_internal(const char *sepstr, int argidx,
4527 : FunctionCallInfo fcinfo)
4528 : {
4529 : text *result;
4530 : StringInfoData str;
4531 : FmgrInfo *foutcache;
4532 264 : bool first_arg = true;
4533 : int i;
4534 :
4535 : /*
4536 : * concat(VARIADIC some-array) is essentially equivalent to
4537 : * array_to_text(), ie concat the array elements with the given separator.
4538 : * So we just pass the case off to that code.
4539 : */
4540 264 : if (get_fn_expr_variadic(fcinfo->flinfo))
4541 : {
4542 : ArrayType *arr;
4543 :
4544 : /* Should have just the one argument */
4545 : Assert(argidx == PG_NARGS() - 1);
4546 :
4547 : /* concat(VARIADIC NULL) is defined as NULL */
4548 30 : if (PG_ARGISNULL(argidx))
4549 12 : return NULL;
4550 :
4551 : /*
4552 : * Non-null argument had better be an array. We assume that any call
4553 : * context that could let get_fn_expr_variadic return true will have
4554 : * checked that a VARIADIC-labeled parameter actually is an array. So
4555 : * it should be okay to just Assert that it's an array rather than
4556 : * doing a full-fledged error check.
4557 : */
4558 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
4559 :
4560 : /* OK, safe to fetch the array value */
4561 18 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
4562 :
4563 : /*
4564 : * And serialize the array. We tell array_to_text to ignore null
4565 : * elements, which matches the behavior of the loop below.
4566 : */
4567 18 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4568 : }
4569 :
4570 : /* Normal case without explicit VARIADIC marker */
4571 234 : initStringInfo(&str);
4572 :
4573 : /* Get output function info, building it if first time through */
4574 234 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
4575 234 : if (foutcache == NULL)
4576 106 : foutcache = build_concat_foutcache(fcinfo, argidx);
4577 :
4578 822 : for (i = argidx; i < PG_NARGS(); i++)
4579 : {
4580 588 : if (!PG_ARGISNULL(i))
4581 : {
4582 510 : Datum value = PG_GETARG_DATUM(i);
4583 :
4584 : /* add separator if appropriate */
4585 510 : if (first_arg)
4586 228 : first_arg = false;
4587 : else
4588 282 : appendStringInfoString(&str, sepstr);
4589 :
4590 : /* call the appropriate type output function, append the result */
4591 510 : appendStringInfoString(&str,
4592 510 : OutputFunctionCall(&foutcache[i], value));
4593 : }
4594 : }
4595 :
4596 234 : result = cstring_to_text_with_len(str.data, str.len);
4597 234 : pfree(str.data);
4598 :
4599 234 : return result;
4600 : }
4601 :
4602 : /*
4603 : * Concatenate all arguments. NULL arguments are ignored.
4604 : */
4605 : Datum
4606 186 : text_concat(PG_FUNCTION_ARGS)
4607 : {
4608 : text *result;
4609 :
4610 186 : result = concat_internal("", 0, fcinfo);
4611 186 : if (result == NULL)
4612 6 : PG_RETURN_NULL();
4613 180 : PG_RETURN_TEXT_P(result);
4614 : }
4615 :
4616 : /*
4617 : * Concatenate all but first argument value with separators. The first
4618 : * parameter is used as the separator. NULL arguments are ignored.
4619 : */
4620 : Datum
4621 84 : text_concat_ws(PG_FUNCTION_ARGS)
4622 : {
4623 : char *sep;
4624 : text *result;
4625 :
4626 : /* return NULL when separator is NULL */
4627 84 : if (PG_ARGISNULL(0))
4628 6 : PG_RETURN_NULL();
4629 78 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
4630 :
4631 78 : result = concat_internal(sep, 1, fcinfo);
4632 78 : if (result == NULL)
4633 6 : PG_RETURN_NULL();
4634 72 : PG_RETURN_TEXT_P(result);
4635 : }
4636 :
4637 : /*
4638 : * Return first n characters in the string. When n is negative,
4639 : * return all but last |n| characters.
4640 : */
4641 : Datum
4642 2148 : text_left(PG_FUNCTION_ARGS)
4643 : {
4644 2148 : int n = PG_GETARG_INT32(1);
4645 :
4646 2148 : if (n < 0)
4647 : {
4648 30 : text *str = PG_GETARG_TEXT_PP(0);
4649 30 : const char *p = VARDATA_ANY(str);
4650 30 : int len = VARSIZE_ANY_EXHDR(str);
4651 : int rlen;
4652 :
4653 30 : n = pg_mbstrlen_with_len(p, len) + n;
4654 30 : rlen = pg_mbcharcliplen(p, len, n);
4655 30 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
4656 : }
4657 : else
4658 2118 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
4659 : }
4660 :
4661 : /*
4662 : * Return last n characters in the string. When n is negative,
4663 : * return all but first |n| characters.
4664 : */
4665 : Datum
4666 66 : text_right(PG_FUNCTION_ARGS)
4667 : {
4668 66 : text *str = PG_GETARG_TEXT_PP(0);
4669 66 : const char *p = VARDATA_ANY(str);
4670 66 : int len = VARSIZE_ANY_EXHDR(str);
4671 66 : int n = PG_GETARG_INT32(1);
4672 : int off;
4673 :
4674 66 : if (n < 0)
4675 30 : n = -n;
4676 : else
4677 36 : n = pg_mbstrlen_with_len(p, len) - n;
4678 66 : off = pg_mbcharcliplen(p, len, n);
4679 :
4680 66 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4681 : }
4682 :
4683 : /*
4684 : * Return reversed string
4685 : */
4686 : Datum
4687 6 : text_reverse(PG_FUNCTION_ARGS)
4688 : {
4689 6 : text *str = PG_GETARG_TEXT_PP(0);
4690 6 : const char *p = VARDATA_ANY(str);
4691 6 : int len = VARSIZE_ANY_EXHDR(str);
4692 6 : const char *endp = p + len;
4693 : text *result;
4694 : char *dst;
4695 :
4696 6 : result = palloc(len + VARHDRSZ);
4697 6 : dst = (char *) VARDATA(result) + len;
4698 6 : SET_VARSIZE(result, len + VARHDRSZ);
4699 :
4700 6 : if (pg_database_encoding_max_length() > 1)
4701 : {
4702 : /* multibyte version */
4703 36 : while (p < endp)
4704 : {
4705 : int sz;
4706 :
4707 30 : sz = pg_mblen(p);
4708 30 : dst -= sz;
4709 30 : memcpy(dst, p, sz);
4710 30 : p += sz;
4711 : }
4712 : }
4713 : else
4714 : {
4715 : /* single byte version */
4716 0 : while (p < endp)
4717 0 : *(--dst) = *p++;
4718 : }
4719 :
4720 6 : PG_RETURN_TEXT_P(result);
4721 : }
4722 :
4723 :
4724 : /*
4725 : * Support macros for text_format()
4726 : */
4727 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4728 :
4729 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4730 : do { \
4731 : if (++(ptr) >= (end_ptr)) \
4732 : ereport(ERROR, \
4733 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4734 : errmsg("unterminated format() type specifier"), \
4735 : errhint("For a single \"%%\" use \"%%%%\"."))); \
4736 : } while (0)
4737 :
4738 : /*
4739 : * Returns a formatted string
4740 : */
4741 : Datum
4742 33178 : text_format(PG_FUNCTION_ARGS)
4743 : {
4744 : text *fmt;
4745 : StringInfoData str;
4746 : const char *cp;
4747 : const char *start_ptr;
4748 : const char *end_ptr;
4749 : text *result;
4750 : int arg;
4751 : bool funcvariadic;
4752 : int nargs;
4753 33178 : Datum *elements = NULL;
4754 33178 : bool *nulls = NULL;
4755 33178 : Oid element_type = InvalidOid;
4756 33178 : Oid prev_type = InvalidOid;
4757 33178 : Oid prev_width_type = InvalidOid;
4758 : FmgrInfo typoutputfinfo;
4759 : FmgrInfo typoutputinfo_width;
4760 :
4761 : /* When format string is null, immediately return null */
4762 33178 : if (PG_ARGISNULL(0))
4763 6 : PG_RETURN_NULL();
4764 :
4765 : /* If argument is marked VARIADIC, expand array into elements */
4766 33172 : if (get_fn_expr_variadic(fcinfo->flinfo))
4767 : {
4768 : ArrayType *arr;
4769 : int16 elmlen;
4770 : bool elmbyval;
4771 : char elmalign;
4772 : int nitems;
4773 :
4774 : /* Should have just the one argument */
4775 : Assert(PG_NARGS() == 2);
4776 :
4777 : /* If argument is NULL, we treat it as zero-length array */
4778 48 : if (PG_ARGISNULL(1))
4779 6 : nitems = 0;
4780 : else
4781 : {
4782 : /*
4783 : * Non-null argument had better be an array. We assume that any
4784 : * call context that could let get_fn_expr_variadic return true
4785 : * will have checked that a VARIADIC-labeled parameter actually is
4786 : * an array. So it should be okay to just Assert that it's an
4787 : * array rather than doing a full-fledged error check.
4788 : */
4789 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
4790 :
4791 : /* OK, safe to fetch the array value */
4792 42 : arr = PG_GETARG_ARRAYTYPE_P(1);
4793 :
4794 : /* Get info about array element type */
4795 42 : element_type = ARR_ELEMTYPE(arr);
4796 42 : get_typlenbyvalalign(element_type,
4797 : &elmlen, &elmbyval, &elmalign);
4798 :
4799 : /* Extract all array elements */
4800 42 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
4801 : &elements, &nulls, &nitems);
4802 : }
4803 :
4804 48 : nargs = nitems + 1;
4805 48 : funcvariadic = true;
4806 : }
4807 : else
4808 : {
4809 : /* Non-variadic case, we'll process the arguments individually */
4810 33124 : nargs = PG_NARGS();
4811 33124 : funcvariadic = false;
4812 : }
4813 :
4814 : /* Setup for main loop. */
4815 33172 : fmt = PG_GETARG_TEXT_PP(0);
4816 33172 : start_ptr = VARDATA_ANY(fmt);
4817 33172 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
4818 33172 : initStringInfo(&str);
4819 33172 : arg = 1; /* next argument position to print */
4820 :
4821 : /* Scan format string, looking for conversion specifiers. */
4822 1012540 : for (cp = start_ptr; cp < end_ptr; cp++)
4823 : {
4824 : int argpos;
4825 : int widthpos;
4826 : int flags;
4827 : int width;
4828 : Datum value;
4829 : bool isNull;
4830 : Oid typid;
4831 :
4832 : /*
4833 : * If it's not the start of a conversion specifier, just copy it to
4834 : * the output buffer.
4835 : */
4836 979428 : if (*cp != '%')
4837 : {
4838 913598 : appendStringInfoCharMacro(&str, *cp);
4839 913616 : continue;
4840 : }
4841 :
4842 65830 : ADVANCE_PARSE_POINTER(cp, end_ptr);
4843 :
4844 : /* Easy case: %% outputs a single % */
4845 65830 : if (*cp == '%')
4846 : {
4847 18 : appendStringInfoCharMacro(&str, *cp);
4848 18 : continue;
4849 : }
4850 :
4851 : /* Parse the optional portions of the format specifier */
4852 65812 : cp = text_format_parse_format(cp, end_ptr,
4853 : &argpos, &widthpos,
4854 : &flags, &width);
4855 :
4856 : /*
4857 : * Next we should see the main conversion specifier. Whether or not
4858 : * an argument position was present, it's known that at least one
4859 : * character remains in the string at this point. Experience suggests
4860 : * that it's worth checking that that character is one of the expected
4861 : * ones before we try to fetch arguments, so as to produce the least
4862 : * confusing response to a mis-formatted specifier.
4863 : */
4864 65788 : if (strchr("sIL", *cp) == NULL)
4865 6 : ereport(ERROR,
4866 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4867 : errmsg("unrecognized format() type specifier \"%.*s\"",
4868 : pg_mblen(cp), cp),
4869 : errhint("For a single \"%%\" use \"%%%%\".")));
4870 :
4871 : /* If indirect width was specified, get its value */
4872 65782 : if (widthpos >= 0)
4873 : {
4874 : /* Collect the specified or next argument position */
4875 42 : if (widthpos > 0)
4876 36 : arg = widthpos;
4877 42 : if (arg >= nargs)
4878 0 : ereport(ERROR,
4879 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4880 : errmsg("too few arguments for format()")));
4881 :
4882 : /* Get the value and type of the selected argument */
4883 42 : if (!funcvariadic)
4884 : {
4885 42 : value = PG_GETARG_DATUM(arg);
4886 42 : isNull = PG_ARGISNULL(arg);
4887 42 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4888 : }
4889 : else
4890 : {
4891 0 : value = elements[arg - 1];
4892 0 : isNull = nulls[arg - 1];
4893 0 : typid = element_type;
4894 : }
4895 42 : if (!OidIsValid(typid))
4896 0 : elog(ERROR, "could not determine data type of format() input");
4897 :
4898 42 : arg++;
4899 :
4900 : /* We can treat NULL width the same as zero */
4901 42 : if (isNull)
4902 6 : width = 0;
4903 36 : else if (typid == INT4OID)
4904 36 : width = DatumGetInt32(value);
4905 0 : else if (typid == INT2OID)
4906 0 : width = DatumGetInt16(value);
4907 : else
4908 : {
4909 : /* For less-usual datatypes, convert to text then to int */
4910 : char *str;
4911 :
4912 0 : if (typid != prev_width_type)
4913 : {
4914 : Oid typoutputfunc;
4915 : bool typIsVarlena;
4916 :
4917 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4918 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
4919 0 : prev_width_type = typid;
4920 : }
4921 :
4922 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
4923 :
4924 : /* pg_strtoint32 will complain about bad data or overflow */
4925 0 : width = pg_strtoint32(str);
4926 :
4927 0 : pfree(str);
4928 : }
4929 : }
4930 :
4931 : /* Collect the specified or next argument position */
4932 65782 : if (argpos > 0)
4933 132 : arg = argpos;
4934 65782 : if (arg >= nargs)
4935 24 : ereport(ERROR,
4936 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4937 : errmsg("too few arguments for format()")));
4938 :
4939 : /* Get the value and type of the selected argument */
4940 65758 : if (!funcvariadic)
4941 : {
4942 64486 : value = PG_GETARG_DATUM(arg);
4943 64486 : isNull = PG_ARGISNULL(arg);
4944 64486 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4945 : }
4946 : else
4947 : {
4948 1272 : value = elements[arg - 1];
4949 1272 : isNull = nulls[arg - 1];
4950 1272 : typid = element_type;
4951 : }
4952 65758 : if (!OidIsValid(typid))
4953 0 : elog(ERROR, "could not determine data type of format() input");
4954 :
4955 65758 : arg++;
4956 :
4957 : /*
4958 : * Get the appropriate typOutput function, reusing previous one if
4959 : * same type as previous argument. That's particularly useful in the
4960 : * variadic-array case, but often saves work even for ordinary calls.
4961 : */
4962 65758 : if (typid != prev_type)
4963 : {
4964 : Oid typoutputfunc;
4965 : bool typIsVarlena;
4966 :
4967 34240 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4968 34240 : fmgr_info(typoutputfunc, &typoutputfinfo);
4969 34240 : prev_type = typid;
4970 : }
4971 :
4972 : /*
4973 : * And now we can format the value.
4974 : */
4975 65758 : switch (*cp)
4976 : {
4977 65758 : case 's':
4978 : case 'I':
4979 : case 'L':
4980 65758 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
4981 : value, isNull,
4982 : flags, width);
4983 65752 : break;
4984 0 : default:
4985 : /* should not get here, because of previous check */
4986 0 : ereport(ERROR,
4987 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4988 : errmsg("unrecognized format() type specifier \"%.*s\"",
4989 : pg_mblen(cp), cp),
4990 : errhint("For a single \"%%\" use \"%%%%\".")));
4991 : break;
4992 : }
4993 : }
4994 :
4995 : /* Don't need deconstruct_array results anymore. */
4996 33112 : if (elements != NULL)
4997 42 : pfree(elements);
4998 33112 : if (nulls != NULL)
4999 42 : pfree(nulls);
5000 :
5001 : /* Generate results. */
5002 33112 : result = cstring_to_text_with_len(str.data, str.len);
5003 33112 : pfree(str.data);
5004 :
5005 33112 : PG_RETURN_TEXT_P(result);
5006 : }
5007 :
5008 : /*
5009 : * Parse contiguous digits as a decimal number.
5010 : *
5011 : * Returns true if some digits could be parsed.
5012 : * The value is returned into *value, and *ptr is advanced to the next
5013 : * character to be parsed.
5014 : *
5015 : * Note parsing invariant: at least one character is known available before
5016 : * string end (end_ptr) at entry, and this is still true at exit.
5017 : */
5018 : static bool
5019 131588 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5020 : {
5021 131588 : bool found = false;
5022 131588 : const char *cp = *ptr;
5023 131588 : int val = 0;
5024 :
5025 131900 : while (*cp >= '0' && *cp <= '9')
5026 : {
5027 318 : int8 digit = (*cp - '0');
5028 :
5029 318 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5030 318 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5031 0 : ereport(ERROR,
5032 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5033 : errmsg("number is out of range")));
5034 318 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5035 312 : found = true;
5036 : }
5037 :
5038 131582 : *ptr = cp;
5039 131582 : *value = val;
5040 :
5041 131582 : return found;
5042 : }
5043 :
5044 : /*
5045 : * Parse a format specifier (generally following the SUS printf spec).
5046 : *
5047 : * We have already advanced over the initial '%', and we are looking for
5048 : * [argpos][flags][width]type (but the type character is not consumed here).
5049 : *
5050 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5051 : * Output parameters:
5052 : * argpos: argument position for value to be printed. -1 means unspecified.
5053 : * widthpos: argument position for width. Zero means the argument position
5054 : * was unspecified (ie, take the next arg) and -1 means no width
5055 : * argument (width was omitted or specified as a constant).
5056 : * flags: bitmask of flags.
5057 : * width: directly-specified width value. Zero means the width was omitted
5058 : * (note it's not necessary to distinguish this case from an explicit
5059 : * zero width value).
5060 : *
5061 : * The function result is the next character position to be parsed, ie, the
5062 : * location where the type character is/should be.
5063 : *
5064 : * Note parsing invariant: at least one character is known available before
5065 : * string end (end_ptr) at entry, and this is still true at exit.
5066 : */
5067 : static const char *
5068 65812 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5069 : int *argpos, int *widthpos,
5070 : int *flags, int *width)
5071 : {
5072 65812 : const char *cp = start_ptr;
5073 : int n;
5074 :
5075 : /* set defaults for output parameters */
5076 65812 : *argpos = -1;
5077 65812 : *widthpos = -1;
5078 65812 : *flags = 0;
5079 65812 : *width = 0;
5080 :
5081 : /* try to identify first number */
5082 65812 : if (text_format_parse_digits(&cp, end_ptr, &n))
5083 : {
5084 174 : if (*cp != '$')
5085 : {
5086 : /* Must be just a width and a type, so we're done */
5087 24 : *width = n;
5088 24 : return cp;
5089 : }
5090 : /* The number was argument position */
5091 150 : *argpos = n;
5092 : /* Explicit 0 for argument index is immediately refused */
5093 150 : if (n == 0)
5094 6 : ereport(ERROR,
5095 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5096 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5097 144 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5098 : }
5099 :
5100 : /* Handle flags (only minus is supported now) */
5101 65806 : while (*cp == '-')
5102 : {
5103 30 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5104 30 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5105 : }
5106 :
5107 65776 : if (*cp == '*')
5108 : {
5109 : /* Handle indirect width */
5110 48 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5111 48 : if (text_format_parse_digits(&cp, end_ptr, &n))
5112 : {
5113 : /* number in this position must be closed by $ */
5114 42 : if (*cp != '$')
5115 0 : ereport(ERROR,
5116 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5117 : errmsg("width argument position must be ended by \"$\"")));
5118 : /* The number was width argument position */
5119 42 : *widthpos = n;
5120 : /* Explicit 0 for argument index is immediately refused */
5121 42 : if (n == 0)
5122 6 : ereport(ERROR,
5123 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5124 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5125 36 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5126 : }
5127 : else
5128 6 : *widthpos = 0; /* width's argument position is unspecified */
5129 : }
5130 : else
5131 : {
5132 : /* Check for direct width specification */
5133 65728 : if (text_format_parse_digits(&cp, end_ptr, &n))
5134 30 : *width = n;
5135 : }
5136 :
5137 : /* cp should now be pointing at type character */
5138 65764 : return cp;
5139 : }
5140 :
5141 : /*
5142 : * Format a %s, %I, or %L conversion
5143 : */
5144 : static void
5145 65758 : text_format_string_conversion(StringInfo buf, char conversion,
5146 : FmgrInfo *typOutputInfo,
5147 : Datum value, bool isNull,
5148 : int flags, int width)
5149 : {
5150 : char *str;
5151 :
5152 : /* Handle NULL arguments before trying to stringify the value. */
5153 65758 : if (isNull)
5154 : {
5155 342 : if (conversion == 's')
5156 270 : text_format_append_string(buf, "", flags, width);
5157 72 : else if (conversion == 'L')
5158 66 : text_format_append_string(buf, "NULL", flags, width);
5159 6 : else if (conversion == 'I')
5160 6 : ereport(ERROR,
5161 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5162 : errmsg("null values cannot be formatted as an SQL identifier")));
5163 336 : return;
5164 : }
5165 :
5166 : /* Stringify. */
5167 65416 : str = OutputFunctionCall(typOutputInfo, value);
5168 :
5169 : /* Escape. */
5170 65416 : if (conversion == 'I')
5171 : {
5172 : /* quote_identifier may or may not allocate a new string. */
5173 4902 : text_format_append_string(buf, quote_identifier(str), flags, width);
5174 : }
5175 60514 : else if (conversion == 'L')
5176 : {
5177 3232 : char *qstr = quote_literal_cstr(str);
5178 :
5179 3232 : text_format_append_string(buf, qstr, flags, width);
5180 : /* quote_literal_cstr() always allocates a new string */
5181 3232 : pfree(qstr);
5182 : }
5183 : else
5184 57282 : text_format_append_string(buf, str, flags, width);
5185 :
5186 : /* Cleanup. */
5187 65416 : pfree(str);
5188 : }
5189 :
5190 : /*
5191 : * Append str to buf, padding as directed by flags/width
5192 : */
5193 : static void
5194 65752 : text_format_append_string(StringInfo buf, const char *str,
5195 : int flags, int width)
5196 : {
5197 65752 : bool align_to_left = false;
5198 : int len;
5199 :
5200 : /* fast path for typical easy case */
5201 65752 : if (width == 0)
5202 : {
5203 65668 : appendStringInfoString(buf, str);
5204 65668 : return;
5205 : }
5206 :
5207 84 : if (width < 0)
5208 : {
5209 : /* Negative width: implicit '-' flag, then take absolute value */
5210 6 : align_to_left = true;
5211 : /* -INT_MIN is undefined */
5212 6 : if (width <= INT_MIN)
5213 0 : ereport(ERROR,
5214 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5215 : errmsg("number is out of range")));
5216 6 : width = -width;
5217 : }
5218 78 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
5219 24 : align_to_left = true;
5220 :
5221 84 : len = pg_mbstrlen(str);
5222 84 : if (align_to_left)
5223 : {
5224 : /* left justify */
5225 30 : appendStringInfoString(buf, str);
5226 30 : if (len < width)
5227 30 : appendStringInfoSpaces(buf, width - len);
5228 : }
5229 : else
5230 : {
5231 : /* right justify */
5232 54 : if (len < width)
5233 54 : appendStringInfoSpaces(buf, width - len);
5234 54 : appendStringInfoString(buf, str);
5235 : }
5236 : }
5237 :
5238 : /*
5239 : * text_format_nv - nonvariadic wrapper for text_format function.
5240 : *
5241 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5242 : * which checks that all built-in functions that share the implementing C
5243 : * function take the same number of arguments.
5244 : */
5245 : Datum
5246 3810 : text_format_nv(PG_FUNCTION_ARGS)
5247 : {
5248 3810 : return text_format(fcinfo);
5249 : }
5250 :
5251 : /*
5252 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
5253 : * for this use case.
5254 : */
5255 : static inline bool
5256 0 : rest_of_char_same(const char *s1, const char *s2, int len)
5257 : {
5258 0 : while (len > 0)
5259 : {
5260 0 : len--;
5261 0 : if (s1[len] != s2[len])
5262 0 : return false;
5263 : }
5264 0 : return true;
5265 : }
5266 :
5267 : /* Expand each Levenshtein distance variant */
5268 : #include "levenshtein.c"
5269 : #define LEVENSHTEIN_LESS_EQUAL
5270 : #include "levenshtein.c"
5271 :
5272 :
5273 : /*
5274 : * The following *ClosestMatch() functions can be used to determine whether a
5275 : * user-provided string resembles any known valid values, which is useful for
5276 : * providing hints in log messages, among other things. Use these functions
5277 : * like so:
5278 : *
5279 : * initClosestMatch(&state, source_string, max_distance);
5280 : *
5281 : * for (int i = 0; i < num_valid_strings; i++)
5282 : * updateClosestMatch(&state, valid_strings[i]);
5283 : *
5284 : * closestMatch = getClosestMatch(&state);
5285 : */
5286 :
5287 : /*
5288 : * Initialize the given state with the source string and maximum Levenshtein
5289 : * distance to consider.
5290 : */
5291 : void
5292 78 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
5293 : {
5294 : Assert(state);
5295 : Assert(max_d >= 0);
5296 :
5297 78 : state->source = source;
5298 78 : state->min_d = -1;
5299 78 : state->max_d = max_d;
5300 78 : state->match = NULL;
5301 78 : }
5302 :
5303 : /*
5304 : * If the candidate string is a closer match than the current one saved (or
5305 : * there is no match saved), save it as the closest match.
5306 : *
5307 : * If the source or candidate string is NULL, empty, or too long, this function
5308 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
5309 : * allowed or more than half the characters are different, no action is taken.
5310 : */
5311 : void
5312 804 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
5313 : {
5314 : int dist;
5315 :
5316 : Assert(state);
5317 :
5318 804 : if (state->source == NULL || state->source[0] == '\0' ||
5319 804 : candidate == NULL || candidate[0] == '\0')
5320 0 : return;
5321 :
5322 : /*
5323 : * To avoid ERROR-ing, we check the lengths here instead of setting
5324 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
5325 : */
5326 804 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
5327 804 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
5328 0 : return;
5329 :
5330 804 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
5331 804 : candidate, strlen(candidate), 1, 1, 1,
5332 : state->max_d, true);
5333 804 : if (dist <= state->max_d &&
5334 62 : dist <= strlen(state->source) / 2 &&
5335 14 : (state->min_d == -1 || dist < state->min_d))
5336 : {
5337 14 : state->min_d = dist;
5338 14 : state->match = candidate;
5339 : }
5340 : }
5341 :
5342 : /*
5343 : * Return the closest match. If no suitable candidates were provided via
5344 : * updateClosestMatch(), return NULL.
5345 : */
5346 : const char *
5347 78 : getClosestMatch(ClosestMatchState *state)
5348 : {
5349 : Assert(state);
5350 :
5351 78 : return state->match;
5352 : }
5353 :
5354 :
5355 : /*
5356 : * Unicode support
5357 : */
5358 :
5359 : static UnicodeNormalizationForm
5360 210 : unicode_norm_form_from_string(const char *formstr)
5361 : {
5362 210 : UnicodeNormalizationForm form = -1;
5363 :
5364 : /*
5365 : * Might as well check this while we're here.
5366 : */
5367 210 : if (GetDatabaseEncoding() != PG_UTF8)
5368 0 : ereport(ERROR,
5369 : (errcode(ERRCODE_SYNTAX_ERROR),
5370 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
5371 :
5372 210 : if (pg_strcasecmp(formstr, "NFC") == 0)
5373 66 : form = UNICODE_NFC;
5374 144 : else if (pg_strcasecmp(formstr, "NFD") == 0)
5375 60 : form = UNICODE_NFD;
5376 84 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
5377 36 : form = UNICODE_NFKC;
5378 48 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
5379 36 : form = UNICODE_NFKD;
5380 : else
5381 12 : ereport(ERROR,
5382 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5383 : errmsg("invalid normalization form: %s", formstr)));
5384 :
5385 198 : return form;
5386 : }
5387 :
5388 : /*
5389 : * Returns version of Unicode used by Postgres in "major.minor" format (the
5390 : * same format as the Unicode version reported by ICU). The third component
5391 : * ("update version") never involves additions to the character repertoire and
5392 : * is unimportant for most purposes.
5393 : *
5394 : * See: https://unicode.org/versions/
5395 : */
5396 : Datum
5397 30 : unicode_version(PG_FUNCTION_ARGS)
5398 : {
5399 30 : PG_RETURN_TEXT_P(cstring_to_text(PG_UNICODE_VERSION));
5400 : }
5401 :
5402 : /*
5403 : * Returns version of Unicode used by ICU, if enabled; otherwise NULL.
5404 : */
5405 : Datum
5406 2 : icu_unicode_version(PG_FUNCTION_ARGS)
5407 : {
5408 : #ifdef USE_ICU
5409 2 : PG_RETURN_TEXT_P(cstring_to_text(U_UNICODE_VERSION));
5410 : #else
5411 : PG_RETURN_NULL();
5412 : #endif
5413 : }
5414 :
5415 : /*
5416 : * Check whether the string contains only assigned Unicode code
5417 : * points. Requires that the database encoding is UTF-8.
5418 : */
5419 : Datum
5420 12 : unicode_assigned(PG_FUNCTION_ARGS)
5421 : {
5422 12 : text *input = PG_GETARG_TEXT_PP(0);
5423 : unsigned char *p;
5424 : int size;
5425 :
5426 12 : if (GetDatabaseEncoding() != PG_UTF8)
5427 0 : ereport(ERROR,
5428 : (errmsg("Unicode categorization can only be performed if server encoding is UTF8")));
5429 :
5430 : /* convert to pg_wchar */
5431 12 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5432 12 : p = (unsigned char *) VARDATA_ANY(input);
5433 48 : for (int i = 0; i < size; i++)
5434 : {
5435 42 : pg_wchar uchar = utf8_to_unicode(p);
5436 42 : int category = unicode_category(uchar);
5437 :
5438 42 : if (category == PG_U_UNASSIGNED)
5439 6 : PG_RETURN_BOOL(false);
5440 :
5441 36 : p += pg_utf_mblen(p);
5442 : }
5443 :
5444 6 : PG_RETURN_BOOL(true);
5445 : }
5446 :
5447 : Datum
5448 72 : unicode_normalize_func(PG_FUNCTION_ARGS)
5449 : {
5450 72 : text *input = PG_GETARG_TEXT_PP(0);
5451 72 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5452 : UnicodeNormalizationForm form;
5453 : int size;
5454 : pg_wchar *input_chars;
5455 : pg_wchar *output_chars;
5456 : unsigned char *p;
5457 : text *result;
5458 : int i;
5459 :
5460 72 : form = unicode_norm_form_from_string(formstr);
5461 :
5462 : /* convert to pg_wchar */
5463 66 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5464 66 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
5465 66 : p = (unsigned char *) VARDATA_ANY(input);
5466 288 : for (i = 0; i < size; i++)
5467 : {
5468 222 : input_chars[i] = utf8_to_unicode(p);
5469 222 : p += pg_utf_mblen(p);
5470 : }
5471 66 : input_chars[i] = (pg_wchar) '\0';
5472 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5473 :
5474 : /* action */
5475 66 : output_chars = unicode_normalize(form, input_chars);
5476 :
5477 : /* convert back to UTF-8 string */
5478 66 : size = 0;
5479 306 : for (pg_wchar *wp = output_chars; *wp; wp++)
5480 : {
5481 : unsigned char buf[4];
5482 :
5483 240 : unicode_to_utf8(*wp, buf);
5484 240 : size += pg_utf_mblen(buf);
5485 : }
5486 :
5487 66 : result = palloc(size + VARHDRSZ);
5488 66 : SET_VARSIZE(result, size + VARHDRSZ);
5489 :
5490 66 : p = (unsigned char *) VARDATA_ANY(result);
5491 306 : for (pg_wchar *wp = output_chars; *wp; wp++)
5492 : {
5493 240 : unicode_to_utf8(*wp, p);
5494 240 : p += pg_utf_mblen(p);
5495 : }
5496 : Assert((char *) p == (char *) result + size + VARHDRSZ);
5497 :
5498 66 : PG_RETURN_TEXT_P(result);
5499 : }
5500 :
5501 : /*
5502 : * Check whether the string is in the specified Unicode normalization form.
5503 : *
5504 : * This is done by converting the string to the specified normal form and then
5505 : * comparing that to the original string. To speed that up, we also apply the
5506 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
5507 : * answer for many strings by just scanning the string once.
5508 : *
5509 : * This function should generally be optimized for the case where the string
5510 : * is in fact normalized. In that case, we'll end up looking at the entire
5511 : * string, so it's probably not worth doing any incremental conversion etc.
5512 : */
5513 : Datum
5514 138 : unicode_is_normalized(PG_FUNCTION_ARGS)
5515 : {
5516 138 : text *input = PG_GETARG_TEXT_PP(0);
5517 138 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5518 : UnicodeNormalizationForm form;
5519 : int size;
5520 : pg_wchar *input_chars;
5521 : pg_wchar *output_chars;
5522 : unsigned char *p;
5523 : int i;
5524 : UnicodeNormalizationQC quickcheck;
5525 : int output_size;
5526 : bool result;
5527 :
5528 138 : form = unicode_norm_form_from_string(formstr);
5529 :
5530 : /* convert to pg_wchar */
5531 132 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5532 132 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
5533 132 : p = (unsigned char *) VARDATA_ANY(input);
5534 504 : for (i = 0; i < size; i++)
5535 : {
5536 372 : input_chars[i] = utf8_to_unicode(p);
5537 372 : p += pg_utf_mblen(p);
5538 : }
5539 132 : input_chars[i] = (pg_wchar) '\0';
5540 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5541 :
5542 : /* quick check (see UAX #15) */
5543 132 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
5544 132 : if (quickcheck == UNICODE_NORM_QC_YES)
5545 42 : PG_RETURN_BOOL(true);
5546 90 : else if (quickcheck == UNICODE_NORM_QC_NO)
5547 12 : PG_RETURN_BOOL(false);
5548 :
5549 : /* normalize and compare with original */
5550 78 : output_chars = unicode_normalize(form, input_chars);
5551 :
5552 78 : output_size = 0;
5553 324 : for (pg_wchar *wp = output_chars; *wp; wp++)
5554 246 : output_size++;
5555 :
5556 114 : result = (size == output_size) &&
5557 36 : (memcmp(input_chars, output_chars, size * sizeof(pg_wchar)) == 0);
5558 :
5559 78 : PG_RETURN_BOOL(result);
5560 : }
5561 :
5562 : /*
5563 : * Check if first n chars are hexadecimal digits
5564 : */
5565 : static bool
5566 156 : isxdigits_n(const char *instr, size_t n)
5567 : {
5568 660 : for (size_t i = 0; i < n; i++)
5569 570 : if (!isxdigit((unsigned char) instr[i]))
5570 66 : return false;
5571 :
5572 90 : return true;
5573 : }
5574 :
5575 : static unsigned int
5576 504 : hexval(unsigned char c)
5577 : {
5578 504 : if (c >= '0' && c <= '9')
5579 384 : return c - '0';
5580 120 : if (c >= 'a' && c <= 'f')
5581 60 : return c - 'a' + 0xA;
5582 60 : if (c >= 'A' && c <= 'F')
5583 60 : return c - 'A' + 0xA;
5584 0 : elog(ERROR, "invalid hexadecimal digit");
5585 : return 0; /* not reached */
5586 : }
5587 :
5588 : /*
5589 : * Translate string with hexadecimal digits to number
5590 : */
5591 : static unsigned int
5592 90 : hexval_n(const char *instr, size_t n)
5593 : {
5594 90 : unsigned int result = 0;
5595 :
5596 594 : for (size_t i = 0; i < n; i++)
5597 504 : result += hexval(instr[i]) << (4 * (n - i - 1));
5598 :
5599 90 : return result;
5600 : }
5601 :
5602 : /*
5603 : * Replaces Unicode escape sequences by Unicode characters
5604 : */
5605 : Datum
5606 66 : unistr(PG_FUNCTION_ARGS)
5607 : {
5608 66 : text *input_text = PG_GETARG_TEXT_PP(0);
5609 : char *instr;
5610 : int len;
5611 : StringInfoData str;
5612 : text *result;
5613 66 : pg_wchar pair_first = 0;
5614 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
5615 :
5616 66 : instr = VARDATA_ANY(input_text);
5617 66 : len = VARSIZE_ANY_EXHDR(input_text);
5618 :
5619 66 : initStringInfo(&str);
5620 :
5621 510 : while (len > 0)
5622 : {
5623 486 : if (instr[0] == '\\')
5624 : {
5625 102 : if (len >= 2 &&
5626 102 : instr[1] == '\\')
5627 : {
5628 6 : if (pair_first)
5629 0 : goto invalid_pair;
5630 6 : appendStringInfoChar(&str, '\\');
5631 6 : instr += 2;
5632 6 : len -= 2;
5633 : }
5634 96 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
5635 66 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
5636 30 : {
5637 : pg_wchar unicode;
5638 42 : int offset = instr[1] == 'u' ? 2 : 1;
5639 :
5640 42 : unicode = hexval_n(instr + offset, 4);
5641 :
5642 42 : if (!is_valid_unicode_codepoint(unicode))
5643 0 : ereport(ERROR,
5644 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5645 : errmsg("invalid Unicode code point: %04X", unicode));
5646 :
5647 42 : if (pair_first)
5648 : {
5649 12 : if (is_utf16_surrogate_second(unicode))
5650 : {
5651 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5652 0 : pair_first = 0;
5653 : }
5654 : else
5655 12 : goto invalid_pair;
5656 : }
5657 30 : else if (is_utf16_surrogate_second(unicode))
5658 0 : goto invalid_pair;
5659 :
5660 30 : if (is_utf16_surrogate_first(unicode))
5661 18 : pair_first = unicode;
5662 : else
5663 : {
5664 12 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5665 12 : appendStringInfoString(&str, cbuf);
5666 : }
5667 :
5668 30 : instr += 4 + offset;
5669 30 : len -= 4 + offset;
5670 : }
5671 54 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
5672 12 : {
5673 : pg_wchar unicode;
5674 :
5675 24 : unicode = hexval_n(instr + 2, 6);
5676 :
5677 24 : if (!is_valid_unicode_codepoint(unicode))
5678 6 : ereport(ERROR,
5679 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5680 : errmsg("invalid Unicode code point: %04X", unicode));
5681 :
5682 18 : if (pair_first)
5683 : {
5684 6 : if (is_utf16_surrogate_second(unicode))
5685 : {
5686 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5687 0 : pair_first = 0;
5688 : }
5689 : else
5690 6 : goto invalid_pair;
5691 : }
5692 12 : else if (is_utf16_surrogate_second(unicode))
5693 0 : goto invalid_pair;
5694 :
5695 12 : if (is_utf16_surrogate_first(unicode))
5696 6 : pair_first = unicode;
5697 : else
5698 : {
5699 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5700 6 : appendStringInfoString(&str, cbuf);
5701 : }
5702 :
5703 12 : instr += 8;
5704 12 : len -= 8;
5705 : }
5706 30 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
5707 12 : {
5708 : pg_wchar unicode;
5709 :
5710 24 : unicode = hexval_n(instr + 2, 8);
5711 :
5712 24 : if (!is_valid_unicode_codepoint(unicode))
5713 6 : ereport(ERROR,
5714 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5715 : errmsg("invalid Unicode code point: %04X", unicode));
5716 :
5717 18 : if (pair_first)
5718 : {
5719 6 : if (is_utf16_surrogate_second(unicode))
5720 : {
5721 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5722 0 : pair_first = 0;
5723 : }
5724 : else
5725 6 : goto invalid_pair;
5726 : }
5727 12 : else if (is_utf16_surrogate_second(unicode))
5728 0 : goto invalid_pair;
5729 :
5730 12 : if (is_utf16_surrogate_first(unicode))
5731 6 : pair_first = unicode;
5732 : else
5733 : {
5734 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5735 6 : appendStringInfoString(&str, cbuf);
5736 : }
5737 :
5738 12 : instr += 10;
5739 12 : len -= 10;
5740 : }
5741 : else
5742 6 : ereport(ERROR,
5743 : (errcode(ERRCODE_SYNTAX_ERROR),
5744 : errmsg("invalid Unicode escape"),
5745 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
5746 : }
5747 : else
5748 : {
5749 384 : if (pair_first)
5750 0 : goto invalid_pair;
5751 :
5752 384 : appendStringInfoChar(&str, *instr++);
5753 384 : len--;
5754 : }
5755 : }
5756 :
5757 : /* unfinished surrogate pair? */
5758 24 : if (pair_first)
5759 6 : goto invalid_pair;
5760 :
5761 18 : result = cstring_to_text_with_len(str.data, str.len);
5762 18 : pfree(str.data);
5763 :
5764 18 : PG_RETURN_TEXT_P(result);
5765 :
5766 30 : invalid_pair:
5767 30 : ereport(ERROR,
5768 : (errcode(ERRCODE_SYNTAX_ERROR),
5769 : errmsg("invalid Unicode surrogate pair")));
5770 : PG_RETURN_NULL(); /* keep compiler quiet */
5771 : }
|
