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 24843362 : cstring_to_text(const char *s)
182 : {
183 24843362 : 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 27548918 : cstring_to_text_with_len(const char *s, int len)
194 : {
195 27548918 : text *result = (text *) palloc(len + VARHDRSZ);
196 :
197 27548918 : SET_VARSIZE(result, len + VARHDRSZ);
198 27548918 : memcpy(VARDATA(result), s, len);
199 :
200 27548918 : 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 17014496 : text_to_cstring(const text *t)
215 : {
216 : /* must cast away the const, unfortunately */
217 17014496 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
218 17014496 : int len = VARSIZE_ANY_EXHDR(tunpacked);
219 : char *result;
220 :
221 17014496 : result = (char *) palloc(len + 1);
222 17014496 : memcpy(result, VARDATA_ANY(tunpacked), len);
223 17014496 : result[len] = '\0';
224 :
225 17014496 : if (tunpacked != t)
226 43616 : pfree(tunpacked);
227 :
228 17014496 : return result;
229 : }
230 :
231 : /*
232 : * text_to_cstring_buffer
233 : *
234 : * Copy a text value into a caller-supplied buffer of size dst_len.
235 : *
236 : * The text string is truncated if necessary to fit. The result is
237 : * guaranteed null-terminated (unless dst_len == 0).
238 : *
239 : * We support being passed a compressed or toasted text value.
240 : * This is a bit bogus since such values shouldn't really be referred to as
241 : * "text *", but it seems useful for robustness. If we didn't handle that
242 : * case here, we'd need another routine that did, anyway.
243 : */
244 : void
245 1006 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
246 : {
247 : /* must cast away the const, unfortunately */
248 1006 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
249 1006 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
250 :
251 1006 : if (dst_len > 0)
252 : {
253 1006 : dst_len--;
254 1006 : if (dst_len >= src_len)
255 1006 : dst_len = src_len;
256 : else /* ensure truncation is encoding-safe */
257 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
258 1006 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
259 1006 : dst[dst_len] = '\0';
260 : }
261 :
262 1006 : if (srcunpacked != src)
263 0 : pfree(srcunpacked);
264 1006 : }
265 :
266 :
267 : /*****************************************************************************
268 : * USER I/O ROUTINES *
269 : *****************************************************************************/
270 :
271 : /*
272 : * textin - converts cstring to internal representation
273 : */
274 : Datum
275 21586520 : textin(PG_FUNCTION_ARGS)
276 : {
277 21586520 : char *inputText = PG_GETARG_CSTRING(0);
278 :
279 21586520 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
280 : }
281 :
282 : /*
283 : * textout - converts internal representation to cstring
284 : */
285 : Datum
286 8118094 : textout(PG_FUNCTION_ARGS)
287 : {
288 8118094 : Datum txt = PG_GETARG_DATUM(0);
289 :
290 8118094 : 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 4944 : textsend(PG_FUNCTION_ARGS)
316 : {
317 4944 : text *t = PG_GETARG_TEXT_PP(0);
318 : StringInfoData buf;
319 :
320 4944 : pq_begintypsend(&buf);
321 4944 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
322 4944 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
323 : }
324 :
325 :
326 : /*
327 : * unknownin - converts cstring to internal representation
328 : */
329 : Datum
330 0 : unknownin(PG_FUNCTION_ARGS)
331 : {
332 0 : char *str = PG_GETARG_CSTRING(0);
333 :
334 : /* representation is same as cstring */
335 0 : PG_RETURN_CSTRING(pstrdup(str));
336 : }
337 :
338 : /*
339 : * unknownout - converts internal representation to cstring
340 : */
341 : Datum
342 940 : unknownout(PG_FUNCTION_ARGS)
343 : {
344 : /* representation is same as cstring */
345 940 : char *str = PG_GETARG_CSTRING(0);
346 :
347 940 : PG_RETURN_CSTRING(pstrdup(str));
348 : }
349 :
350 : /*
351 : * unknownrecv - converts external binary format to unknown
352 : */
353 : Datum
354 0 : unknownrecv(PG_FUNCTION_ARGS)
355 : {
356 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
357 : char *str;
358 : int nbytes;
359 :
360 0 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
361 : /* representation is same as cstring */
362 0 : PG_RETURN_CSTRING(str);
363 : }
364 :
365 : /*
366 : * unknownsend - converts unknown to binary format
367 : */
368 : Datum
369 0 : unknownsend(PG_FUNCTION_ARGS)
370 : {
371 : /* representation is same as cstring */
372 0 : char *str = PG_GETARG_CSTRING(0);
373 : StringInfoData buf;
374 :
375 0 : pq_begintypsend(&buf);
376 0 : pq_sendtext(&buf, str, strlen(str));
377 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
378 : }
379 :
380 :
381 : /* ========== PUBLIC ROUTINES ========== */
382 :
383 : /*
384 : * textlen -
385 : * returns the logical length of a text*
386 : * (which is less than the VARSIZE of the text*)
387 : */
388 : Datum
389 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 : return (toast_raw_datum_size(str) - VARHDRSZ);
412 : else
413 : {
414 430792 : text *t = DatumGetTextPP(str);
415 :
416 430792 : return (pg_mbstrlen_with_len(VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t)));
417 : }
418 : }
419 :
420 : /*
421 : * textoctetlen -
422 : * returns the physical length of a text*
423 : * (which is less than the VARSIZE of the text*)
424 : */
425 : Datum
426 70 : textoctetlen(PG_FUNCTION_ARGS)
427 : {
428 70 : Datum str = PG_GETARG_DATUM(0);
429 :
430 : /* We need not detoast the input at all */
431 70 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
432 : }
433 :
434 : /*
435 : * textcat -
436 : * takes two text* and returns a text* that is the concatenation of
437 : * the two.
438 : *
439 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
440 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
441 : * Allocate space for output in all cases.
442 : * XXX - thomas 1997-07-10
443 : */
444 : Datum
445 1950240 : textcat(PG_FUNCTION_ARGS)
446 : {
447 1950240 : text *t1 = PG_GETARG_TEXT_PP(0);
448 1950240 : text *t2 = PG_GETARG_TEXT_PP(1);
449 :
450 1950240 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
451 : }
452 :
453 : /*
454 : * text_catenate
455 : * Guts of textcat(), broken out so it can be used by other functions
456 : *
457 : * Arguments can be in short-header form, but not compressed or out-of-line
458 : */
459 : static text *
460 1950320 : text_catenate(text *t1, text *t2)
461 : {
462 : text *result;
463 : int len1,
464 : len2,
465 : len;
466 : char *ptr;
467 :
468 1950320 : len1 = VARSIZE_ANY_EXHDR(t1);
469 1950320 : len2 = VARSIZE_ANY_EXHDR(t2);
470 :
471 : /* paranoia ... probably should throw error instead? */
472 1950320 : if (len1 < 0)
473 0 : len1 = 0;
474 1950320 : if (len2 < 0)
475 0 : len2 = 0;
476 :
477 1950320 : len = len1 + len2 + VARHDRSZ;
478 1950320 : result = (text *) palloc(len);
479 :
480 : /* Set size of result string... */
481 1950320 : SET_VARSIZE(result, len);
482 :
483 : /* Fill data field of result string... */
484 1950320 : ptr = VARDATA(result);
485 1950320 : if (len1 > 0)
486 1949496 : memcpy(ptr, VARDATA_ANY(t1), len1);
487 1950320 : if (len2 > 0)
488 1950110 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
489 :
490 1950320 : return result;
491 : }
492 :
493 : /*
494 : * charlen_to_bytelen()
495 : * Compute the number of bytes occupied by n characters starting at *p
496 : *
497 : * It is caller's responsibility that there actually are n characters;
498 : * the string need not be null-terminated.
499 : */
500 : static int
501 17222 : charlen_to_bytelen(const char *p, int n)
502 : {
503 17222 : if (pg_database_encoding_max_length() == 1)
504 : {
505 : /* Optimization for single-byte encodings */
506 180 : return n;
507 : }
508 : else
509 : {
510 : const char *s;
511 :
512 6060778 : for (s = p; n > 0; n--)
513 6043736 : s += pg_mblen(s);
514 :
515 17042 : return s - p;
516 : }
517 : }
518 :
519 : /*
520 : * text_substr()
521 : * Return a substring starting at the specified position.
522 : * - thomas 1997-12-31
523 : *
524 : * Input:
525 : * - string
526 : * - starting position (is one-based)
527 : * - string length
528 : *
529 : * If the starting position is zero or less, then return from the start of the string
530 : * adjusting the length to be consistent with the "negative start" per SQL.
531 : * If the length is less than zero, return the remaining string.
532 : *
533 : * Added multibyte support.
534 : * - Tatsuo Ishii 1998-4-21
535 : * Changed behavior if starting position is less than one to conform to SQL behavior.
536 : * Formerly returned the entire string; now returns a portion.
537 : * - Thomas Lockhart 1998-12-10
538 : * Now uses faster TOAST-slicing interface
539 : * - John Gray 2002-02-22
540 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
541 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
542 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
543 : * S > LC and < LC + 4 sometimes garbage characters are returned.
544 : * - Joe Conway 2002-08-10
545 : */
546 : Datum
547 658418 : text_substr(PG_FUNCTION_ARGS)
548 : {
549 658418 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
550 : PG_GETARG_INT32(1),
551 : PG_GETARG_INT32(2),
552 : false));
553 : }
554 :
555 : /*
556 : * text_substr_no_len -
557 : * Wrapper to avoid opr_sanity failure due to
558 : * one function accepting a different number of args.
559 : */
560 : Datum
561 36 : text_substr_no_len(PG_FUNCTION_ARGS)
562 : {
563 36 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
564 : PG_GETARG_INT32(1),
565 : -1, true));
566 : }
567 :
568 : /*
569 : * text_substring -
570 : * Does the real work for text_substr() and text_substr_no_len()
571 : *
572 : * This is broken out so it can be called directly by other string processing
573 : * functions. Note that the argument is passed as a Datum, to indicate that
574 : * it may still be in compressed/toasted form. We can avoid detoasting all
575 : * of it in some cases.
576 : *
577 : * The result is always a freshly palloc'd datum.
578 : */
579 : static text *
580 698566 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
581 : {
582 698566 : int32 eml = pg_database_encoding_max_length();
583 698566 : int32 S = start; /* start position */
584 : int32 S1; /* adjusted start position */
585 : int32 L1; /* adjusted substring length */
586 : int32 E; /* end position */
587 :
588 : /*
589 : * SQL99 says S can be zero or negative (which we don't document), but we
590 : * still must fetch from the start of the string.
591 : * https://www.postgresql.org/message-id/170905442373.643.11536838320909376197%40wrigleys.postgresql.org
592 : */
593 698566 : S1 = Max(S, 1);
594 :
595 : /* life is easy if the encoding max length is 1 */
596 698566 : if (eml == 1)
597 : {
598 22 : if (length_not_specified) /* special case - get length to end of
599 : * string */
600 0 : L1 = -1;
601 22 : else if (length < 0)
602 : {
603 : /* SQL99 says to throw an error for E < S, i.e., negative length */
604 0 : ereport(ERROR,
605 : (errcode(ERRCODE_SUBSTRING_ERROR),
606 : errmsg("negative substring length not allowed")));
607 : L1 = -1; /* silence stupider compilers */
608 : }
609 22 : else if (pg_add_s32_overflow(S, length, &E))
610 : {
611 : /*
612 : * L could be large enough for S + L to overflow, in which case
613 : * the substring must run to end of string.
614 : */
615 0 : L1 = -1;
616 : }
617 : else
618 : {
619 : /*
620 : * A zero or negative value for the end position can happen if the
621 : * start was negative or one. SQL99 says to return a zero-length
622 : * string.
623 : */
624 22 : if (E < 1)
625 0 : return cstring_to_text("");
626 :
627 22 : L1 = E - S1;
628 : }
629 :
630 : /*
631 : * If the start position is past the end of the string, SQL99 says to
632 : * return a zero-length string -- DatumGetTextPSlice() will do that
633 : * for us. We need only convert S1 to zero-based starting position.
634 : */
635 22 : return DatumGetTextPSlice(str, S1 - 1, L1);
636 : }
637 698544 : else if (eml > 1)
638 : {
639 : /*
640 : * When encoding max length is > 1, we can't get LC without
641 : * detoasting, so we'll grab a conservatively large slice now and go
642 : * back later to do the right thing
643 : */
644 : int32 slice_start;
645 : int32 slice_size;
646 : int32 slice_strlen;
647 : text *slice;
648 : int32 E1;
649 : int32 i;
650 : char *p;
651 : char *s;
652 : text *ret;
653 :
654 : /*
655 : * We need to start at position zero because there is no way to know
656 : * in advance which byte offset corresponds to the supplied start
657 : * position.
658 : */
659 698544 : slice_start = 0;
660 :
661 698544 : if (length_not_specified) /* special case - get length to end of
662 : * string */
663 76 : slice_size = L1 = -1;
664 698468 : else if (length < 0)
665 : {
666 : /* SQL99 says to throw an error for E < S, i.e., negative length */
667 12 : ereport(ERROR,
668 : (errcode(ERRCODE_SUBSTRING_ERROR),
669 : errmsg("negative substring length not allowed")));
670 : slice_size = L1 = -1; /* silence stupider compilers */
671 : }
672 698456 : else if (pg_add_s32_overflow(S, length, &E))
673 : {
674 : /*
675 : * L could be large enough for S + L to overflow, in which case
676 : * the substring must run to end of string.
677 : */
678 6 : slice_size = L1 = -1;
679 : }
680 : else
681 : {
682 : /*
683 : * A zero or negative value for the end position can happen if the
684 : * start was negative or one. SQL99 says to return a zero-length
685 : * string.
686 : */
687 698450 : if (E < 1)
688 0 : return cstring_to_text("");
689 :
690 : /*
691 : * if E is past the end of the string, the tuple toaster will
692 : * truncate the length for us
693 : */
694 698450 : L1 = E - S1;
695 :
696 : /*
697 : * Total slice size in bytes can't be any longer than the start
698 : * position plus substring length times the encoding max length.
699 : * If that overflows, we can just use -1.
700 : */
701 698450 : if (pg_mul_s32_overflow(E, eml, &slice_size))
702 6 : slice_size = -1;
703 : }
704 :
705 : /*
706 : * If we're working with an untoasted source, no need to do an extra
707 : * copying step.
708 : */
709 1397010 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
710 698478 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
711 354 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
712 : else
713 698178 : slice = (text *) DatumGetPointer(str);
714 :
715 : /* see if we got back an empty string */
716 698532 : if (VARSIZE_ANY_EXHDR(slice) == 0)
717 : {
718 0 : if (slice != (text *) DatumGetPointer(str))
719 0 : pfree(slice);
720 0 : return cstring_to_text("");
721 : }
722 :
723 : /* Now we can get the actual length of the slice in MB characters */
724 698532 : slice_strlen = pg_mbstrlen_with_len(VARDATA_ANY(slice),
725 698532 : VARSIZE_ANY_EXHDR(slice));
726 :
727 : /*
728 : * Check that the start position wasn't > slice_strlen. If so, SQL99
729 : * says to return a zero-length string.
730 : */
731 698532 : if (S1 > slice_strlen)
732 : {
733 22 : if (slice != (text *) DatumGetPointer(str))
734 0 : pfree(slice);
735 22 : return cstring_to_text("");
736 : }
737 :
738 : /*
739 : * Adjust L1 and E1 now that we know the slice string length. Again
740 : * remember that S1 is one based, and slice_start is zero based.
741 : */
742 698510 : if (L1 > -1)
743 698450 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
744 : else
745 60 : E1 = slice_start + 1 + slice_strlen;
746 :
747 : /*
748 : * Find the start position in the slice; remember S1 is not zero based
749 : */
750 698510 : p = VARDATA_ANY(slice);
751 6719902 : for (i = 0; i < S1 - 1; i++)
752 6021392 : p += pg_mblen(p);
753 :
754 : /* hang onto a pointer to our start position */
755 698510 : s = p;
756 :
757 : /*
758 : * Count the actual bytes used by the substring of the requested
759 : * length.
760 : */
761 9939132 : for (i = S1; i < E1; i++)
762 9240622 : p += pg_mblen(p);
763 :
764 698510 : ret = (text *) palloc(VARHDRSZ + (p - s));
765 698510 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
766 698510 : memcpy(VARDATA(ret), s, (p - s));
767 :
768 698510 : if (slice != (text *) DatumGetPointer(str))
769 354 : pfree(slice);
770 :
771 698510 : return ret;
772 : }
773 : else
774 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
775 :
776 : /* not reached: suppress compiler warning */
777 : return NULL;
778 : }
779 :
780 : /*
781 : * textoverlay
782 : * Replace specified substring of first string with second
783 : *
784 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
785 : * This code is a direct implementation of what the standard says.
786 : */
787 : Datum
788 28 : textoverlay(PG_FUNCTION_ARGS)
789 : {
790 28 : text *t1 = PG_GETARG_TEXT_PP(0);
791 28 : text *t2 = PG_GETARG_TEXT_PP(1);
792 28 : int sp = PG_GETARG_INT32(2); /* substring start position */
793 28 : int sl = PG_GETARG_INT32(3); /* substring length */
794 :
795 28 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
796 : }
797 :
798 : Datum
799 12 : textoverlay_no_len(PG_FUNCTION_ARGS)
800 : {
801 12 : text *t1 = PG_GETARG_TEXT_PP(0);
802 12 : text *t2 = PG_GETARG_TEXT_PP(1);
803 12 : int sp = PG_GETARG_INT32(2); /* substring start position */
804 : int sl;
805 :
806 12 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
807 12 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
808 : }
809 :
810 : static text *
811 40 : text_overlay(text *t1, text *t2, int sp, int sl)
812 : {
813 : text *result;
814 : text *s1;
815 : text *s2;
816 : int sp_pl_sl;
817 :
818 : /*
819 : * Check for possible integer-overflow cases. For negative sp, throw a
820 : * "substring length" error because that's what should be expected
821 : * according to the spec's definition of OVERLAY().
822 : */
823 40 : if (sp <= 0)
824 0 : ereport(ERROR,
825 : (errcode(ERRCODE_SUBSTRING_ERROR),
826 : errmsg("negative substring length not allowed")));
827 40 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
828 0 : ereport(ERROR,
829 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
830 : errmsg("integer out of range")));
831 :
832 40 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
833 40 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
834 40 : result = text_catenate(s1, t2);
835 40 : result = text_catenate(result, s2);
836 :
837 40 : return result;
838 : }
839 :
840 : /*
841 : * textpos -
842 : * Return the position of the specified substring.
843 : * Implements the SQL POSITION() function.
844 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
845 : * - thomas 1997-07-27
846 : */
847 : Datum
848 130 : textpos(PG_FUNCTION_ARGS)
849 : {
850 130 : text *str = PG_GETARG_TEXT_PP(0);
851 130 : text *search_str = PG_GETARG_TEXT_PP(1);
852 :
853 130 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
854 : }
855 :
856 : /*
857 : * text_position -
858 : * Does the real work for textpos()
859 : *
860 : * Inputs:
861 : * t1 - string to be searched
862 : * t2 - pattern to match within t1
863 : * Result:
864 : * Character index of the first matched char, starting from 1,
865 : * or 0 if no match.
866 : *
867 : * This is broken out so it can be called directly by other string processing
868 : * functions.
869 : */
870 : static int
871 130 : text_position(text *t1, text *t2, Oid collid)
872 : {
873 : TextPositionState state;
874 : int result;
875 :
876 130 : check_collation_set(collid);
877 :
878 : /* Empty needle always matches at position 1 */
879 130 : if (VARSIZE_ANY_EXHDR(t2) < 1)
880 12 : return 1;
881 :
882 : /* Otherwise, can't match if haystack is shorter than needle */
883 118 : if (VARSIZE_ANY_EXHDR(t1) < VARSIZE_ANY_EXHDR(t2) &&
884 22 : pg_newlocale_from_collation(collid)->deterministic)
885 22 : return 0;
886 :
887 96 : text_position_setup(t1, t2, collid, &state);
888 : /* don't need greedy mode here */
889 96 : state.greedy = false;
890 :
891 96 : if (!text_position_next(&state))
892 24 : result = 0;
893 : else
894 72 : result = text_position_get_match_pos(&state);
895 96 : text_position_cleanup(&state);
896 96 : return result;
897 : }
898 :
899 :
900 : /*
901 : * text_position_setup, text_position_next, text_position_cleanup -
902 : * Component steps of text_position()
903 : *
904 : * These are broken out so that a string can be efficiently searched for
905 : * multiple occurrences of the same pattern. text_position_next may be
906 : * called multiple times, and it advances to the next match on each call.
907 : * text_position_get_match_ptr() and text_position_get_match_pos() return
908 : * a pointer or 1-based character position of the last match, respectively.
909 : *
910 : * The "state" variable is normally just a local variable in the caller.
911 : *
912 : * NOTE: text_position_next skips over the matched portion. For example,
913 : * searching for "xx" in "xxx" returns only one match, not two.
914 : */
915 :
916 : static void
917 1818 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
918 : {
919 1818 : int len1 = VARSIZE_ANY_EXHDR(t1);
920 1818 : int len2 = VARSIZE_ANY_EXHDR(t2);
921 :
922 1818 : check_collation_set(collid);
923 :
924 1818 : state->locale = pg_newlocale_from_collation(collid);
925 :
926 : /*
927 : * Most callers need greedy mode, but some might want to unset this to
928 : * optimize.
929 : */
930 1818 : state->greedy = true;
931 :
932 : Assert(len2 > 0);
933 :
934 : /*
935 : * Even with a multi-byte encoding, we perform the search using the raw
936 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
937 : * because in UTF-8 the byte sequence of one character cannot contain
938 : * another character. For other multi-byte encodings, we do the search
939 : * initially as a simple byte search, ignoring multibyte issues, but
940 : * verify afterwards that the match we found is at a character boundary,
941 : * and continue the search if it was a false match.
942 : */
943 1818 : if (pg_database_encoding_max_length() == 1)
944 108 : state->is_multibyte_char_in_char = false;
945 1710 : else if (GetDatabaseEncoding() == PG_UTF8)
946 1710 : state->is_multibyte_char_in_char = false;
947 : else
948 0 : state->is_multibyte_char_in_char = true;
949 :
950 1818 : state->str1 = VARDATA_ANY(t1);
951 1818 : state->str2 = VARDATA_ANY(t2);
952 1818 : state->len1 = len1;
953 1818 : state->len2 = len2;
954 1818 : state->last_match = NULL;
955 1818 : state->refpoint = state->str1;
956 1818 : state->refpos = 0;
957 :
958 : /*
959 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
960 : * notes we use the terminology that the "haystack" is the string to be
961 : * searched (t1) and the "needle" is the pattern being sought (t2).
962 : *
963 : * If the needle is empty or bigger than the haystack then there is no
964 : * point in wasting cycles initializing the table. We also choose not to
965 : * use B-M-H for needles of length 1, since the skip table can't possibly
966 : * save anything in that case.
967 : *
968 : * (With nondeterministic collations, the search is already
969 : * multibyte-aware, so we don't need this.)
970 : */
971 1818 : if (len1 >= len2 && len2 > 1 && state->locale->deterministic)
972 : {
973 1490 : int searchlength = len1 - len2;
974 : int skiptablemask;
975 : int last;
976 : int i;
977 1490 : const char *str2 = state->str2;
978 :
979 : /*
980 : * First we must determine how much of the skip table to use. The
981 : * declaration of TextPositionState allows up to 256 elements, but for
982 : * short search problems we don't really want to have to initialize so
983 : * many elements --- it would take too long in comparison to the
984 : * actual search time. So we choose a useful skip table size based on
985 : * the haystack length minus the needle length. The closer the needle
986 : * length is to the haystack length the less useful skipping becomes.
987 : *
988 : * Note: since we use bit-masking to select table elements, the skip
989 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
990 : */
991 1490 : if (searchlength < 16)
992 114 : skiptablemask = 3;
993 1376 : else if (searchlength < 64)
994 16 : skiptablemask = 7;
995 1360 : else if (searchlength < 128)
996 14 : skiptablemask = 15;
997 1346 : else if (searchlength < 512)
998 256 : skiptablemask = 31;
999 1090 : else if (searchlength < 2048)
1000 804 : skiptablemask = 63;
1001 286 : else if (searchlength < 4096)
1002 192 : skiptablemask = 127;
1003 : else
1004 94 : skiptablemask = 255;
1005 1490 : state->skiptablemask = skiptablemask;
1006 :
1007 : /*
1008 : * Initialize the skip table. We set all elements to the needle
1009 : * length, since this is the correct skip distance for any character
1010 : * not found in the needle.
1011 : */
1012 110586 : for (i = 0; i <= skiptablemask; i++)
1013 109096 : state->skiptable[i] = len2;
1014 :
1015 : /*
1016 : * Now examine the needle. For each character except the last one,
1017 : * set the corresponding table element to the appropriate skip
1018 : * distance. Note that when two characters share the same skip table
1019 : * entry, the one later in the needle must determine the skip
1020 : * distance.
1021 : */
1022 1490 : last = len2 - 1;
1023 :
1024 19872 : for (i = 0; i < last; i++)
1025 18382 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1026 : }
1027 1818 : }
1028 :
1029 : /*
1030 : * Advance to the next match, starting from the end of the previous match
1031 : * (or the beginning of the string, on first call). Returns true if a match
1032 : * is found.
1033 : *
1034 : * Note that this refuses to match an empty-string needle. Most callers
1035 : * will have handled that case specially and we'll never see it here.
1036 : */
1037 : static bool
1038 9224 : text_position_next(TextPositionState *state)
1039 : {
1040 9224 : int needle_len = state->len2;
1041 : char *start_ptr;
1042 : char *matchptr;
1043 :
1044 9224 : if (needle_len <= 0)
1045 0 : return false; /* result for empty pattern */
1046 :
1047 : /* Start from the point right after the previous match. */
1048 9224 : if (state->last_match)
1049 7394 : start_ptr = state->last_match + state->last_match_len;
1050 : else
1051 1830 : start_ptr = state->str1;
1052 :
1053 9224 : retry:
1054 9224 : matchptr = text_position_next_internal(start_ptr, state);
1055 :
1056 9224 : if (!matchptr)
1057 1734 : return false;
1058 :
1059 : /*
1060 : * Found a match for the byte sequence. If this is a multibyte encoding,
1061 : * where one character's byte sequence can appear inside a longer
1062 : * multi-byte character, we need to verify that the match was at a
1063 : * character boundary, not in the middle of a multi-byte character.
1064 : */
1065 7490 : if (state->is_multibyte_char_in_char && state->locale->deterministic)
1066 : {
1067 : /* Walk one character at a time, until we reach the match. */
1068 :
1069 : /* the search should never move backwards. */
1070 : Assert(state->refpoint <= matchptr);
1071 :
1072 0 : while (state->refpoint < matchptr)
1073 : {
1074 : /* step to next character. */
1075 0 : state->refpoint += pg_mblen(state->refpoint);
1076 0 : state->refpos++;
1077 :
1078 : /*
1079 : * If we stepped over the match's start position, then it was a
1080 : * false positive, where the byte sequence appeared in the middle
1081 : * of a multi-byte character. Skip it, and continue the search at
1082 : * the next character boundary.
1083 : */
1084 0 : if (state->refpoint > matchptr)
1085 : {
1086 0 : start_ptr = state->refpoint;
1087 0 : goto retry;
1088 : }
1089 : }
1090 : }
1091 :
1092 7490 : state->last_match = matchptr;
1093 7490 : state->last_match_len = state->last_match_len_tmp;
1094 7490 : return true;
1095 : }
1096 :
1097 : /*
1098 : * Subroutine of text_position_next(). This searches for the raw byte
1099 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1100 : * match starting at 'start_ptr', or NULL if no match is found.
1101 : */
1102 : static char *
1103 9224 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1104 : {
1105 9224 : int haystack_len = state->len1;
1106 9224 : int needle_len = state->len2;
1107 9224 : int skiptablemask = state->skiptablemask;
1108 9224 : const char *haystack = state->str1;
1109 9224 : const char *needle = state->str2;
1110 9224 : const char *haystack_end = &haystack[haystack_len];
1111 : const char *hptr;
1112 :
1113 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1114 :
1115 9224 : state->last_match_len_tmp = needle_len;
1116 :
1117 9224 : if (!state->locale->deterministic)
1118 : {
1119 : /*
1120 : * With a nondeterministic collation, we have to use an unoptimized
1121 : * route. We walk through the haystack and see if at each position
1122 : * there is a substring of the remaining string that is equal to the
1123 : * needle under the given collation.
1124 : *
1125 : * Note, the found substring could have a different length than the
1126 : * needle, including being empty. Callers that want to skip over the
1127 : * found string need to read the length of the found substring from
1128 : * last_match_len rather than just using the length of their needle.
1129 : *
1130 : * Most callers will require "greedy" semantics, meaning that we need
1131 : * to find the longest such substring, not the shortest. For callers
1132 : * that don't need greedy semantics, we can finish on the first match.
1133 : */
1134 240 : const char *result_hptr = NULL;
1135 :
1136 240 : hptr = start_ptr;
1137 642 : while (hptr < haystack_end)
1138 : {
1139 : /*
1140 : * First check the common case that there is a match in the
1141 : * haystack of exactly the length of the needle.
1142 : */
1143 534 : if (!state->greedy &&
1144 108 : haystack_end - hptr >= needle_len &&
1145 54 : pg_strncoll(hptr, needle_len, needle, needle_len, state->locale) == 0)
1146 12 : return (char *) hptr;
1147 :
1148 : /*
1149 : * Else check if any of the possible substrings starting at hptr
1150 : * are equal to the needle.
1151 : */
1152 2586 : for (const char *test_end = hptr; test_end < haystack_end; test_end += pg_mblen(test_end))
1153 : {
1154 2064 : if (pg_strncoll(hptr, (test_end - hptr), needle, needle_len, state->locale) == 0)
1155 : {
1156 132 : state->last_match_len_tmp = (test_end - hptr);
1157 132 : result_hptr = hptr;
1158 132 : if (!state->greedy)
1159 0 : break;
1160 : }
1161 : }
1162 522 : if (result_hptr)
1163 120 : break;
1164 :
1165 402 : hptr += pg_mblen(hptr);
1166 : }
1167 :
1168 228 : return (char *) result_hptr;
1169 : }
1170 8984 : else if (needle_len == 1)
1171 : {
1172 : /* No point in using B-M-H for a one-character needle */
1173 760 : char nchar = *needle;
1174 :
1175 760 : hptr = start_ptr;
1176 5878 : while (hptr < haystack_end)
1177 : {
1178 5712 : if (*hptr == nchar)
1179 594 : return (char *) hptr;
1180 5118 : hptr++;
1181 : }
1182 : }
1183 : else
1184 : {
1185 8224 : const char *needle_last = &needle[needle_len - 1];
1186 :
1187 : /* Start at startpos plus the length of the needle */
1188 8224 : hptr = start_ptr + needle_len - 1;
1189 213838 : while (hptr < haystack_end)
1190 : {
1191 : /* Match the needle scanning *backward* */
1192 : const char *nptr;
1193 : const char *p;
1194 :
1195 212378 : nptr = needle_last;
1196 212378 : p = hptr;
1197 313316 : while (*nptr == *p)
1198 : {
1199 : /* Matched it all? If so, return 1-based position */
1200 107702 : if (nptr == needle)
1201 6764 : return (char *) p;
1202 100938 : nptr--, p--;
1203 : }
1204 :
1205 : /*
1206 : * No match, so use the haystack char at hptr to decide how far to
1207 : * advance. If the needle had any occurrence of that character
1208 : * (or more precisely, one sharing the same skiptable entry)
1209 : * before its last character, then we advance far enough to align
1210 : * the last such needle character with that haystack position.
1211 : * Otherwise we can advance by the whole needle length.
1212 : */
1213 205614 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1214 : }
1215 : }
1216 :
1217 1626 : return 0; /* not found */
1218 : }
1219 :
1220 : /*
1221 : * Return a pointer to the current match.
1222 : *
1223 : * The returned pointer points into the original haystack string.
1224 : */
1225 : static char *
1226 7388 : text_position_get_match_ptr(TextPositionState *state)
1227 : {
1228 7388 : return state->last_match;
1229 : }
1230 :
1231 : /*
1232 : * Return the offset of the current match.
1233 : *
1234 : * The offset is in characters, 1-based.
1235 : */
1236 : static int
1237 72 : text_position_get_match_pos(TextPositionState *state)
1238 : {
1239 : /* Convert the byte position to char position. */
1240 144 : state->refpos += pg_mbstrlen_with_len(state->refpoint,
1241 72 : state->last_match - state->refpoint);
1242 72 : state->refpoint = state->last_match;
1243 72 : return state->refpos + 1;
1244 : }
1245 :
1246 : /*
1247 : * Reset search state to the initial state installed by text_position_setup.
1248 : *
1249 : * The next call to text_position_next will search from the beginning
1250 : * of the string.
1251 : */
1252 : static void
1253 12 : text_position_reset(TextPositionState *state)
1254 : {
1255 12 : state->last_match = NULL;
1256 12 : state->refpoint = state->str1;
1257 12 : state->refpos = 0;
1258 12 : }
1259 :
1260 : static void
1261 1818 : text_position_cleanup(TextPositionState *state)
1262 : {
1263 : /* no cleanup needed */
1264 1818 : }
1265 :
1266 :
1267 : static void
1268 17082318 : check_collation_set(Oid collid)
1269 : {
1270 17082318 : if (!OidIsValid(collid))
1271 : {
1272 : /*
1273 : * This typically means that the parser could not resolve a conflict
1274 : * of implicit collations, so report it that way.
1275 : */
1276 30 : ereport(ERROR,
1277 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1278 : errmsg("could not determine which collation to use for string comparison"),
1279 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1280 : }
1281 17082288 : }
1282 :
1283 : /*
1284 : * varstr_cmp()
1285 : *
1286 : * Comparison function for text strings with given lengths, using the
1287 : * appropriate locale. Returns an integer less than, equal to, or greater than
1288 : * zero, indicating whether arg1 is less than, equal to, or greater than arg2.
1289 : *
1290 : * Note: many functions that depend on this are marked leakproof; therefore,
1291 : * avoid reporting the actual contents of the input when throwing errors.
1292 : * All errors herein should be things that can't happen except on corrupt
1293 : * data, anyway; otherwise we will have trouble with indexing strings that
1294 : * would cause them.
1295 : */
1296 : int
1297 10033796 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1298 : {
1299 : int result;
1300 : pg_locale_t mylocale;
1301 :
1302 10033796 : check_collation_set(collid);
1303 :
1304 10033778 : mylocale = pg_newlocale_from_collation(collid);
1305 :
1306 10033778 : if (mylocale->collate_is_c)
1307 : {
1308 4031714 : result = memcmp(arg1, arg2, Min(len1, len2));
1309 4031714 : if ((result == 0) && (len1 != len2))
1310 137516 : result = (len1 < len2) ? -1 : 1;
1311 : }
1312 : else
1313 : {
1314 : /*
1315 : * memcmp() can't tell us which of two unequal strings sorts first,
1316 : * but it's a cheap way to tell if they're equal. Testing shows that
1317 : * memcmp() followed by strcoll() is only trivially slower than
1318 : * strcoll() by itself, so we don't lose much if this doesn't work out
1319 : * very often, and if it does - for example, because there are many
1320 : * equal strings in the input - then we win big by avoiding expensive
1321 : * collation-aware comparisons.
1322 : */
1323 6002064 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1324 1557532 : return 0;
1325 :
1326 4444532 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1327 :
1328 : /* Break tie if necessary. */
1329 4444532 : if (result == 0 && mylocale->deterministic)
1330 : {
1331 0 : result = memcmp(arg1, arg2, Min(len1, len2));
1332 0 : if ((result == 0) && (len1 != len2))
1333 0 : result = (len1 < len2) ? -1 : 1;
1334 : }
1335 : }
1336 :
1337 8476246 : return result;
1338 : }
1339 :
1340 : /* text_cmp()
1341 : * Internal comparison function for text strings.
1342 : * Returns -1, 0 or 1
1343 : */
1344 : static int
1345 7745178 : text_cmp(text *arg1, text *arg2, Oid collid)
1346 : {
1347 : char *a1p,
1348 : *a2p;
1349 : int len1,
1350 : len2;
1351 :
1352 7745178 : a1p = VARDATA_ANY(arg1);
1353 7745178 : a2p = VARDATA_ANY(arg2);
1354 :
1355 7745178 : len1 = VARSIZE_ANY_EXHDR(arg1);
1356 7745178 : len2 = VARSIZE_ANY_EXHDR(arg2);
1357 :
1358 7745178 : return varstr_cmp(a1p, len1, a2p, len2, collid);
1359 : }
1360 :
1361 : /*
1362 : * Comparison functions for text strings.
1363 : *
1364 : * Note: btree indexes need these routines not to leak memory; therefore,
1365 : * be careful to free working copies of toasted datums. Most places don't
1366 : * need to be so careful.
1367 : */
1368 :
1369 : Datum
1370 6640308 : texteq(PG_FUNCTION_ARGS)
1371 : {
1372 6640308 : Oid collid = PG_GET_COLLATION();
1373 6640308 : pg_locale_t mylocale = 0;
1374 : bool result;
1375 :
1376 6640308 : check_collation_set(collid);
1377 :
1378 6640308 : mylocale = pg_newlocale_from_collation(collid);
1379 :
1380 6640308 : if (mylocale->deterministic)
1381 : {
1382 6631868 : Datum arg1 = PG_GETARG_DATUM(0);
1383 6631868 : Datum arg2 = PG_GETARG_DATUM(1);
1384 : Size len1,
1385 : len2;
1386 :
1387 : /*
1388 : * Since we only care about equality or not-equality, we can avoid all
1389 : * the expense of strcoll() here, and just do bitwise comparison. In
1390 : * fact, we don't even have to do a bitwise comparison if we can show
1391 : * the lengths of the strings are unequal; which might save us from
1392 : * having to detoast one or both values.
1393 : */
1394 6631868 : len1 = toast_raw_datum_size(arg1);
1395 6631868 : len2 = toast_raw_datum_size(arg2);
1396 6631868 : if (len1 != len2)
1397 3155128 : result = false;
1398 : else
1399 : {
1400 3476740 : text *targ1 = DatumGetTextPP(arg1);
1401 3476740 : text *targ2 = DatumGetTextPP(arg2);
1402 :
1403 3476740 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1404 : len1 - VARHDRSZ) == 0);
1405 :
1406 3476740 : PG_FREE_IF_COPY(targ1, 0);
1407 3476740 : PG_FREE_IF_COPY(targ2, 1);
1408 : }
1409 : }
1410 : else
1411 : {
1412 8440 : text *arg1 = PG_GETARG_TEXT_PP(0);
1413 8440 : text *arg2 = PG_GETARG_TEXT_PP(1);
1414 :
1415 8440 : result = (text_cmp(arg1, arg2, collid) == 0);
1416 :
1417 8440 : PG_FREE_IF_COPY(arg1, 0);
1418 8440 : PG_FREE_IF_COPY(arg2, 1);
1419 : }
1420 :
1421 6640308 : PG_RETURN_BOOL(result);
1422 : }
1423 :
1424 : Datum
1425 22798 : textne(PG_FUNCTION_ARGS)
1426 : {
1427 22798 : Oid collid = PG_GET_COLLATION();
1428 : pg_locale_t mylocale;
1429 : bool result;
1430 :
1431 22798 : check_collation_set(collid);
1432 :
1433 22798 : mylocale = pg_newlocale_from_collation(collid);
1434 :
1435 22798 : if (mylocale->deterministic)
1436 : {
1437 22774 : Datum arg1 = PG_GETARG_DATUM(0);
1438 22774 : Datum arg2 = PG_GETARG_DATUM(1);
1439 : Size len1,
1440 : len2;
1441 :
1442 : /* See comment in texteq() */
1443 22774 : len1 = toast_raw_datum_size(arg1);
1444 22774 : len2 = toast_raw_datum_size(arg2);
1445 22774 : if (len1 != len2)
1446 4448 : result = true;
1447 : else
1448 : {
1449 18326 : text *targ1 = DatumGetTextPP(arg1);
1450 18326 : text *targ2 = DatumGetTextPP(arg2);
1451 :
1452 18326 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1453 : len1 - VARHDRSZ) != 0);
1454 :
1455 18326 : PG_FREE_IF_COPY(targ1, 0);
1456 18326 : PG_FREE_IF_COPY(targ2, 1);
1457 : }
1458 : }
1459 : else
1460 : {
1461 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
1462 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
1463 :
1464 24 : result = (text_cmp(arg1, arg2, collid) != 0);
1465 :
1466 24 : PG_FREE_IF_COPY(arg1, 0);
1467 24 : PG_FREE_IF_COPY(arg2, 1);
1468 : }
1469 :
1470 22798 : PG_RETURN_BOOL(result);
1471 : }
1472 :
1473 : Datum
1474 211430 : text_lt(PG_FUNCTION_ARGS)
1475 : {
1476 211430 : text *arg1 = PG_GETARG_TEXT_PP(0);
1477 211430 : text *arg2 = PG_GETARG_TEXT_PP(1);
1478 : bool result;
1479 :
1480 211430 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1481 :
1482 211412 : PG_FREE_IF_COPY(arg1, 0);
1483 211412 : PG_FREE_IF_COPY(arg2, 1);
1484 :
1485 211412 : PG_RETURN_BOOL(result);
1486 : }
1487 :
1488 : Datum
1489 318602 : text_le(PG_FUNCTION_ARGS)
1490 : {
1491 318602 : text *arg1 = PG_GETARG_TEXT_PP(0);
1492 318602 : text *arg2 = PG_GETARG_TEXT_PP(1);
1493 : bool result;
1494 :
1495 318602 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1496 :
1497 318602 : PG_FREE_IF_COPY(arg1, 0);
1498 318602 : PG_FREE_IF_COPY(arg2, 1);
1499 :
1500 318602 : PG_RETURN_BOOL(result);
1501 : }
1502 :
1503 : Datum
1504 195908 : text_gt(PG_FUNCTION_ARGS)
1505 : {
1506 195908 : text *arg1 = PG_GETARG_TEXT_PP(0);
1507 195908 : text *arg2 = PG_GETARG_TEXT_PP(1);
1508 : bool result;
1509 :
1510 195908 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1511 :
1512 195908 : PG_FREE_IF_COPY(arg1, 0);
1513 195908 : PG_FREE_IF_COPY(arg2, 1);
1514 :
1515 195908 : PG_RETURN_BOOL(result);
1516 : }
1517 :
1518 : Datum
1519 175720 : text_ge(PG_FUNCTION_ARGS)
1520 : {
1521 175720 : text *arg1 = PG_GETARG_TEXT_PP(0);
1522 175720 : text *arg2 = PG_GETARG_TEXT_PP(1);
1523 : bool result;
1524 :
1525 175720 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1526 :
1527 175720 : PG_FREE_IF_COPY(arg1, 0);
1528 175720 : PG_FREE_IF_COPY(arg2, 1);
1529 :
1530 175720 : PG_RETURN_BOOL(result);
1531 : }
1532 :
1533 : Datum
1534 37914 : text_starts_with(PG_FUNCTION_ARGS)
1535 : {
1536 37914 : Datum arg1 = PG_GETARG_DATUM(0);
1537 37914 : Datum arg2 = PG_GETARG_DATUM(1);
1538 37914 : Oid collid = PG_GET_COLLATION();
1539 : pg_locale_t mylocale;
1540 : bool result;
1541 : Size len1,
1542 : len2;
1543 :
1544 37914 : check_collation_set(collid);
1545 :
1546 37914 : mylocale = pg_newlocale_from_collation(collid);
1547 :
1548 37914 : if (!mylocale->deterministic)
1549 0 : ereport(ERROR,
1550 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1551 : errmsg("nondeterministic collations are not supported for substring searches")));
1552 :
1553 37914 : len1 = toast_raw_datum_size(arg1);
1554 37914 : len2 = toast_raw_datum_size(arg2);
1555 37914 : if (len2 > len1)
1556 0 : result = false;
1557 : else
1558 : {
1559 37914 : text *targ1 = text_substring(arg1, 1, len2, false);
1560 37914 : text *targ2 = DatumGetTextPP(arg2);
1561 :
1562 37914 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1563 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1564 :
1565 37914 : PG_FREE_IF_COPY(targ1, 0);
1566 37914 : PG_FREE_IF_COPY(targ2, 1);
1567 : }
1568 :
1569 37914 : PG_RETURN_BOOL(result);
1570 : }
1571 :
1572 : Datum
1573 6519418 : bttextcmp(PG_FUNCTION_ARGS)
1574 : {
1575 6519418 : text *arg1 = PG_GETARG_TEXT_PP(0);
1576 6519418 : text *arg2 = PG_GETARG_TEXT_PP(1);
1577 : int32 result;
1578 :
1579 6519418 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1580 :
1581 6519418 : PG_FREE_IF_COPY(arg1, 0);
1582 6519418 : PG_FREE_IF_COPY(arg2, 1);
1583 :
1584 6519418 : PG_RETURN_INT32(result);
1585 : }
1586 :
1587 : Datum
1588 87316 : bttextsortsupport(PG_FUNCTION_ARGS)
1589 : {
1590 87316 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1591 87316 : Oid collid = ssup->ssup_collation;
1592 : MemoryContext oldcontext;
1593 :
1594 87316 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1595 :
1596 : /* Use generic string SortSupport */
1597 87316 : varstr_sortsupport(ssup, TEXTOID, collid);
1598 :
1599 87304 : MemoryContextSwitchTo(oldcontext);
1600 :
1601 87304 : PG_RETURN_VOID();
1602 : }
1603 :
1604 : /*
1605 : * Generic sortsupport interface for character type's operator classes.
1606 : * Includes locale support, and support for BpChar semantics (i.e. removing
1607 : * trailing spaces before comparison).
1608 : *
1609 : * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
1610 : * same representation. Callers that always use the C collation (e.g.
1611 : * non-collatable type callers like bytea) may have NUL bytes in their strings;
1612 : * this will not work with any other collation, though.
1613 : */
1614 : void
1615 137560 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1616 : {
1617 137560 : bool abbreviate = ssup->abbreviate;
1618 137560 : bool collate_c = false;
1619 : VarStringSortSupport *sss;
1620 : pg_locale_t locale;
1621 :
1622 137560 : check_collation_set(collid);
1623 :
1624 137548 : locale = pg_newlocale_from_collation(collid);
1625 :
1626 : /*
1627 : * If possible, set ssup->comparator to a function which can be used to
1628 : * directly compare two datums. If we can do this, we'll avoid the
1629 : * overhead of a trip through the fmgr layer for every comparison, which
1630 : * can be substantial.
1631 : *
1632 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1633 : * which uses strcoll() to perform comparisons. We use that for the
1634 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1635 : * LC_COLLATE = C, we can make things quite a bit faster with
1636 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1637 : * memcmp() rather than strcoll().
1638 : */
1639 137548 : if (locale->collate_is_c)
1640 : {
1641 91330 : if (typid == BPCHAROID)
1642 308 : ssup->comparator = bpcharfastcmp_c;
1643 91022 : else if (typid == NAMEOID)
1644 : {
1645 49178 : ssup->comparator = namefastcmp_c;
1646 : /* Not supporting abbreviation with type NAME, for now */
1647 49178 : abbreviate = false;
1648 : }
1649 : else
1650 41844 : ssup->comparator = varstrfastcmp_c;
1651 :
1652 91330 : collate_c = true;
1653 : }
1654 : else
1655 : {
1656 : /*
1657 : * We use varlenafastcmp_locale except for type NAME.
1658 : */
1659 46218 : if (typid == NAMEOID)
1660 : {
1661 0 : ssup->comparator = namefastcmp_locale;
1662 : /* Not supporting abbreviation with type NAME, for now */
1663 0 : abbreviate = false;
1664 : }
1665 : else
1666 46218 : ssup->comparator = varlenafastcmp_locale;
1667 :
1668 : /*
1669 : * Unfortunately, it seems that abbreviation for non-C collations is
1670 : * broken on many common platforms; see pg_strxfrm_enabled().
1671 : *
1672 : * Even apart from the risk of broken locales, it's possible that
1673 : * there are platforms where the use of abbreviated keys should be
1674 : * disabled at compile time. For example, macOS's strxfrm()
1675 : * implementation is known to not effectively concentrate a
1676 : * significant amount of entropy from the original string in earlier
1677 : * transformed blobs. It's possible that other supported platforms
1678 : * are similarly encumbered. So, if we ever get past disabling this
1679 : * categorically, we may still want or need to disable it for
1680 : * particular platforms.
1681 : */
1682 46218 : if (!pg_strxfrm_enabled(locale))
1683 45422 : abbreviate = false;
1684 : }
1685 :
1686 : /*
1687 : * If we're using abbreviated keys, or if we're using a locale-aware
1688 : * comparison, we need to initialize a VarStringSortSupport object. Both
1689 : * cases will make use of the temporary buffers we initialize here for
1690 : * scratch space (and to detect requirement for BpChar semantics from
1691 : * caller), and the abbreviation case requires additional state.
1692 : */
1693 137548 : if (abbreviate || !collate_c)
1694 : {
1695 69952 : sss = palloc(sizeof(VarStringSortSupport));
1696 69952 : sss->buf1 = palloc(TEXTBUFLEN);
1697 69952 : sss->buflen1 = TEXTBUFLEN;
1698 69952 : sss->buf2 = palloc(TEXTBUFLEN);
1699 69952 : sss->buflen2 = TEXTBUFLEN;
1700 : /* Start with invalid values */
1701 69952 : sss->last_len1 = -1;
1702 69952 : sss->last_len2 = -1;
1703 : /* Initialize */
1704 69952 : sss->last_returned = 0;
1705 69952 : if (collate_c)
1706 23734 : sss->locale = NULL;
1707 : else
1708 46218 : sss->locale = locale;
1709 :
1710 : /*
1711 : * To avoid somehow confusing a strxfrm() blob and an original string,
1712 : * constantly keep track of the variety of data that buf1 and buf2
1713 : * currently contain.
1714 : *
1715 : * Comparisons may be interleaved with conversion calls. Frequently,
1716 : * conversions and comparisons are batched into two distinct phases,
1717 : * but the correctness of caching cannot hinge upon this. For
1718 : * comparison caching, buffer state is only trusted if cache_blob is
1719 : * found set to false, whereas strxfrm() caching only trusts the state
1720 : * when cache_blob is found set to true.
1721 : *
1722 : * Arbitrarily initialize cache_blob to true.
1723 : */
1724 69952 : sss->cache_blob = true;
1725 69952 : sss->collate_c = collate_c;
1726 69952 : sss->typid = typid;
1727 69952 : ssup->ssup_extra = sss;
1728 :
1729 : /*
1730 : * If possible, plan to use the abbreviated keys optimization. The
1731 : * core code may switch back to authoritative comparator should
1732 : * abbreviation be aborted.
1733 : */
1734 69952 : if (abbreviate)
1735 : {
1736 24332 : sss->prop_card = 0.20;
1737 24332 : initHyperLogLog(&sss->abbr_card, 10);
1738 24332 : initHyperLogLog(&sss->full_card, 10);
1739 24332 : ssup->abbrev_full_comparator = ssup->comparator;
1740 24332 : ssup->comparator = ssup_datum_unsigned_cmp;
1741 24332 : ssup->abbrev_converter = varstr_abbrev_convert;
1742 24332 : ssup->abbrev_abort = varstr_abbrev_abort;
1743 : }
1744 : }
1745 137548 : }
1746 :
1747 : /*
1748 : * sortsupport comparison func (for C locale case)
1749 : */
1750 : static int
1751 44172558 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
1752 : {
1753 44172558 : VarString *arg1 = DatumGetVarStringPP(x);
1754 44172558 : VarString *arg2 = DatumGetVarStringPP(y);
1755 : char *a1p,
1756 : *a2p;
1757 : int len1,
1758 : len2,
1759 : result;
1760 :
1761 44172558 : a1p = VARDATA_ANY(arg1);
1762 44172558 : a2p = VARDATA_ANY(arg2);
1763 :
1764 44172558 : len1 = VARSIZE_ANY_EXHDR(arg1);
1765 44172558 : len2 = VARSIZE_ANY_EXHDR(arg2);
1766 :
1767 44172558 : result = memcmp(a1p, a2p, Min(len1, len2));
1768 44172558 : if ((result == 0) && (len1 != len2))
1769 1182682 : result = (len1 < len2) ? -1 : 1;
1770 :
1771 : /* We can't afford to leak memory here. */
1772 44172558 : if (PointerGetDatum(arg1) != x)
1773 0 : pfree(arg1);
1774 44172558 : if (PointerGetDatum(arg2) != y)
1775 0 : pfree(arg2);
1776 :
1777 44172558 : return result;
1778 : }
1779 :
1780 : /*
1781 : * sortsupport comparison func (for BpChar C locale case)
1782 : *
1783 : * BpChar outsources its sortsupport to this module. Specialization for the
1784 : * varstr_sortsupport BpChar case, modeled on
1785 : * internal_bpchar_pattern_compare().
1786 : */
1787 : static int
1788 62420 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
1789 : {
1790 62420 : BpChar *arg1 = DatumGetBpCharPP(x);
1791 62420 : BpChar *arg2 = DatumGetBpCharPP(y);
1792 : char *a1p,
1793 : *a2p;
1794 : int len1,
1795 : len2,
1796 : result;
1797 :
1798 62420 : a1p = VARDATA_ANY(arg1);
1799 62420 : a2p = VARDATA_ANY(arg2);
1800 :
1801 62420 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
1802 62420 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
1803 :
1804 62420 : result = memcmp(a1p, a2p, Min(len1, len2));
1805 62420 : if ((result == 0) && (len1 != len2))
1806 4 : result = (len1 < len2) ? -1 : 1;
1807 :
1808 : /* We can't afford to leak memory here. */
1809 62420 : if (PointerGetDatum(arg1) != x)
1810 0 : pfree(arg1);
1811 62420 : if (PointerGetDatum(arg2) != y)
1812 0 : pfree(arg2);
1813 :
1814 62420 : return result;
1815 : }
1816 :
1817 : /*
1818 : * sortsupport comparison func (for NAME C locale case)
1819 : */
1820 : static int
1821 39683466 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
1822 : {
1823 39683466 : Name arg1 = DatumGetName(x);
1824 39683466 : Name arg2 = DatumGetName(y);
1825 :
1826 39683466 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
1827 : }
1828 :
1829 : /*
1830 : * sortsupport comparison func (for locale case with all varlena types)
1831 : */
1832 : static int
1833 35932382 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
1834 : {
1835 35932382 : VarString *arg1 = DatumGetVarStringPP(x);
1836 35932382 : VarString *arg2 = DatumGetVarStringPP(y);
1837 : char *a1p,
1838 : *a2p;
1839 : int len1,
1840 : len2,
1841 : result;
1842 :
1843 35932382 : a1p = VARDATA_ANY(arg1);
1844 35932382 : a2p = VARDATA_ANY(arg2);
1845 :
1846 35932382 : len1 = VARSIZE_ANY_EXHDR(arg1);
1847 35932382 : len2 = VARSIZE_ANY_EXHDR(arg2);
1848 :
1849 35932382 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
1850 :
1851 : /* We can't afford to leak memory here. */
1852 35932382 : if (PointerGetDatum(arg1) != x)
1853 0 : pfree(arg1);
1854 35932382 : if (PointerGetDatum(arg2) != y)
1855 0 : pfree(arg2);
1856 :
1857 35932382 : return result;
1858 : }
1859 :
1860 : /*
1861 : * sortsupport comparison func (for locale case with NAME type)
1862 : */
1863 : static int
1864 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
1865 : {
1866 0 : Name arg1 = DatumGetName(x);
1867 0 : Name arg2 = DatumGetName(y);
1868 :
1869 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
1870 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
1871 : ssup);
1872 : }
1873 :
1874 : /*
1875 : * sortsupport comparison func for locale cases
1876 : */
1877 : static int
1878 35932382 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
1879 : {
1880 35932382 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1881 : int result;
1882 : bool arg1_match;
1883 :
1884 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
1885 35932382 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
1886 : {
1887 : /*
1888 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
1889 : * last_len2. Existing contents of buffers might still be used by
1890 : * next call.
1891 : *
1892 : * It's fine to allow the comparison of BpChar padding bytes here,
1893 : * even though that implies that the memcmp() will usually be
1894 : * performed for BpChar callers (though multibyte characters could
1895 : * still prevent that from occurring). The memcmp() is still very
1896 : * cheap, and BpChar's funny semantics have us remove trailing spaces
1897 : * (not limited to padding), so we need make no distinction between
1898 : * padding space characters and "real" space characters.
1899 : */
1900 9223622 : return 0;
1901 : }
1902 :
1903 26708760 : if (sss->typid == BPCHAROID)
1904 : {
1905 : /* Get true number of bytes, ignoring trailing spaces */
1906 34536 : len1 = bpchartruelen(a1p, len1);
1907 34536 : len2 = bpchartruelen(a2p, len2);
1908 : }
1909 :
1910 26708760 : if (len1 >= sss->buflen1)
1911 : {
1912 10 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
1913 10 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
1914 : }
1915 26708760 : if (len2 >= sss->buflen2)
1916 : {
1917 6 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
1918 6 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
1919 : }
1920 :
1921 : /*
1922 : * We're likely to be asked to compare the same strings repeatedly, and
1923 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
1924 : * comparisons, even though in general there is no reason to think that
1925 : * that will work out (every string datum may be unique). Caching does
1926 : * not slow things down measurably when it doesn't work out, and can speed
1927 : * things up by rather a lot when it does. In part, this is because the
1928 : * memcmp() compares data from cachelines that are needed in L1 cache even
1929 : * when the last comparison's result cannot be reused.
1930 : */
1931 26708760 : arg1_match = true;
1932 26708760 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
1933 : {
1934 24705656 : arg1_match = false;
1935 24705656 : memcpy(sss->buf1, a1p, len1);
1936 24705656 : sss->buf1[len1] = '\0';
1937 24705656 : sss->last_len1 = len1;
1938 : }
1939 :
1940 : /*
1941 : * If we're comparing the same two strings as last time, we can return the
1942 : * same answer without calling strcoll() again. This is more likely than
1943 : * it seems (at least with moderate to low cardinality sets), because
1944 : * quicksort compares the same pivot against many values.
1945 : */
1946 26708760 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
1947 : {
1948 4049018 : memcpy(sss->buf2, a2p, len2);
1949 4049018 : sss->buf2[len2] = '\0';
1950 4049018 : sss->last_len2 = len2;
1951 : }
1952 22659742 : else if (arg1_match && !sss->cache_blob)
1953 : {
1954 : /* Use result cached following last actual strcoll() call */
1955 1576324 : return sss->last_returned;
1956 : }
1957 :
1958 25132436 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
1959 :
1960 : /* Break tie if necessary. */
1961 25132436 : if (result == 0 && sss->locale->deterministic)
1962 0 : result = strcmp(sss->buf1, sss->buf2);
1963 :
1964 : /* Cache result, perhaps saving an expensive strcoll() call next time */
1965 25132436 : sss->cache_blob = false;
1966 25132436 : sss->last_returned = result;
1967 25132436 : return result;
1968 : }
1969 :
1970 : /*
1971 : * Conversion routine for sortsupport. Converts original to abbreviated key
1972 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
1973 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
1974 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
1975 : * locale is used, or in case of bytea, just memcpy() from original instead.
1976 : */
1977 : static Datum
1978 833022 : varstr_abbrev_convert(Datum original, SortSupport ssup)
1979 : {
1980 833022 : const size_t max_prefix_bytes = sizeof(Datum);
1981 833022 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1982 833022 : VarString *authoritative = DatumGetVarStringPP(original);
1983 833022 : char *authoritative_data = VARDATA_ANY(authoritative);
1984 :
1985 : /* working state */
1986 : Datum res;
1987 : char *pres;
1988 : int len;
1989 : uint32 hash;
1990 :
1991 833022 : pres = (char *) &res;
1992 : /* memset(), so any non-overwritten bytes are NUL */
1993 833022 : memset(pres, 0, max_prefix_bytes);
1994 833022 : len = VARSIZE_ANY_EXHDR(authoritative);
1995 :
1996 : /* Get number of bytes, ignoring trailing spaces */
1997 833022 : if (sss->typid == BPCHAROID)
1998 1010 : len = bpchartruelen(authoritative_data, len);
1999 :
2000 : /*
2001 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2002 : * abbreviate keys. The full comparator for the C locale is always
2003 : * memcmp(). It would be incorrect to allow bytea callers (callers that
2004 : * always force the C collation -- bytea isn't a collatable type, but this
2005 : * approach is convenient) to use strxfrm(). This is because bytea
2006 : * strings may contain NUL bytes. Besides, this should be faster, too.
2007 : *
2008 : * More generally, it's okay that bytea callers can have NUL bytes in
2009 : * strings because abbreviated cmp need not make a distinction between
2010 : * terminating NUL bytes, and NUL bytes representing actual NULs in the
2011 : * authoritative representation. Hopefully a comparison at or past one
2012 : * abbreviated key's terminating NUL byte will resolve the comparison
2013 : * without consulting the authoritative representation; specifically, some
2014 : * later non-NUL byte in the longer string can resolve the comparison
2015 : * against a subsequent terminating NUL in the shorter string. There will
2016 : * usually be what is effectively a "length-wise" resolution there and
2017 : * then.
2018 : *
2019 : * If that doesn't work out -- if all bytes in the longer string
2020 : * positioned at or past the offset of the smaller string's (first)
2021 : * terminating NUL are actually representative of NUL bytes in the
2022 : * authoritative binary string (perhaps with some *terminating* NUL bytes
2023 : * towards the end of the longer string iff it happens to still be small)
2024 : * -- then an authoritative tie-breaker will happen, and do the right
2025 : * thing: explicitly consider string length.
2026 : */
2027 833022 : if (sss->collate_c)
2028 831186 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
2029 : else
2030 : {
2031 : Size bsize;
2032 :
2033 : /*
2034 : * We're not using the C collation, so fall back on strxfrm or ICU
2035 : * analogs.
2036 : */
2037 :
2038 : /* By convention, we use buffer 1 to store and NUL-terminate */
2039 1836 : if (len >= sss->buflen1)
2040 : {
2041 0 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2042 0 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2043 : }
2044 :
2045 : /* Might be able to reuse strxfrm() blob from last call */
2046 1836 : if (sss->last_len1 == len && sss->cache_blob &&
2047 918 : memcmp(sss->buf1, authoritative_data, len) == 0)
2048 : {
2049 168 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, sss->last_len2));
2050 : /* No change affecting cardinality, so no hashing required */
2051 168 : goto done;
2052 : }
2053 :
2054 1668 : memcpy(sss->buf1, authoritative_data, len);
2055 :
2056 : /*
2057 : * pg_strxfrm() and pg_strxfrm_prefix expect NUL-terminated strings.
2058 : */
2059 1668 : sss->buf1[len] = '\0';
2060 1668 : sss->last_len1 = len;
2061 :
2062 1668 : if (pg_strxfrm_prefix_enabled(sss->locale))
2063 : {
2064 1668 : if (sss->buflen2 < max_prefix_bytes)
2065 : {
2066 0 : sss->buflen2 = Max(max_prefix_bytes,
2067 : Min(sss->buflen2 * 2, MaxAllocSize));
2068 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2069 : }
2070 :
2071 1668 : bsize = pg_strxfrm_prefix(sss->buf2, sss->buf1,
2072 : max_prefix_bytes, sss->locale);
2073 1668 : sss->last_len2 = bsize;
2074 : }
2075 : else
2076 : {
2077 : /*
2078 : * Loop: Call pg_strxfrm(), possibly enlarge buffer, and try
2079 : * again. The pg_strxfrm() function leaves the result buffer
2080 : * content undefined if the result did not fit, so we need to
2081 : * retry until everything fits, even though we only need the first
2082 : * few bytes in the end.
2083 : */
2084 : for (;;)
2085 : {
2086 0 : bsize = pg_strxfrm(sss->buf2, sss->buf1, sss->buflen2,
2087 : sss->locale);
2088 :
2089 0 : sss->last_len2 = bsize;
2090 0 : if (bsize < sss->buflen2)
2091 0 : break;
2092 :
2093 : /*
2094 : * Grow buffer and retry.
2095 : */
2096 0 : sss->buflen2 = Max(bsize + 1,
2097 : Min(sss->buflen2 * 2, MaxAllocSize));
2098 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2099 : }
2100 : }
2101 :
2102 : /*
2103 : * Every Datum byte is always compared. This is safe because the
2104 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2105 : * misinterpreting any NUL bytes not intended to be interpreted as
2106 : * logically representing termination.
2107 : *
2108 : * (Actually, even if there were NUL bytes in the blob it would be
2109 : * okay. See remarks on bytea case above.)
2110 : */
2111 1668 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, bsize));
2112 : }
2113 :
2114 : /*
2115 : * Maintain approximate cardinality of both abbreviated keys and original,
2116 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2117 : * the worst case, where we do many string transformations for no saving
2118 : * in full strcoll()-based comparisons. These statistics are used by
2119 : * varstr_abbrev_abort().
2120 : *
2121 : * First, Hash key proper, or a significant fraction of it. Mix in length
2122 : * in order to compensate for cases where differences are past
2123 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2124 : */
2125 832854 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2126 : Min(len, PG_CACHE_LINE_SIZE)));
2127 :
2128 832854 : if (len > PG_CACHE_LINE_SIZE)
2129 190 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2130 :
2131 832854 : addHyperLogLog(&sss->full_card, hash);
2132 :
2133 : /* Hash abbreviated key */
2134 : {
2135 : uint32 tmp;
2136 :
2137 832854 : tmp = DatumGetUInt32(res) ^ (uint32) (DatumGetUInt64(res) >> 32);
2138 832854 : hash = DatumGetUInt32(hash_uint32(tmp));
2139 : }
2140 :
2141 832854 : addHyperLogLog(&sss->abbr_card, hash);
2142 :
2143 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2144 832854 : sss->cache_blob = true;
2145 833022 : done:
2146 :
2147 : /*
2148 : * Byteswap on little-endian machines.
2149 : *
2150 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2151 : * 3-way comparator) works correctly on all platforms. If we didn't do
2152 : * this, the comparator would have to call memcmp() with a pair of
2153 : * pointers to the first byte of each abbreviated key, which is slower.
2154 : */
2155 833022 : res = DatumBigEndianToNative(res);
2156 :
2157 : /* Don't leak memory here */
2158 833022 : if (PointerGetDatum(authoritative) != original)
2159 2 : pfree(authoritative);
2160 :
2161 833022 : return res;
2162 : }
2163 :
2164 : /*
2165 : * Callback for estimating effectiveness of abbreviated key optimization, using
2166 : * heuristic rules. Returns value indicating if the abbreviation optimization
2167 : * should be aborted, based on its projected effectiveness.
2168 : */
2169 : static bool
2170 2262 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2171 : {
2172 2262 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2173 : double abbrev_distinct,
2174 : key_distinct;
2175 :
2176 : Assert(ssup->abbreviate);
2177 :
2178 : /* Have a little patience */
2179 2262 : if (memtupcount < 100)
2180 1282 : return false;
2181 :
2182 980 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2183 980 : key_distinct = estimateHyperLogLog(&sss->full_card);
2184 :
2185 : /*
2186 : * Clamp cardinality estimates to at least one distinct value. While
2187 : * NULLs are generally disregarded, if only NULL values were seen so far,
2188 : * that might misrepresent costs if we failed to clamp.
2189 : */
2190 980 : if (abbrev_distinct <= 1.0)
2191 0 : abbrev_distinct = 1.0;
2192 :
2193 980 : if (key_distinct <= 1.0)
2194 0 : key_distinct = 1.0;
2195 :
2196 : /*
2197 : * In the worst case all abbreviated keys are identical, while at the same
2198 : * time there are differences within full key strings not captured in
2199 : * abbreviations.
2200 : */
2201 980 : if (trace_sort)
2202 : {
2203 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2204 :
2205 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2206 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2207 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2208 : sss->prop_card);
2209 : }
2210 :
2211 : /*
2212 : * If the number of distinct abbreviated keys approximately matches the
2213 : * number of distinct authoritative original keys, that's reason enough to
2214 : * proceed. We can win even with a very low cardinality set if most
2215 : * tie-breakers only memcmp(). This is by far the most important
2216 : * consideration.
2217 : *
2218 : * While comparisons that are resolved at the abbreviated key level are
2219 : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2220 : * those two outcomes are so much cheaper than a full strcoll() once
2221 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2222 : * cardinality against the overall size of the set in order to more
2223 : * accurately model costs. Assume that an abbreviated comparison, and an
2224 : * abbreviated comparison with a cheap memcmp()-based authoritative
2225 : * resolution are equivalent.
2226 : */
2227 980 : if (abbrev_distinct > key_distinct * sss->prop_card)
2228 : {
2229 : /*
2230 : * When we have exceeded 10,000 tuples, decay required cardinality
2231 : * aggressively for next call.
2232 : *
2233 : * This is useful because the number of comparisons required on
2234 : * average increases at a linearithmic rate, and at roughly 10,000
2235 : * tuples that factor will start to dominate over the linear costs of
2236 : * string transformation (this is a conservative estimate). The decay
2237 : * rate is chosen to be a little less aggressive than halving -- which
2238 : * (since we're called at points at which memtupcount has doubled)
2239 : * would never see the cost model actually abort past the first call
2240 : * following a decay. This decay rate is mostly a precaution against
2241 : * a sudden, violent swing in how well abbreviated cardinality tracks
2242 : * full key cardinality. The decay also serves to prevent a marginal
2243 : * case from being aborted too late, when too much has already been
2244 : * invested in string transformation.
2245 : *
2246 : * It's possible for sets of several million distinct strings with
2247 : * mere tens of thousands of distinct abbreviated keys to still
2248 : * benefit very significantly. This will generally occur provided
2249 : * each abbreviated key is a proxy for a roughly uniform number of the
2250 : * set's full keys. If it isn't so, we hope to catch that early and
2251 : * abort. If it isn't caught early, by the time the problem is
2252 : * apparent it's probably not worth aborting.
2253 : */
2254 980 : if (memtupcount > 10000)
2255 4 : sss->prop_card *= 0.65;
2256 :
2257 980 : return false;
2258 : }
2259 :
2260 : /*
2261 : * Abort abbreviation strategy.
2262 : *
2263 : * The worst case, where all abbreviated keys are identical while all
2264 : * original strings differ will typically only see a regression of about
2265 : * 10% in execution time for small to medium sized lists of strings.
2266 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2267 : * often expect very large improvements, particularly with sets of strings
2268 : * of moderately high to high abbreviated cardinality. There is little to
2269 : * lose but much to gain, which our strategy reflects.
2270 : */
2271 0 : if (trace_sort)
2272 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2273 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2274 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2275 :
2276 0 : return true;
2277 : }
2278 :
2279 : /*
2280 : * Generic equalimage support function for character type's operator classes.
2281 : * Disables the use of deduplication with nondeterministic collations.
2282 : */
2283 : Datum
2284 8766 : btvarstrequalimage(PG_FUNCTION_ARGS)
2285 : {
2286 : /* Oid opcintype = PG_GETARG_OID(0); */
2287 8766 : Oid collid = PG_GET_COLLATION();
2288 : pg_locale_t locale;
2289 :
2290 8766 : check_collation_set(collid);
2291 :
2292 8766 : locale = pg_newlocale_from_collation(collid);
2293 :
2294 8766 : PG_RETURN_BOOL(locale->deterministic);
2295 : }
2296 :
2297 : Datum
2298 229560 : text_larger(PG_FUNCTION_ARGS)
2299 : {
2300 229560 : text *arg1 = PG_GETARG_TEXT_PP(0);
2301 229560 : text *arg2 = PG_GETARG_TEXT_PP(1);
2302 : text *result;
2303 :
2304 229560 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2305 :
2306 229560 : PG_RETURN_TEXT_P(result);
2307 : }
2308 :
2309 : Datum
2310 86076 : text_smaller(PG_FUNCTION_ARGS)
2311 : {
2312 86076 : text *arg1 = PG_GETARG_TEXT_PP(0);
2313 86076 : text *arg2 = PG_GETARG_TEXT_PP(1);
2314 : text *result;
2315 :
2316 86076 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2317 :
2318 86076 : PG_RETURN_TEXT_P(result);
2319 : }
2320 :
2321 :
2322 : /*
2323 : * Cross-type comparison functions for types text and name.
2324 : */
2325 :
2326 : Datum
2327 191254 : nameeqtext(PG_FUNCTION_ARGS)
2328 : {
2329 191254 : Name arg1 = PG_GETARG_NAME(0);
2330 191254 : text *arg2 = PG_GETARG_TEXT_PP(1);
2331 191254 : size_t len1 = strlen(NameStr(*arg1));
2332 191254 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2333 191254 : Oid collid = PG_GET_COLLATION();
2334 : bool result;
2335 :
2336 191254 : check_collation_set(collid);
2337 :
2338 191254 : if (collid == C_COLLATION_OID)
2339 255728 : result = (len1 == len2 &&
2340 124122 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2341 : else
2342 59648 : result = (varstr_cmp(NameStr(*arg1), len1,
2343 59648 : VARDATA_ANY(arg2), len2,
2344 : collid) == 0);
2345 :
2346 191254 : PG_FREE_IF_COPY(arg2, 1);
2347 :
2348 191254 : PG_RETURN_BOOL(result);
2349 : }
2350 :
2351 : Datum
2352 7938 : texteqname(PG_FUNCTION_ARGS)
2353 : {
2354 7938 : text *arg1 = PG_GETARG_TEXT_PP(0);
2355 7938 : Name arg2 = PG_GETARG_NAME(1);
2356 7938 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2357 7938 : size_t len2 = strlen(NameStr(*arg2));
2358 7938 : Oid collid = PG_GET_COLLATION();
2359 : bool result;
2360 :
2361 7938 : check_collation_set(collid);
2362 :
2363 7938 : if (collid == C_COLLATION_OID)
2364 568 : result = (len1 == len2 &&
2365 182 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2366 : else
2367 7552 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2368 7552 : NameStr(*arg2), len2,
2369 : collid) == 0);
2370 :
2371 7938 : PG_FREE_IF_COPY(arg1, 0);
2372 :
2373 7938 : PG_RETURN_BOOL(result);
2374 : }
2375 :
2376 : Datum
2377 18 : namenetext(PG_FUNCTION_ARGS)
2378 : {
2379 18 : Name arg1 = PG_GETARG_NAME(0);
2380 18 : text *arg2 = PG_GETARG_TEXT_PP(1);
2381 18 : size_t len1 = strlen(NameStr(*arg1));
2382 18 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2383 18 : Oid collid = PG_GET_COLLATION();
2384 : bool result;
2385 :
2386 18 : check_collation_set(collid);
2387 :
2388 18 : if (collid == C_COLLATION_OID)
2389 0 : result = !(len1 == len2 &&
2390 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2391 : else
2392 18 : result = !(varstr_cmp(NameStr(*arg1), len1,
2393 18 : VARDATA_ANY(arg2), len2,
2394 : collid) == 0);
2395 :
2396 18 : PG_FREE_IF_COPY(arg2, 1);
2397 :
2398 18 : PG_RETURN_BOOL(result);
2399 : }
2400 :
2401 : Datum
2402 18 : textnename(PG_FUNCTION_ARGS)
2403 : {
2404 18 : text *arg1 = PG_GETARG_TEXT_PP(0);
2405 18 : Name arg2 = PG_GETARG_NAME(1);
2406 18 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2407 18 : size_t len2 = strlen(NameStr(*arg2));
2408 18 : Oid collid = PG_GET_COLLATION();
2409 : bool result;
2410 :
2411 18 : check_collation_set(collid);
2412 :
2413 18 : if (collid == C_COLLATION_OID)
2414 0 : result = !(len1 == len2 &&
2415 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2416 : else
2417 18 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2418 18 : NameStr(*arg2), len2,
2419 : collid) == 0);
2420 :
2421 18 : PG_FREE_IF_COPY(arg1, 0);
2422 :
2423 18 : PG_RETURN_BOOL(result);
2424 : }
2425 :
2426 : Datum
2427 126368 : btnametextcmp(PG_FUNCTION_ARGS)
2428 : {
2429 126368 : Name arg1 = PG_GETARG_NAME(0);
2430 126368 : text *arg2 = PG_GETARG_TEXT_PP(1);
2431 : int32 result;
2432 :
2433 126368 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2434 126368 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2435 : PG_GET_COLLATION());
2436 :
2437 126368 : PG_FREE_IF_COPY(arg2, 1);
2438 :
2439 126368 : PG_RETURN_INT32(result);
2440 : }
2441 :
2442 : Datum
2443 44 : bttextnamecmp(PG_FUNCTION_ARGS)
2444 : {
2445 44 : text *arg1 = PG_GETARG_TEXT_PP(0);
2446 44 : Name arg2 = PG_GETARG_NAME(1);
2447 : int32 result;
2448 :
2449 44 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2450 44 : NameStr(*arg2), strlen(NameStr(*arg2)),
2451 : PG_GET_COLLATION());
2452 :
2453 44 : PG_FREE_IF_COPY(arg1, 0);
2454 :
2455 44 : PG_RETURN_INT32(result);
2456 : }
2457 :
2458 : #define CmpCall(cmpfunc) \
2459 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2460 : PG_GET_COLLATION(), \
2461 : PG_GETARG_DATUM(0), \
2462 : PG_GETARG_DATUM(1)))
2463 :
2464 : Datum
2465 60976 : namelttext(PG_FUNCTION_ARGS)
2466 : {
2467 60976 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2468 : }
2469 :
2470 : Datum
2471 0 : nameletext(PG_FUNCTION_ARGS)
2472 : {
2473 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2474 : }
2475 :
2476 : Datum
2477 0 : namegttext(PG_FUNCTION_ARGS)
2478 : {
2479 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2480 : }
2481 :
2482 : Datum
2483 52518 : namegetext(PG_FUNCTION_ARGS)
2484 : {
2485 52518 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2486 : }
2487 :
2488 : Datum
2489 0 : textltname(PG_FUNCTION_ARGS)
2490 : {
2491 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
2492 : }
2493 :
2494 : Datum
2495 0 : textlename(PG_FUNCTION_ARGS)
2496 : {
2497 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2498 : }
2499 :
2500 : Datum
2501 0 : textgtname(PG_FUNCTION_ARGS)
2502 : {
2503 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2504 : }
2505 :
2506 : Datum
2507 0 : textgename(PG_FUNCTION_ARGS)
2508 : {
2509 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2510 : }
2511 :
2512 : #undef CmpCall
2513 :
2514 :
2515 : /*
2516 : * The following operators support character-by-character comparison
2517 : * of text datums, to allow building indexes suitable for LIKE clauses.
2518 : * Note that the regular texteq/textne comparison operators, and regular
2519 : * support functions 1 and 2 with "C" collation are assumed to be
2520 : * compatible with these!
2521 : */
2522 :
2523 : static int
2524 160444 : internal_text_pattern_compare(text *arg1, text *arg2)
2525 : {
2526 : int result;
2527 : int len1,
2528 : len2;
2529 :
2530 160444 : len1 = VARSIZE_ANY_EXHDR(arg1);
2531 160444 : len2 = VARSIZE_ANY_EXHDR(arg2);
2532 :
2533 160444 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2534 160444 : if (result != 0)
2535 160312 : return result;
2536 132 : else if (len1 < len2)
2537 0 : return -1;
2538 132 : else if (len1 > len2)
2539 84 : return 1;
2540 : else
2541 48 : return 0;
2542 : }
2543 :
2544 :
2545 : Datum
2546 47866 : text_pattern_lt(PG_FUNCTION_ARGS)
2547 : {
2548 47866 : text *arg1 = PG_GETARG_TEXT_PP(0);
2549 47866 : text *arg2 = PG_GETARG_TEXT_PP(1);
2550 : int result;
2551 :
2552 47866 : result = internal_text_pattern_compare(arg1, arg2);
2553 :
2554 47866 : PG_FREE_IF_COPY(arg1, 0);
2555 47866 : PG_FREE_IF_COPY(arg2, 1);
2556 :
2557 47866 : PG_RETURN_BOOL(result < 0);
2558 : }
2559 :
2560 :
2561 : Datum
2562 37510 : text_pattern_le(PG_FUNCTION_ARGS)
2563 : {
2564 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2565 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2566 : int result;
2567 :
2568 37510 : result = internal_text_pattern_compare(arg1, arg2);
2569 :
2570 37510 : PG_FREE_IF_COPY(arg1, 0);
2571 37510 : PG_FREE_IF_COPY(arg2, 1);
2572 :
2573 37510 : PG_RETURN_BOOL(result <= 0);
2574 : }
2575 :
2576 :
2577 : Datum
2578 37534 : text_pattern_ge(PG_FUNCTION_ARGS)
2579 : {
2580 37534 : text *arg1 = PG_GETARG_TEXT_PP(0);
2581 37534 : text *arg2 = PG_GETARG_TEXT_PP(1);
2582 : int result;
2583 :
2584 37534 : result = internal_text_pattern_compare(arg1, arg2);
2585 :
2586 37534 : PG_FREE_IF_COPY(arg1, 0);
2587 37534 : PG_FREE_IF_COPY(arg2, 1);
2588 :
2589 37534 : PG_RETURN_BOOL(result >= 0);
2590 : }
2591 :
2592 :
2593 : Datum
2594 37510 : text_pattern_gt(PG_FUNCTION_ARGS)
2595 : {
2596 37510 : text *arg1 = PG_GETARG_TEXT_PP(0);
2597 37510 : text *arg2 = PG_GETARG_TEXT_PP(1);
2598 : int result;
2599 :
2600 37510 : result = internal_text_pattern_compare(arg1, arg2);
2601 :
2602 37510 : PG_FREE_IF_COPY(arg1, 0);
2603 37510 : PG_FREE_IF_COPY(arg2, 1);
2604 :
2605 37510 : PG_RETURN_BOOL(result > 0);
2606 : }
2607 :
2608 :
2609 : Datum
2610 24 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
2611 : {
2612 24 : text *arg1 = PG_GETARG_TEXT_PP(0);
2613 24 : text *arg2 = PG_GETARG_TEXT_PP(1);
2614 : int result;
2615 :
2616 24 : result = internal_text_pattern_compare(arg1, arg2);
2617 :
2618 24 : PG_FREE_IF_COPY(arg1, 0);
2619 24 : PG_FREE_IF_COPY(arg2, 1);
2620 :
2621 24 : PG_RETURN_INT32(result);
2622 : }
2623 :
2624 :
2625 : Datum
2626 116 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
2627 : {
2628 116 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
2629 : MemoryContext oldcontext;
2630 :
2631 116 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2632 :
2633 : /* Use generic string SortSupport, forcing "C" collation */
2634 116 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
2635 :
2636 116 : MemoryContextSwitchTo(oldcontext);
2637 :
2638 116 : PG_RETURN_VOID();
2639 : }
2640 :
2641 :
2642 : /* text_name()
2643 : * Converts a text type to a Name type.
2644 : */
2645 : Datum
2646 30778 : text_name(PG_FUNCTION_ARGS)
2647 : {
2648 30778 : text *s = PG_GETARG_TEXT_PP(0);
2649 : Name result;
2650 : int len;
2651 :
2652 30778 : len = VARSIZE_ANY_EXHDR(s);
2653 :
2654 : /* Truncate oversize input */
2655 30778 : if (len >= NAMEDATALEN)
2656 6 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
2657 :
2658 : /* We use palloc0 here to ensure result is zero-padded */
2659 30778 : result = (Name) palloc0(NAMEDATALEN);
2660 30778 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
2661 :
2662 30778 : PG_RETURN_NAME(result);
2663 : }
2664 :
2665 : /* name_text()
2666 : * Converts a Name type to a text type.
2667 : */
2668 : Datum
2669 654876 : name_text(PG_FUNCTION_ARGS)
2670 : {
2671 654876 : Name s = PG_GETARG_NAME(0);
2672 :
2673 654876 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
2674 : }
2675 :
2676 :
2677 : /*
2678 : * textToQualifiedNameList - convert a text object to list of names
2679 : *
2680 : * This implements the input parsing needed by nextval() and other
2681 : * functions that take a text parameter representing a qualified name.
2682 : * We split the name at dots, downcase if not double-quoted, and
2683 : * truncate names if they're too long.
2684 : */
2685 : List *
2686 5426 : textToQualifiedNameList(text *textval)
2687 : {
2688 : char *rawname;
2689 5426 : List *result = NIL;
2690 : List *namelist;
2691 : ListCell *l;
2692 :
2693 : /* Convert to C string (handles possible detoasting). */
2694 : /* Note we rely on being able to modify rawname below. */
2695 5426 : rawname = text_to_cstring(textval);
2696 :
2697 5426 : if (!SplitIdentifierString(rawname, '.', &namelist))
2698 0 : ereport(ERROR,
2699 : (errcode(ERRCODE_INVALID_NAME),
2700 : errmsg("invalid name syntax")));
2701 :
2702 5426 : if (namelist == NIL)
2703 0 : ereport(ERROR,
2704 : (errcode(ERRCODE_INVALID_NAME),
2705 : errmsg("invalid name syntax")));
2706 :
2707 10968 : foreach(l, namelist)
2708 : {
2709 5542 : char *curname = (char *) lfirst(l);
2710 :
2711 5542 : result = lappend(result, makeString(pstrdup(curname)));
2712 : }
2713 :
2714 5426 : pfree(rawname);
2715 5426 : list_free(namelist);
2716 :
2717 5426 : return result;
2718 : }
2719 :
2720 : /*
2721 : * SplitIdentifierString --- parse a string containing identifiers
2722 : *
2723 : * This is the guts of textToQualifiedNameList, and is exported for use in
2724 : * other situations such as parsing GUC variables. In the GUC case, it's
2725 : * important to avoid memory leaks, so the API is designed to minimize the
2726 : * amount of stuff that needs to be allocated and freed.
2727 : *
2728 : * Inputs:
2729 : * rawstring: the input string; must be overwritable! On return, it's
2730 : * been modified to contain the separated identifiers.
2731 : * separator: the separator punctuation expected between identifiers
2732 : * (typically '.' or ','). Whitespace may also appear around
2733 : * identifiers.
2734 : * Outputs:
2735 : * namelist: filled with a palloc'd list of pointers to identifiers within
2736 : * rawstring. Caller should list_free() this even on error return.
2737 : *
2738 : * Returns true if okay, false if there is a syntax error in the string.
2739 : *
2740 : * Note that an empty string is considered okay here, though not in
2741 : * textToQualifiedNameList.
2742 : */
2743 : bool
2744 373614 : SplitIdentifierString(char *rawstring, char separator,
2745 : List **namelist)
2746 : {
2747 373614 : char *nextp = rawstring;
2748 373614 : bool done = false;
2749 :
2750 373614 : *namelist = NIL;
2751 :
2752 373620 : while (scanner_isspace(*nextp))
2753 6 : nextp++; /* skip leading whitespace */
2754 :
2755 373614 : if (*nextp == '\0')
2756 30272 : return true; /* allow empty string */
2757 :
2758 : /* At the top of the loop, we are at start of a new identifier. */
2759 : do
2760 : {
2761 : char *curname;
2762 : char *endp;
2763 :
2764 642768 : if (*nextp == '"')
2765 : {
2766 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
2767 40230 : curname = nextp + 1;
2768 : for (;;)
2769 : {
2770 40234 : endp = strchr(nextp + 1, '"');
2771 40232 : if (endp == NULL)
2772 0 : return false; /* mismatched quotes */
2773 40232 : if (endp[1] != '"')
2774 40230 : break; /* found end of quoted name */
2775 : /* Collapse adjacent quotes into one quote, and look again */
2776 2 : memmove(endp, endp + 1, strlen(endp));
2777 2 : nextp = endp;
2778 : }
2779 : /* endp now points at the terminating quote */
2780 40230 : nextp = endp + 1;
2781 : }
2782 : else
2783 : {
2784 : /* Unquoted name --- extends to separator or whitespace */
2785 : char *downname;
2786 : int len;
2787 :
2788 602538 : curname = nextp;
2789 5528250 : while (*nextp && *nextp != separator &&
2790 4925714 : !scanner_isspace(*nextp))
2791 4925712 : nextp++;
2792 602538 : endp = nextp;
2793 602538 : if (curname == nextp)
2794 0 : return false; /* empty unquoted name not allowed */
2795 :
2796 : /*
2797 : * Downcase the identifier, using same code as main lexer does.
2798 : *
2799 : * XXX because we want to overwrite the input in-place, we cannot
2800 : * support a downcasing transformation that increases the string
2801 : * length. This is not a problem given the current implementation
2802 : * of downcase_truncate_identifier, but we'll probably have to do
2803 : * something about this someday.
2804 : */
2805 602538 : len = endp - curname;
2806 602538 : downname = downcase_truncate_identifier(curname, len, false);
2807 : Assert(strlen(downname) <= len);
2808 602538 : strncpy(curname, downname, len); /* strncpy is required here */
2809 602538 : pfree(downname);
2810 : }
2811 :
2812 642770 : while (scanner_isspace(*nextp))
2813 2 : nextp++; /* skip trailing whitespace */
2814 :
2815 642768 : if (*nextp == separator)
2816 : {
2817 299426 : nextp++;
2818 574286 : while (scanner_isspace(*nextp))
2819 274860 : nextp++; /* skip leading whitespace for next */
2820 : /* we expect another name, so done remains false */
2821 : }
2822 343342 : else if (*nextp == '\0')
2823 343340 : done = true;
2824 : else
2825 2 : return false; /* invalid syntax */
2826 :
2827 : /* Now safe to overwrite separator with a null */
2828 642766 : *endp = '\0';
2829 :
2830 : /* Truncate name if it's overlength */
2831 642766 : truncate_identifier(curname, strlen(curname), false);
2832 :
2833 : /*
2834 : * Finished isolating current name --- add it to list
2835 : */
2836 642766 : *namelist = lappend(*namelist, curname);
2837 :
2838 : /* Loop back if we didn't reach end of string */
2839 642766 : } while (!done);
2840 :
2841 343340 : return true;
2842 : }
2843 :
2844 :
2845 : /*
2846 : * SplitDirectoriesString --- parse a string containing file/directory names
2847 : *
2848 : * This works fine on file names too; the function name is historical.
2849 : *
2850 : * This is similar to SplitIdentifierString, except that the parsing
2851 : * rules are meant to handle pathnames instead of identifiers: there is
2852 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
2853 : * and we apply canonicalize_path() to each extracted string. Because of the
2854 : * last, the returned strings are separately palloc'd rather than being
2855 : * pointers into rawstring --- but we still scribble on rawstring.
2856 : *
2857 : * Inputs:
2858 : * rawstring: the input string; must be modifiable!
2859 : * separator: the separator punctuation expected between directories
2860 : * (typically ',' or ';'). Whitespace may also appear around
2861 : * directories.
2862 : * Outputs:
2863 : * namelist: filled with a palloc'd list of directory names.
2864 : * Caller should list_free_deep() this even on error return.
2865 : *
2866 : * Returns true if okay, false if there is a syntax error in the string.
2867 : *
2868 : * Note that an empty string is considered okay here.
2869 : */
2870 : bool
2871 1776 : SplitDirectoriesString(char *rawstring, char separator,
2872 : List **namelist)
2873 : {
2874 1776 : char *nextp = rawstring;
2875 1776 : bool done = false;
2876 :
2877 1776 : *namelist = NIL;
2878 :
2879 1776 : while (scanner_isspace(*nextp))
2880 0 : nextp++; /* skip leading whitespace */
2881 :
2882 1776 : if (*nextp == '\0')
2883 2 : return true; /* allow empty string */
2884 :
2885 : /* At the top of the loop, we are at start of a new directory. */
2886 : do
2887 : {
2888 : char *curname;
2889 : char *endp;
2890 :
2891 1776 : if (*nextp == '"')
2892 : {
2893 : /* Quoted name --- collapse quote-quote pairs */
2894 0 : curname = nextp + 1;
2895 : for (;;)
2896 : {
2897 0 : endp = strchr(nextp + 1, '"');
2898 0 : if (endp == NULL)
2899 0 : return false; /* mismatched quotes */
2900 0 : if (endp[1] != '"')
2901 0 : break; /* found end of quoted name */
2902 : /* Collapse adjacent quotes into one quote, and look again */
2903 0 : memmove(endp, endp + 1, strlen(endp));
2904 0 : nextp = endp;
2905 : }
2906 : /* endp now points at the terminating quote */
2907 0 : nextp = endp + 1;
2908 : }
2909 : else
2910 : {
2911 : /* Unquoted name --- extends to separator or end of string */
2912 1776 : curname = endp = nextp;
2913 29698 : while (*nextp && *nextp != separator)
2914 : {
2915 : /* trailing whitespace should not be included in name */
2916 27922 : if (!scanner_isspace(*nextp))
2917 27922 : endp = nextp + 1;
2918 27922 : nextp++;
2919 : }
2920 1776 : if (curname == endp)
2921 0 : return false; /* empty unquoted name not allowed */
2922 : }
2923 :
2924 1776 : while (scanner_isspace(*nextp))
2925 0 : nextp++; /* skip trailing whitespace */
2926 :
2927 1776 : if (*nextp == separator)
2928 : {
2929 2 : nextp++;
2930 2 : while (scanner_isspace(*nextp))
2931 0 : nextp++; /* skip leading whitespace for next */
2932 : /* we expect another name, so done remains false */
2933 : }
2934 1774 : else if (*nextp == '\0')
2935 1774 : done = true;
2936 : else
2937 0 : return false; /* invalid syntax */
2938 :
2939 : /* Now safe to overwrite separator with a null */
2940 1776 : *endp = '\0';
2941 :
2942 : /* Truncate path if it's overlength */
2943 1776 : if (strlen(curname) >= MAXPGPATH)
2944 0 : curname[MAXPGPATH - 1] = '\0';
2945 :
2946 : /*
2947 : * Finished isolating current name --- add it to list
2948 : */
2949 1776 : curname = pstrdup(curname);
2950 1776 : canonicalize_path(curname);
2951 1776 : *namelist = lappend(*namelist, curname);
2952 :
2953 : /* Loop back if we didn't reach end of string */
2954 1776 : } while (!done);
2955 :
2956 1774 : return true;
2957 : }
2958 :
2959 :
2960 : /*
2961 : * SplitGUCList --- parse a string containing identifiers or file names
2962 : *
2963 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
2964 : * presuming whether the elements will be taken as identifiers or file names.
2965 : * We assume the input has already been through flatten_set_variable_args(),
2966 : * so that we need never downcase (if appropriate, that was done already).
2967 : * Nor do we ever truncate, since we don't know the correct max length.
2968 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
2969 : * because any embedded whitespace should have led to double-quoting).
2970 : * Otherwise the API is identical to SplitIdentifierString.
2971 : *
2972 : * XXX it's annoying to have so many copies of this string-splitting logic.
2973 : * However, it's not clear that having one function with a bunch of option
2974 : * flags would be much better.
2975 : *
2976 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
2977 : * Be sure to update that if you have to change this.
2978 : *
2979 : * Inputs:
2980 : * rawstring: the input string; must be overwritable! On return, it's
2981 : * been modified to contain the separated identifiers.
2982 : * separator: the separator punctuation expected between identifiers
2983 : * (typically '.' or ','). Whitespace may also appear around
2984 : * identifiers.
2985 : * Outputs:
2986 : * namelist: filled with a palloc'd list of pointers to identifiers within
2987 : * rawstring. Caller should list_free() this even on error return.
2988 : *
2989 : * Returns true if okay, false if there is a syntax error in the string.
2990 : */
2991 : bool
2992 3920 : SplitGUCList(char *rawstring, char separator,
2993 : List **namelist)
2994 : {
2995 3920 : char *nextp = rawstring;
2996 3920 : bool done = false;
2997 :
2998 3920 : *namelist = NIL;
2999 :
3000 3920 : while (scanner_isspace(*nextp))
3001 0 : nextp++; /* skip leading whitespace */
3002 :
3003 3920 : if (*nextp == '\0')
3004 3846 : return true; /* allow empty string */
3005 :
3006 : /* At the top of the loop, we are at start of a new identifier. */
3007 : do
3008 : {
3009 : char *curname;
3010 : char *endp;
3011 :
3012 100 : if (*nextp == '"')
3013 : {
3014 : /* Quoted name --- collapse quote-quote pairs */
3015 24 : curname = nextp + 1;
3016 : for (;;)
3017 : {
3018 36 : endp = strchr(nextp + 1, '"');
3019 30 : if (endp == NULL)
3020 0 : return false; /* mismatched quotes */
3021 30 : if (endp[1] != '"')
3022 24 : break; /* found end of quoted name */
3023 : /* Collapse adjacent quotes into one quote, and look again */
3024 6 : memmove(endp, endp + 1, strlen(endp));
3025 6 : nextp = endp;
3026 : }
3027 : /* endp now points at the terminating quote */
3028 24 : nextp = endp + 1;
3029 : }
3030 : else
3031 : {
3032 : /* Unquoted name --- extends to separator or whitespace */
3033 76 : curname = nextp;
3034 718 : while (*nextp && *nextp != separator &&
3035 642 : !scanner_isspace(*nextp))
3036 642 : nextp++;
3037 76 : endp = nextp;
3038 76 : if (curname == nextp)
3039 0 : return false; /* empty unquoted name not allowed */
3040 : }
3041 :
3042 100 : while (scanner_isspace(*nextp))
3043 0 : nextp++; /* skip trailing whitespace */
3044 :
3045 100 : if (*nextp == separator)
3046 : {
3047 26 : nextp++;
3048 44 : while (scanner_isspace(*nextp))
3049 18 : nextp++; /* skip leading whitespace for next */
3050 : /* we expect another name, so done remains false */
3051 : }
3052 74 : else if (*nextp == '\0')
3053 74 : done = true;
3054 : else
3055 0 : return false; /* invalid syntax */
3056 :
3057 : /* Now safe to overwrite separator with a null */
3058 100 : *endp = '\0';
3059 :
3060 : /*
3061 : * Finished isolating current name --- add it to list
3062 : */
3063 100 : *namelist = lappend(*namelist, curname);
3064 :
3065 : /* Loop back if we didn't reach end of string */
3066 100 : } while (!done);
3067 :
3068 74 : return true;
3069 : }
3070 :
3071 : /*
3072 : * appendStringInfoText
3073 : *
3074 : * Append a text to str.
3075 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3076 : */
3077 : static void
3078 1926418 : appendStringInfoText(StringInfo str, const text *t)
3079 : {
3080 1926418 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
3081 1926418 : }
3082 :
3083 : /*
3084 : * replace_text
3085 : * replace all occurrences of 'old_sub_str' in 'orig_str'
3086 : * with 'new_sub_str' to form 'new_str'
3087 : *
3088 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3089 : * otherwise returns 'new_str'
3090 : */
3091 : Datum
3092 1462 : replace_text(PG_FUNCTION_ARGS)
3093 : {
3094 1462 : text *src_text = PG_GETARG_TEXT_PP(0);
3095 1462 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
3096 1462 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
3097 : int src_text_len;
3098 : int from_sub_text_len;
3099 : TextPositionState state;
3100 : text *ret_text;
3101 : int chunk_len;
3102 : char *curr_ptr;
3103 : char *start_ptr;
3104 : StringInfoData str;
3105 : bool found;
3106 :
3107 1462 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
3108 1462 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
3109 :
3110 : /* Return unmodified source string if empty source or pattern */
3111 1462 : if (src_text_len < 1 || from_sub_text_len < 1)
3112 : {
3113 0 : PG_RETURN_TEXT_P(src_text);
3114 : }
3115 :
3116 1462 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
3117 :
3118 1462 : found = text_position_next(&state);
3119 :
3120 : /* When the from_sub_text is not found, there is nothing to do. */
3121 1462 : if (!found)
3122 : {
3123 324 : text_position_cleanup(&state);
3124 324 : PG_RETURN_TEXT_P(src_text);
3125 : }
3126 1138 : curr_ptr = text_position_get_match_ptr(&state);
3127 1138 : start_ptr = VARDATA_ANY(src_text);
3128 :
3129 1138 : initStringInfo(&str);
3130 :
3131 : do
3132 : {
3133 6780 : CHECK_FOR_INTERRUPTS();
3134 :
3135 : /* copy the data skipped over by last text_position_next() */
3136 6780 : chunk_len = curr_ptr - start_ptr;
3137 6780 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3138 :
3139 6780 : appendStringInfoText(&str, to_sub_text);
3140 :
3141 6780 : start_ptr = curr_ptr + state.last_match_len;
3142 :
3143 6780 : found = text_position_next(&state);
3144 6780 : if (found)
3145 5642 : curr_ptr = text_position_get_match_ptr(&state);
3146 : }
3147 6780 : while (found);
3148 :
3149 : /* copy trailing data */
3150 1138 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3151 1138 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3152 :
3153 1138 : text_position_cleanup(&state);
3154 :
3155 1138 : ret_text = cstring_to_text_with_len(str.data, str.len);
3156 1138 : pfree(str.data);
3157 :
3158 1138 : PG_RETURN_TEXT_P(ret_text);
3159 : }
3160 :
3161 : /*
3162 : * check_replace_text_has_escape
3163 : *
3164 : * Returns 0 if text contains no backslashes that need processing.
3165 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
3166 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
3167 : */
3168 : static int
3169 18736 : check_replace_text_has_escape(const text *replace_text)
3170 : {
3171 18736 : int result = 0;
3172 18736 : const char *p = VARDATA_ANY(replace_text);
3173 18736 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3174 :
3175 37516 : while (p < p_end)
3176 : {
3177 : /* Find next escape char, if any. */
3178 17628 : p = memchr(p, '\\', p_end - p);
3179 17628 : if (p == NULL)
3180 16806 : break;
3181 822 : p++;
3182 : /* Note: a backslash at the end doesn't require extra processing. */
3183 822 : if (p < p_end)
3184 : {
3185 822 : if (*p >= '1' && *p <= '9')
3186 778 : return 2; /* Found a submatch specifier, so done */
3187 44 : result = 1; /* Found some other sequence, keep looking */
3188 44 : p++;
3189 : }
3190 : }
3191 17958 : return result;
3192 : }
3193 :
3194 : /*
3195 : * appendStringInfoRegexpSubstr
3196 : *
3197 : * Append replace_text to str, substituting regexp back references for
3198 : * \n escapes. start_ptr is the start of the match in the source string,
3199 : * at logical character position data_pos.
3200 : */
3201 : static void
3202 236 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
3203 : regmatch_t *pmatch,
3204 : char *start_ptr, int data_pos)
3205 : {
3206 236 : const char *p = VARDATA_ANY(replace_text);
3207 236 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3208 :
3209 574 : while (p < p_end)
3210 : {
3211 518 : const char *chunk_start = p;
3212 : int so;
3213 : int eo;
3214 :
3215 : /* Find next escape char, if any. */
3216 518 : p = memchr(p, '\\', p_end - p);
3217 518 : if (p == NULL)
3218 174 : p = p_end;
3219 :
3220 : /* Copy the text we just scanned over, if any. */
3221 518 : if (p > chunk_start)
3222 318 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3223 :
3224 : /* Done if at end of string, else advance over escape char. */
3225 518 : if (p >= p_end)
3226 174 : break;
3227 344 : p++;
3228 :
3229 344 : if (p >= p_end)
3230 : {
3231 : /* Escape at very end of input. Treat same as unexpected char */
3232 6 : appendStringInfoChar(str, '\\');
3233 6 : break;
3234 : }
3235 :
3236 338 : if (*p >= '1' && *p <= '9')
3237 278 : {
3238 : /* Use the back reference of regexp. */
3239 278 : int idx = *p - '0';
3240 :
3241 278 : so = pmatch[idx].rm_so;
3242 278 : eo = pmatch[idx].rm_eo;
3243 278 : p++;
3244 : }
3245 60 : else if (*p == '&')
3246 : {
3247 : /* Use the entire matched string. */
3248 18 : so = pmatch[0].rm_so;
3249 18 : eo = pmatch[0].rm_eo;
3250 18 : p++;
3251 : }
3252 42 : else if (*p == '\\')
3253 : {
3254 : /* \\ means transfer one \ to output. */
3255 36 : appendStringInfoChar(str, '\\');
3256 36 : p++;
3257 36 : continue;
3258 : }
3259 : else
3260 : {
3261 : /*
3262 : * If escape char is not followed by any expected char, just treat
3263 : * it as ordinary data to copy. (XXX would it be better to throw
3264 : * an error?)
3265 : */
3266 6 : appendStringInfoChar(str, '\\');
3267 6 : continue;
3268 : }
3269 :
3270 296 : if (so >= 0 && eo >= 0)
3271 : {
3272 : /*
3273 : * Copy the text that is back reference of regexp. Note so and eo
3274 : * are counted in characters not bytes.
3275 : */
3276 : char *chunk_start;
3277 : int chunk_len;
3278 :
3279 : Assert(so >= data_pos);
3280 296 : chunk_start = start_ptr;
3281 296 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3282 296 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3283 296 : appendBinaryStringInfo(str, chunk_start, chunk_len);
3284 : }
3285 : }
3286 236 : }
3287 :
3288 : /*
3289 : * replace_text_regexp
3290 : *
3291 : * replace substring(s) in src_text that match pattern with replace_text.
3292 : * The replace_text can contain backslash markers to substitute
3293 : * (parts of) the matched text.
3294 : *
3295 : * cflags: regexp compile flags.
3296 : * collation: collation to use.
3297 : * search_start: the character (not byte) offset in src_text at which to
3298 : * begin searching.
3299 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
3300 : */
3301 : text *
3302 18736 : replace_text_regexp(text *src_text, text *pattern_text,
3303 : text *replace_text,
3304 : int cflags, Oid collation,
3305 : int search_start, int n)
3306 : {
3307 : text *ret_text;
3308 : regex_t *re;
3309 18736 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3310 18736 : int nmatches = 0;
3311 : StringInfoData buf;
3312 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
3313 18736 : int nmatch = lengthof(pmatch);
3314 : pg_wchar *data;
3315 : size_t data_len;
3316 : int data_pos;
3317 : char *start_ptr;
3318 : int escape_status;
3319 :
3320 18736 : initStringInfo(&buf);
3321 :
3322 : /* Convert data string to wide characters. */
3323 18736 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3324 18736 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3325 :
3326 : /* Check whether replace_text has escapes, especially regexp submatches. */
3327 18736 : escape_status = check_replace_text_has_escape(replace_text);
3328 :
3329 : /* If no regexp submatches, we can use REG_NOSUB. */
3330 18736 : if (escape_status < 2)
3331 : {
3332 17958 : cflags |= REG_NOSUB;
3333 : /* Also tell pg_regexec we only want the whole-match location. */
3334 17958 : nmatch = 1;
3335 : }
3336 :
3337 : /* Prepare the regexp. */
3338 18736 : re = RE_compile_and_cache(pattern_text, cflags, collation);
3339 :
3340 : /* start_ptr points to the data_pos'th character of src_text */
3341 18736 : start_ptr = (char *) VARDATA_ANY(src_text);
3342 18736 : data_pos = 0;
3343 :
3344 25148 : while (search_start <= data_len)
3345 : {
3346 : int regexec_result;
3347 :
3348 25142 : CHECK_FOR_INTERRUPTS();
3349 :
3350 25142 : regexec_result = pg_regexec(re,
3351 : data,
3352 : data_len,
3353 : search_start,
3354 : NULL, /* no details */
3355 : nmatch,
3356 : pmatch,
3357 : 0);
3358 :
3359 25142 : if (regexec_result == REG_NOMATCH)
3360 16688 : break;
3361 :
3362 8454 : if (regexec_result != REG_OKAY)
3363 : {
3364 : char errMsg[100];
3365 :
3366 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3367 0 : ereport(ERROR,
3368 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3369 : errmsg("regular expression failed: %s", errMsg)));
3370 : }
3371 :
3372 : /*
3373 : * Count matches, and decide whether to replace this match.
3374 : */
3375 8454 : nmatches++;
3376 8454 : if (n > 0 && nmatches != n)
3377 : {
3378 : /*
3379 : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
3380 : * we treat the matched text as if it weren't matched, and copy it
3381 : * to the output later.)
3382 : */
3383 60 : search_start = pmatch[0].rm_eo;
3384 60 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3385 0 : search_start++;
3386 60 : continue;
3387 : }
3388 :
3389 : /*
3390 : * Copy the text to the left of the match position. Note we are given
3391 : * character not byte indexes.
3392 : */
3393 8394 : if (pmatch[0].rm_so - data_pos > 0)
3394 : {
3395 : int chunk_len;
3396 :
3397 8236 : chunk_len = charlen_to_bytelen(start_ptr,
3398 8236 : pmatch[0].rm_so - data_pos);
3399 8236 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3400 :
3401 : /*
3402 : * Advance start_ptr over that text, to avoid multiple rescans of
3403 : * it if the replace_text contains multiple back-references.
3404 : */
3405 8236 : start_ptr += chunk_len;
3406 8236 : data_pos = pmatch[0].rm_so;
3407 : }
3408 :
3409 : /*
3410 : * Copy the replace_text, processing escapes if any are present.
3411 : */
3412 8394 : if (escape_status > 0)
3413 236 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3414 : start_ptr, data_pos);
3415 : else
3416 8158 : appendStringInfoText(&buf, replace_text);
3417 :
3418 : /* Advance start_ptr and data_pos over the matched text. */
3419 16788 : start_ptr += charlen_to_bytelen(start_ptr,
3420 8394 : pmatch[0].rm_eo - data_pos);
3421 8394 : data_pos = pmatch[0].rm_eo;
3422 :
3423 : /*
3424 : * If we only want to replace one occurrence, we're done.
3425 : */
3426 8394 : if (n > 0)
3427 2042 : break;
3428 :
3429 : /*
3430 : * Advance search position. Normally we start the next search at the
3431 : * end of the previous match; but if the match was of zero length, we
3432 : * have to advance by one character, or we'd just find the same match
3433 : * again.
3434 : */
3435 6352 : search_start = data_pos;
3436 6352 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3437 12 : search_start++;
3438 : }
3439 :
3440 : /*
3441 : * Copy the text to the right of the last match.
3442 : */
3443 18736 : if (data_pos < data_len)
3444 : {
3445 : int chunk_len;
3446 :
3447 17862 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3448 17862 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3449 : }
3450 :
3451 18736 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
3452 18736 : pfree(buf.data);
3453 18736 : pfree(data);
3454 :
3455 18736 : return ret_text;
3456 : }
3457 :
3458 : /*
3459 : * split_part
3460 : * parse input string based on provided field separator
3461 : * return N'th item (1 based, negative counts from end)
3462 : */
3463 : Datum
3464 150 : split_part(PG_FUNCTION_ARGS)
3465 : {
3466 150 : text *inputstring = PG_GETARG_TEXT_PP(0);
3467 150 : text *fldsep = PG_GETARG_TEXT_PP(1);
3468 150 : int fldnum = PG_GETARG_INT32(2);
3469 : int inputstring_len;
3470 : int fldsep_len;
3471 : TextPositionState state;
3472 : char *start_ptr;
3473 : char *end_ptr;
3474 : text *result_text;
3475 : bool found;
3476 :
3477 : /* field number is 1 based */
3478 150 : if (fldnum == 0)
3479 6 : ereport(ERROR,
3480 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3481 : errmsg("field position must not be zero")));
3482 :
3483 144 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3484 144 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3485 :
3486 : /* return empty string for empty input string */
3487 144 : if (inputstring_len < 1)
3488 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3489 :
3490 : /* handle empty field separator */
3491 132 : if (fldsep_len < 1)
3492 : {
3493 : /* if first or last field, return input string, else empty string */
3494 24 : if (fldnum == 1 || fldnum == -1)
3495 12 : PG_RETURN_TEXT_P(inputstring);
3496 : else
3497 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3498 : }
3499 :
3500 : /* find the first field separator */
3501 108 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
3502 :
3503 108 : found = text_position_next(&state);
3504 :
3505 : /* special case if fldsep not found at all */
3506 108 : if (!found)
3507 : {
3508 24 : text_position_cleanup(&state);
3509 : /* if first or last field, return input string, else empty string */
3510 24 : if (fldnum == 1 || fldnum == -1)
3511 12 : PG_RETURN_TEXT_P(inputstring);
3512 : else
3513 12 : PG_RETURN_TEXT_P(cstring_to_text(""));
3514 : }
3515 :
3516 : /*
3517 : * take care of a negative field number (i.e. count from the right) by
3518 : * converting to a positive field number; we need total number of fields
3519 : */
3520 84 : if (fldnum < 0)
3521 : {
3522 : /* we found a fldsep, so there are at least two fields */
3523 42 : int numfields = 2;
3524 :
3525 54 : while (text_position_next(&state))
3526 12 : numfields++;
3527 :
3528 : /* special case of last field does not require an extra pass */
3529 42 : if (fldnum == -1)
3530 : {
3531 24 : start_ptr = text_position_get_match_ptr(&state) + state.last_match_len;
3532 24 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
3533 24 : text_position_cleanup(&state);
3534 24 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
3535 : end_ptr - start_ptr));
3536 : }
3537 :
3538 : /* else, convert fldnum to positive notation */
3539 18 : fldnum += numfields + 1;
3540 :
3541 : /* if nonexistent field, return empty string */
3542 18 : if (fldnum <= 0)
3543 : {
3544 6 : text_position_cleanup(&state);
3545 6 : PG_RETURN_TEXT_P(cstring_to_text(""));
3546 : }
3547 :
3548 : /* reset to pointing at first match, but now with positive fldnum */
3549 12 : text_position_reset(&state);
3550 12 : found = text_position_next(&state);
3551 : Assert(found);
3552 : }
3553 :
3554 : /* identify bounds of first field */
3555 54 : start_ptr = VARDATA_ANY(inputstring);
3556 54 : end_ptr = text_position_get_match_ptr(&state);
3557 :
3558 102 : while (found && --fldnum > 0)
3559 : {
3560 : /* identify bounds of next field */
3561 48 : start_ptr = end_ptr + state.last_match_len;
3562 48 : found = text_position_next(&state);
3563 48 : if (found)
3564 18 : end_ptr = text_position_get_match_ptr(&state);
3565 : }
3566 :
3567 54 : text_position_cleanup(&state);
3568 :
3569 54 : if (fldnum > 0)
3570 : {
3571 : /* N'th field separator not found */
3572 : /* if last field requested, return it, else empty string */
3573 30 : if (fldnum == 1)
3574 : {
3575 24 : int last_len = start_ptr - VARDATA_ANY(inputstring);
3576 :
3577 24 : result_text = cstring_to_text_with_len(start_ptr,
3578 : inputstring_len - last_len);
3579 : }
3580 : else
3581 6 : result_text = cstring_to_text("");
3582 : }
3583 : else
3584 : {
3585 : /* non-last field requested */
3586 24 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
3587 : }
3588 :
3589 54 : PG_RETURN_TEXT_P(result_text);
3590 : }
3591 :
3592 : /*
3593 : * Convenience function to return true when two text params are equal.
3594 : */
3595 : static bool
3596 384 : text_isequal(text *txt1, text *txt2, Oid collid)
3597 : {
3598 384 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
3599 : collid,
3600 : PointerGetDatum(txt1),
3601 : PointerGetDatum(txt2)));
3602 : }
3603 :
3604 : /*
3605 : * text_to_array
3606 : * parse input string and return text array of elements,
3607 : * based on provided field separator
3608 : */
3609 : Datum
3610 170 : text_to_array(PG_FUNCTION_ARGS)
3611 : {
3612 : SplitTextOutputData tstate;
3613 :
3614 : /* For array output, tstate should start as all zeroes */
3615 170 : memset(&tstate, 0, sizeof(tstate));
3616 :
3617 170 : if (!split_text(fcinfo, &tstate))
3618 6 : PG_RETURN_NULL();
3619 :
3620 164 : if (tstate.astate == NULL)
3621 6 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
3622 :
3623 158 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
3624 : CurrentMemoryContext));
3625 : }
3626 :
3627 : /*
3628 : * text_to_array_null
3629 : * parse input string and return text array of elements,
3630 : * based on provided field separator and null string
3631 : *
3632 : * This is a separate entry point only to prevent the regression tests from
3633 : * complaining about different argument sets for the same internal function.
3634 : */
3635 : Datum
3636 60 : text_to_array_null(PG_FUNCTION_ARGS)
3637 : {
3638 60 : return text_to_array(fcinfo);
3639 : }
3640 :
3641 : /*
3642 : * text_to_table
3643 : * parse input string and return table of elements,
3644 : * based on provided field separator
3645 : */
3646 : Datum
3647 84 : text_to_table(PG_FUNCTION_ARGS)
3648 : {
3649 84 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
3650 : SplitTextOutputData tstate;
3651 :
3652 84 : tstate.astate = NULL;
3653 84 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
3654 84 : tstate.tupstore = rsi->setResult;
3655 84 : tstate.tupdesc = rsi->setDesc;
3656 :
3657 84 : (void) split_text(fcinfo, &tstate);
3658 :
3659 84 : return (Datum) 0;
3660 : }
3661 :
3662 : /*
3663 : * text_to_table_null
3664 : * parse input string and return table of elements,
3665 : * based on provided field separator and null string
3666 : *
3667 : * This is a separate entry point only to prevent the regression tests from
3668 : * complaining about different argument sets for the same internal function.
3669 : */
3670 : Datum
3671 24 : text_to_table_null(PG_FUNCTION_ARGS)
3672 : {
3673 24 : return text_to_table(fcinfo);
3674 : }
3675 :
3676 : /*
3677 : * Common code for text_to_array, text_to_array_null, text_to_table
3678 : * and text_to_table_null functions.
3679 : *
3680 : * These are not strict so we have to test for null inputs explicitly.
3681 : * Returns false if result is to be null, else returns true.
3682 : *
3683 : * Note that if the result is valid but empty (zero elements), we return
3684 : * without changing *tstate --- caller must handle that case, too.
3685 : */
3686 : static bool
3687 254 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
3688 : {
3689 : text *inputstring;
3690 : text *fldsep;
3691 : text *null_string;
3692 254 : Oid collation = PG_GET_COLLATION();
3693 : int inputstring_len;
3694 : int fldsep_len;
3695 : char *start_ptr;
3696 : text *result_text;
3697 :
3698 : /* when input string is NULL, then result is NULL too */
3699 254 : if (PG_ARGISNULL(0))
3700 12 : return false;
3701 :
3702 242 : inputstring = PG_GETARG_TEXT_PP(0);
3703 :
3704 : /* fldsep can be NULL */
3705 242 : if (!PG_ARGISNULL(1))
3706 212 : fldsep = PG_GETARG_TEXT_PP(1);
3707 : else
3708 30 : fldsep = NULL;
3709 :
3710 : /* null_string can be NULL or omitted */
3711 242 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
3712 84 : null_string = PG_GETARG_TEXT_PP(2);
3713 : else
3714 158 : null_string = NULL;
3715 :
3716 242 : if (fldsep != NULL)
3717 : {
3718 : /*
3719 : * Normal case with non-null fldsep. Use the text_position machinery
3720 : * to search for occurrences of fldsep.
3721 : */
3722 : TextPositionState state;
3723 :
3724 212 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3725 212 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3726 :
3727 : /* return empty set for empty input string */
3728 212 : if (inputstring_len < 1)
3729 60 : return true;
3730 :
3731 : /* empty field separator: return input string as a one-element set */
3732 200 : if (fldsep_len < 1)
3733 : {
3734 48 : split_text_accum_result(tstate, inputstring,
3735 : null_string, collation);
3736 48 : return true;
3737 : }
3738 :
3739 152 : text_position_setup(inputstring, fldsep, collation, &state);
3740 :
3741 152 : start_ptr = VARDATA_ANY(inputstring);
3742 :
3743 : for (;;)
3744 512 : {
3745 : bool found;
3746 : char *end_ptr;
3747 : int chunk_len;
3748 :
3749 664 : CHECK_FOR_INTERRUPTS();
3750 :
3751 664 : found = text_position_next(&state);
3752 664 : if (!found)
3753 : {
3754 : /* fetch last field */
3755 152 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
3756 152 : end_ptr = NULL; /* not used, but some compilers complain */
3757 : }
3758 : else
3759 : {
3760 : /* fetch non-last field */
3761 512 : end_ptr = text_position_get_match_ptr(&state);
3762 512 : chunk_len = end_ptr - start_ptr;
3763 : }
3764 :
3765 : /* build a temp text datum to pass to split_text_accum_result */
3766 664 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3767 :
3768 : /* stash away this field */
3769 664 : split_text_accum_result(tstate, result_text,
3770 : null_string, collation);
3771 :
3772 664 : pfree(result_text);
3773 :
3774 664 : if (!found)
3775 152 : break;
3776 :
3777 512 : start_ptr = end_ptr + state.last_match_len;
3778 : }
3779 :
3780 152 : text_position_cleanup(&state);
3781 : }
3782 : else
3783 : {
3784 : /*
3785 : * When fldsep is NULL, each character in the input string becomes a
3786 : * separate element in the result set. The separator is effectively
3787 : * the space between characters.
3788 : */
3789 30 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3790 :
3791 30 : start_ptr = VARDATA_ANY(inputstring);
3792 :
3793 252 : while (inputstring_len > 0)
3794 : {
3795 222 : int chunk_len = pg_mblen(start_ptr);
3796 :
3797 222 : CHECK_FOR_INTERRUPTS();
3798 :
3799 : /* build a temp text datum to pass to split_text_accum_result */
3800 222 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3801 :
3802 : /* stash away this field */
3803 222 : split_text_accum_result(tstate, result_text,
3804 : null_string, collation);
3805 :
3806 222 : pfree(result_text);
3807 :
3808 222 : start_ptr += chunk_len;
3809 222 : inputstring_len -= chunk_len;
3810 : }
3811 : }
3812 :
3813 182 : return true;
3814 : }
3815 :
3816 : /*
3817 : * Add text item to result set (table or array).
3818 : *
3819 : * This is also responsible for checking to see if the item matches
3820 : * the null_string, in which case we should emit NULL instead.
3821 : */
3822 : static void
3823 934 : split_text_accum_result(SplitTextOutputData *tstate,
3824 : text *field_value,
3825 : text *null_string,
3826 : Oid collation)
3827 : {
3828 934 : bool is_null = false;
3829 :
3830 934 : if (null_string && text_isequal(field_value, null_string, collation))
3831 72 : is_null = true;
3832 :
3833 934 : if (tstate->tupstore)
3834 : {
3835 : Datum values[1];
3836 : bool nulls[1];
3837 :
3838 228 : values[0] = PointerGetDatum(field_value);
3839 228 : nulls[0] = is_null;
3840 :
3841 228 : tuplestore_putvalues(tstate->tupstore,
3842 : tstate->tupdesc,
3843 : values,
3844 : nulls);
3845 : }
3846 : else
3847 : {
3848 706 : tstate->astate = accumArrayResult(tstate->astate,
3849 : PointerGetDatum(field_value),
3850 : is_null,
3851 : TEXTOID,
3852 : CurrentMemoryContext);
3853 : }
3854 934 : }
3855 :
3856 : /*
3857 : * array_to_text
3858 : * concatenate Cstring representation of input array elements
3859 : * using provided field separator
3860 : */
3861 : Datum
3862 77142 : array_to_text(PG_FUNCTION_ARGS)
3863 : {
3864 77142 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
3865 77142 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3866 :
3867 77142 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
3868 : }
3869 :
3870 : /*
3871 : * array_to_text_null
3872 : * concatenate Cstring representation of input array elements
3873 : * using provided field separator and null string
3874 : *
3875 : * This version is not strict so we have to test for null inputs explicitly.
3876 : */
3877 : Datum
3878 12 : array_to_text_null(PG_FUNCTION_ARGS)
3879 : {
3880 : ArrayType *v;
3881 : char *fldsep;
3882 : char *null_string;
3883 :
3884 : /* returns NULL when first or second parameter is NULL */
3885 12 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
3886 0 : PG_RETURN_NULL();
3887 :
3888 12 : v = PG_GETARG_ARRAYTYPE_P(0);
3889 12 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3890 :
3891 : /* NULL null string is passed through as a null pointer */
3892 12 : if (!PG_ARGISNULL(2))
3893 6 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
3894 : else
3895 6 : null_string = NULL;
3896 :
3897 12 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
3898 : }
3899 :
3900 : /*
3901 : * common code for array_to_text and array_to_text_null functions
3902 : */
3903 : static text *
3904 77172 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
3905 : const char *fldsep, const char *null_string)
3906 : {
3907 : text *result;
3908 : int nitems,
3909 : *dims,
3910 : ndims;
3911 : Oid element_type;
3912 : int typlen;
3913 : bool typbyval;
3914 : char typalign;
3915 : StringInfoData buf;
3916 77172 : bool printed = false;
3917 : char *p;
3918 : bits8 *bitmap;
3919 : int bitmask;
3920 : int i;
3921 : ArrayMetaState *my_extra;
3922 :
3923 77172 : ndims = ARR_NDIM(v);
3924 77172 : dims = ARR_DIMS(v);
3925 77172 : nitems = ArrayGetNItems(ndims, dims);
3926 :
3927 : /* if there are no elements, return an empty string */
3928 77172 : if (nitems == 0)
3929 52168 : return cstring_to_text_with_len("", 0);
3930 :
3931 25004 : element_type = ARR_ELEMTYPE(v);
3932 25004 : initStringInfo(&buf);
3933 :
3934 : /*
3935 : * We arrange to look up info about element type, including its output
3936 : * conversion proc, only once per series of calls, assuming the element
3937 : * type doesn't change underneath us.
3938 : */
3939 25004 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3940 25004 : if (my_extra == NULL)
3941 : {
3942 1444 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
3943 : sizeof(ArrayMetaState));
3944 1444 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3945 1444 : my_extra->element_type = ~element_type;
3946 : }
3947 :
3948 25004 : if (my_extra->element_type != element_type)
3949 : {
3950 : /*
3951 : * Get info about element type, including its output conversion proc
3952 : */
3953 1444 : get_type_io_data(element_type, IOFunc_output,
3954 : &my_extra->typlen, &my_extra->typbyval,
3955 : &my_extra->typalign, &my_extra->typdelim,
3956 : &my_extra->typioparam, &my_extra->typiofunc);
3957 1444 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
3958 1444 : fcinfo->flinfo->fn_mcxt);
3959 1444 : my_extra->element_type = element_type;
3960 : }
3961 25004 : typlen = my_extra->typlen;
3962 25004 : typbyval = my_extra->typbyval;
3963 25004 : typalign = my_extra->typalign;
3964 :
3965 25004 : p = ARR_DATA_PTR(v);
3966 25004 : bitmap = ARR_NULLBITMAP(v);
3967 25004 : bitmask = 1;
3968 :
3969 85028 : for (i = 0; i < nitems; i++)
3970 : {
3971 : Datum itemvalue;
3972 : char *value;
3973 :
3974 : /* Get source element, checking for NULL */
3975 60024 : if (bitmap && (*bitmap & bitmask) == 0)
3976 : {
3977 : /* if null_string is NULL, we just ignore null elements */
3978 18 : if (null_string != NULL)
3979 : {
3980 6 : if (printed)
3981 6 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
3982 : else
3983 0 : appendStringInfoString(&buf, null_string);
3984 6 : printed = true;
3985 : }
3986 : }
3987 : else
3988 : {
3989 60006 : itemvalue = fetch_att(p, typbyval, typlen);
3990 :
3991 60006 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
3992 :
3993 60006 : if (printed)
3994 35002 : appendStringInfo(&buf, "%s%s", fldsep, value);
3995 : else
3996 25004 : appendStringInfoString(&buf, value);
3997 60006 : printed = true;
3998 :
3999 60006 : p = att_addlength_pointer(p, typlen, p);
4000 60006 : p = (char *) att_align_nominal(p, typalign);
4001 : }
4002 :
4003 : /* advance bitmap pointer if any */
4004 60024 : if (bitmap)
4005 : {
4006 108 : bitmask <<= 1;
4007 108 : if (bitmask == 0x100)
4008 : {
4009 0 : bitmap++;
4010 0 : bitmask = 1;
4011 : }
4012 : }
4013 : }
4014 :
4015 25004 : result = cstring_to_text_with_len(buf.data, buf.len);
4016 25004 : pfree(buf.data);
4017 :
4018 25004 : return result;
4019 : }
4020 :
4021 : /*
4022 : * Workhorse for to_bin, to_oct, and to_hex. Note that base must be > 1 and <=
4023 : * 16.
4024 : */
4025 : static inline text *
4026 38750 : convert_to_base(uint64 value, int base)
4027 : {
4028 38750 : const char *digits = "0123456789abcdef";
4029 :
4030 : /* We size the buffer for to_bin's longest possible return value. */
4031 : char buf[sizeof(uint64) * BITS_PER_BYTE];
4032 38750 : char *const end = buf + sizeof(buf);
4033 38750 : char *ptr = end;
4034 :
4035 : Assert(base > 1);
4036 : Assert(base <= 16);
4037 :
4038 : do
4039 : {
4040 75970 : *--ptr = digits[value % base];
4041 75970 : value /= base;
4042 75970 : } while (ptr > buf && value);
4043 :
4044 38750 : return cstring_to_text_with_len(ptr, end - ptr);
4045 : }
4046 :
4047 : /*
4048 : * Convert an integer to a string containing a base-2 (binary) representation
4049 : * of the number.
4050 : */
4051 : Datum
4052 12 : to_bin32(PG_FUNCTION_ARGS)
4053 : {
4054 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4055 :
4056 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4057 : }
4058 : Datum
4059 12 : to_bin64(PG_FUNCTION_ARGS)
4060 : {
4061 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4062 :
4063 12 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4064 : }
4065 :
4066 : /*
4067 : * Convert an integer to a string containing a base-8 (oct) representation of
4068 : * the number.
4069 : */
4070 : Datum
4071 12 : to_oct32(PG_FUNCTION_ARGS)
4072 : {
4073 12 : uint64 value = (uint32) PG_GETARG_INT32(0);
4074 :
4075 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4076 : }
4077 : Datum
4078 12 : to_oct64(PG_FUNCTION_ARGS)
4079 : {
4080 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4081 :
4082 12 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4083 : }
4084 :
4085 : /*
4086 : * Convert an integer to a string containing a base-16 (hex) representation of
4087 : * the number.
4088 : */
4089 : Datum
4090 38690 : to_hex32(PG_FUNCTION_ARGS)
4091 : {
4092 38690 : uint64 value = (uint32) PG_GETARG_INT32(0);
4093 :
4094 38690 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4095 : }
4096 : Datum
4097 12 : to_hex64(PG_FUNCTION_ARGS)
4098 : {
4099 12 : uint64 value = (uint64) PG_GETARG_INT64(0);
4100 :
4101 12 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4102 : }
4103 :
4104 : /*
4105 : * Return the size of a datum, possibly compressed
4106 : *
4107 : * Works on any data type
4108 : */
4109 : Datum
4110 122 : pg_column_size(PG_FUNCTION_ARGS)
4111 : {
4112 122 : Datum value = PG_GETARG_DATUM(0);
4113 : int32 result;
4114 : int typlen;
4115 :
4116 : /* On first call, get the input type's typlen, and save at *fn_extra */
4117 122 : if (fcinfo->flinfo->fn_extra == NULL)
4118 : {
4119 : /* Lookup the datatype of the supplied argument */
4120 122 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4121 :
4122 122 : typlen = get_typlen(argtypeid);
4123 122 : if (typlen == 0) /* should not happen */
4124 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4125 :
4126 122 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4127 : sizeof(int));
4128 122 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4129 : }
4130 : else
4131 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4132 :
4133 122 : if (typlen == -1)
4134 : {
4135 : /* varlena type, possibly toasted */
4136 122 : result = toast_datum_size(value);
4137 : }
4138 0 : else if (typlen == -2)
4139 : {
4140 : /* cstring */
4141 0 : result = strlen(DatumGetCString(value)) + 1;
4142 : }
4143 : else
4144 : {
4145 : /* ordinary fixed-width type */
4146 0 : result = typlen;
4147 : }
4148 :
4149 122 : PG_RETURN_INT32(result);
4150 : }
4151 :
4152 : /*
4153 : * Return the compression method stored in the compressed attribute. Return
4154 : * NULL for non varlena type or uncompressed data.
4155 : */
4156 : Datum
4157 174 : pg_column_compression(PG_FUNCTION_ARGS)
4158 : {
4159 : int typlen;
4160 : char *result;
4161 : ToastCompressionId cmid;
4162 :
4163 : /* On first call, get the input type's typlen, and save at *fn_extra */
4164 174 : if (fcinfo->flinfo->fn_extra == NULL)
4165 : {
4166 : /* Lookup the datatype of the supplied argument */
4167 138 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4168 :
4169 138 : typlen = get_typlen(argtypeid);
4170 138 : if (typlen == 0) /* should not happen */
4171 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4172 :
4173 138 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4174 : sizeof(int));
4175 138 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4176 : }
4177 : else
4178 36 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4179 :
4180 174 : if (typlen != -1)
4181 0 : PG_RETURN_NULL();
4182 :
4183 : /* get the compression method id stored in the compressed varlena */
4184 174 : cmid = toast_get_compression_id((struct varlena *)
4185 174 : DatumGetPointer(PG_GETARG_DATUM(0)));
4186 174 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
4187 42 : PG_RETURN_NULL();
4188 :
4189 : /* convert compression method id to compression method name */
4190 132 : switch (cmid)
4191 : {
4192 66 : case TOAST_PGLZ_COMPRESSION_ID:
4193 66 : result = "pglz";
4194 66 : break;
4195 66 : case TOAST_LZ4_COMPRESSION_ID:
4196 66 : result = "lz4";
4197 66 : break;
4198 0 : default:
4199 0 : elog(ERROR, "invalid compression method id %d", cmid);
4200 : }
4201 :
4202 132 : PG_RETURN_TEXT_P(cstring_to_text(result));
4203 : }
4204 :
4205 : /*
4206 : * Return the chunk_id of the on-disk TOASTed value. Return NULL if the value
4207 : * is un-TOASTed or not on-disk.
4208 : */
4209 : Datum
4210 48 : pg_column_toast_chunk_id(PG_FUNCTION_ARGS)
4211 : {
4212 : int typlen;
4213 : struct varlena *attr;
4214 : struct varatt_external toast_pointer;
4215 :
4216 : /* On first call, get the input type's typlen, and save at *fn_extra */
4217 48 : if (fcinfo->flinfo->fn_extra == NULL)
4218 : {
4219 : /* Lookup the datatype of the supplied argument */
4220 36 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4221 :
4222 36 : typlen = get_typlen(argtypeid);
4223 36 : if (typlen == 0) /* should not happen */
4224 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4225 :
4226 36 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4227 : sizeof(int));
4228 36 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4229 : }
4230 : else
4231 12 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4232 :
4233 48 : if (typlen != -1)
4234 0 : PG_RETURN_NULL();
4235 :
4236 48 : attr = (struct varlena *) DatumGetPointer(PG_GETARG_DATUM(0));
4237 :
4238 48 : if (!VARATT_IS_EXTERNAL_ONDISK(attr))
4239 12 : PG_RETURN_NULL();
4240 :
4241 36 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
4242 :
4243 36 : PG_RETURN_OID(toast_pointer.va_valueid);
4244 : }
4245 :
4246 : /*
4247 : * string_agg - Concatenates values and returns string.
4248 : *
4249 : * Syntax: string_agg(value text, delimiter text) RETURNS text
4250 : *
4251 : * Note: Any NULL values are ignored. The first-call delimiter isn't
4252 : * actually used at all, and on subsequent calls the delimiter precedes
4253 : * the associated value.
4254 : */
4255 :
4256 : /* subroutine to initialize state */
4257 : static StringInfo
4258 2370 : makeStringAggState(FunctionCallInfo fcinfo)
4259 : {
4260 : StringInfo state;
4261 : MemoryContext aggcontext;
4262 : MemoryContext oldcontext;
4263 :
4264 2370 : if (!AggCheckCallContext(fcinfo, &aggcontext))
4265 : {
4266 : /* cannot be called directly because of internal-type argument */
4267 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
4268 : }
4269 :
4270 : /*
4271 : * Create state in aggregate context. It'll stay there across subsequent
4272 : * calls.
4273 : */
4274 2370 : oldcontext = MemoryContextSwitchTo(aggcontext);
4275 2370 : state = makeStringInfo();
4276 2370 : MemoryContextSwitchTo(oldcontext);
4277 :
4278 2370 : return state;
4279 : }
4280 :
4281 : Datum
4282 970788 : string_agg_transfn(PG_FUNCTION_ARGS)
4283 : {
4284 : StringInfo state;
4285 :
4286 970788 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4287 :
4288 : /* Append the value unless null, preceding it with the delimiter. */
4289 970788 : if (!PG_ARGISNULL(1))
4290 : {
4291 955740 : text *value = PG_GETARG_TEXT_PP(1);
4292 955740 : bool isfirst = false;
4293 :
4294 : /*
4295 : * You might think we can just throw away the first delimiter, however
4296 : * we must keep it as we may be a parallel worker doing partial
4297 : * aggregation building a state to send to the main process. We need
4298 : * to keep the delimiter of every aggregation so that the combine
4299 : * function can properly join up the strings of two separately
4300 : * partially aggregated results. The first delimiter is only stripped
4301 : * off in the final function. To know how much to strip off the front
4302 : * of the string, we store the length of the first delimiter in the
4303 : * StringInfo's cursor field, which we don't otherwise need here.
4304 : */
4305 955740 : if (state == NULL)
4306 : {
4307 2050 : state = makeStringAggState(fcinfo);
4308 2050 : isfirst = true;
4309 : }
4310 :
4311 955740 : if (!PG_ARGISNULL(2))
4312 : {
4313 955740 : text *delim = PG_GETARG_TEXT_PP(2);
4314 :
4315 955740 : appendStringInfoText(state, delim);
4316 955740 : if (isfirst)
4317 2050 : state->cursor = VARSIZE_ANY_EXHDR(delim);
4318 : }
4319 :
4320 955740 : appendStringInfoText(state, value);
4321 : }
4322 :
4323 : /*
4324 : * The transition type for string_agg() is declared to be "internal",
4325 : * which is a pass-by-value type the same size as a pointer.
4326 : */
4327 970788 : if (state)
4328 970698 : PG_RETURN_POINTER(state);
4329 90 : PG_RETURN_NULL();
4330 : }
4331 :
4332 : /*
4333 : * string_agg_combine
4334 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
4335 : */
4336 : Datum
4337 200 : string_agg_combine(PG_FUNCTION_ARGS)
4338 : {
4339 : StringInfo state1;
4340 : StringInfo state2;
4341 : MemoryContext agg_context;
4342 :
4343 200 : if (!AggCheckCallContext(fcinfo, &agg_context))
4344 0 : elog(ERROR, "aggregate function called in non-aggregate context");
4345 :
4346 200 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4347 200 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
4348 :
4349 200 : if (state2 == NULL)
4350 : {
4351 : /*
4352 : * NULL state2 is easy, just return state1, which we know is already
4353 : * in the agg_context
4354 : */
4355 0 : if (state1 == NULL)
4356 0 : PG_RETURN_NULL();
4357 0 : PG_RETURN_POINTER(state1);
4358 : }
4359 :
4360 200 : if (state1 == NULL)
4361 : {
4362 : /* We must copy state2's data into the agg_context */
4363 : MemoryContext old_context;
4364 :
4365 120 : old_context = MemoryContextSwitchTo(agg_context);
4366 120 : state1 = makeStringAggState(fcinfo);
4367 120 : appendBinaryStringInfo(state1, state2->data, state2->len);
4368 120 : state1->cursor = state2->cursor;
4369 120 : MemoryContextSwitchTo(old_context);
4370 : }
4371 80 : else if (state2->len > 0)
4372 : {
4373 : /* Combine ... state1->cursor does not change in this case */
4374 80 : appendBinaryStringInfo(state1, state2->data, state2->len);
4375 : }
4376 :
4377 200 : PG_RETURN_POINTER(state1);
4378 : }
4379 :
4380 : /*
4381 : * string_agg_serialize
4382 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
4383 : *
4384 : * This is strict, so we need not handle NULL input
4385 : */
4386 : Datum
4387 200 : string_agg_serialize(PG_FUNCTION_ARGS)
4388 : {
4389 : StringInfo state;
4390 : StringInfoData buf;
4391 : bytea *result;
4392 :
4393 : /* cannot be called directly because of internal-type argument */
4394 : Assert(AggCheckCallContext(fcinfo, NULL));
4395 :
4396 200 : state = (StringInfo) PG_GETARG_POINTER(0);
4397 :
4398 200 : pq_begintypsend(&buf);
4399 :
4400 : /* cursor */
4401 200 : pq_sendint(&buf, state->cursor, 4);
4402 :
4403 : /* data */
4404 200 : pq_sendbytes(&buf, state->data, state->len);
4405 :
4406 200 : result = pq_endtypsend(&buf);
4407 :
4408 200 : PG_RETURN_BYTEA_P(result);
4409 : }
4410 :
4411 : /*
4412 : * string_agg_deserialize
4413 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
4414 : *
4415 : * This is strict, so we need not handle NULL input
4416 : */
4417 : Datum
4418 200 : string_agg_deserialize(PG_FUNCTION_ARGS)
4419 : {
4420 : bytea *sstate;
4421 : StringInfo result;
4422 : StringInfoData buf;
4423 : char *data;
4424 : int datalen;
4425 :
4426 : /* cannot be called directly because of internal-type argument */
4427 : Assert(AggCheckCallContext(fcinfo, NULL));
4428 :
4429 200 : sstate = PG_GETARG_BYTEA_PP(0);
4430 :
4431 : /*
4432 : * Initialize a StringInfo so that we can "receive" it using the standard
4433 : * recv-function infrastructure.
4434 : */
4435 200 : initReadOnlyStringInfo(&buf, VARDATA_ANY(sstate),
4436 200 : VARSIZE_ANY_EXHDR(sstate));
4437 :
4438 200 : result = makeStringAggState(fcinfo);
4439 :
4440 : /* cursor */
4441 200 : result->cursor = pq_getmsgint(&buf, 4);
4442 :
4443 : /* data */
4444 200 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
4445 200 : data = (char *) pq_getmsgbytes(&buf, datalen);
4446 200 : appendBinaryStringInfo(result, data, datalen);
4447 :
4448 200 : pq_getmsgend(&buf);
4449 :
4450 200 : PG_RETURN_POINTER(result);
4451 : }
4452 :
4453 : Datum
4454 2094 : string_agg_finalfn(PG_FUNCTION_ARGS)
4455 : {
4456 : StringInfo state;
4457 :
4458 : /* cannot be called directly because of internal-type argument */
4459 : Assert(AggCheckCallContext(fcinfo, NULL));
4460 :
4461 2094 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4462 :
4463 2094 : if (state != NULL)
4464 : {
4465 : /* As per comment in transfn, strip data before the cursor position */
4466 2010 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
4467 : state->len - state->cursor));
4468 : }
4469 : else
4470 84 : PG_RETURN_NULL();
4471 : }
4472 :
4473 : /*
4474 : * Prepare cache with fmgr info for the output functions of the datatypes of
4475 : * the arguments of a concat-like function, beginning with argument "argidx".
4476 : * (Arguments before that will have corresponding slots in the resulting
4477 : * FmgrInfo array, but we don't fill those slots.)
4478 : */
4479 : static FmgrInfo *
4480 106 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
4481 : {
4482 : FmgrInfo *foutcache;
4483 : int i;
4484 :
4485 : /* We keep the info in fn_mcxt so it survives across calls */
4486 106 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4487 106 : PG_NARGS() * sizeof(FmgrInfo));
4488 :
4489 400 : for (i = argidx; i < PG_NARGS(); i++)
4490 : {
4491 : Oid valtype;
4492 : Oid typOutput;
4493 : bool typIsVarlena;
4494 :
4495 294 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4496 294 : if (!OidIsValid(valtype))
4497 0 : elog(ERROR, "could not determine data type of concat() input");
4498 :
4499 294 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4500 294 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
4501 : }
4502 :
4503 106 : fcinfo->flinfo->fn_extra = foutcache;
4504 :
4505 106 : return foutcache;
4506 : }
4507 :
4508 : /*
4509 : * Implementation of both concat() and concat_ws().
4510 : *
4511 : * sepstr is the separator string to place between values.
4512 : * argidx identifies the first argument to concatenate (counting from zero);
4513 : * note that this must be constant across any one series of calls.
4514 : *
4515 : * Returns NULL if result should be NULL, else text value.
4516 : */
4517 : static text *
4518 264 : concat_internal(const char *sepstr, int argidx,
4519 : FunctionCallInfo fcinfo)
4520 : {
4521 : text *result;
4522 : StringInfoData str;
4523 : FmgrInfo *foutcache;
4524 264 : bool first_arg = true;
4525 : int i;
4526 :
4527 : /*
4528 : * concat(VARIADIC some-array) is essentially equivalent to
4529 : * array_to_text(), ie concat the array elements with the given separator.
4530 : * So we just pass the case off to that code.
4531 : */
4532 264 : if (get_fn_expr_variadic(fcinfo->flinfo))
4533 : {
4534 : ArrayType *arr;
4535 :
4536 : /* Should have just the one argument */
4537 : Assert(argidx == PG_NARGS() - 1);
4538 :
4539 : /* concat(VARIADIC NULL) is defined as NULL */
4540 30 : if (PG_ARGISNULL(argidx))
4541 12 : return NULL;
4542 :
4543 : /*
4544 : * Non-null argument had better be an array. We assume that any call
4545 : * context that could let get_fn_expr_variadic return true will have
4546 : * checked that a VARIADIC-labeled parameter actually is an array. So
4547 : * it should be okay to just Assert that it's an array rather than
4548 : * doing a full-fledged error check.
4549 : */
4550 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
4551 :
4552 : /* OK, safe to fetch the array value */
4553 18 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
4554 :
4555 : /*
4556 : * And serialize the array. We tell array_to_text to ignore null
4557 : * elements, which matches the behavior of the loop below.
4558 : */
4559 18 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4560 : }
4561 :
4562 : /* Normal case without explicit VARIADIC marker */
4563 234 : initStringInfo(&str);
4564 :
4565 : /* Get output function info, building it if first time through */
4566 234 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
4567 234 : if (foutcache == NULL)
4568 106 : foutcache = build_concat_foutcache(fcinfo, argidx);
4569 :
4570 822 : for (i = argidx; i < PG_NARGS(); i++)
4571 : {
4572 588 : if (!PG_ARGISNULL(i))
4573 : {
4574 510 : Datum value = PG_GETARG_DATUM(i);
4575 :
4576 : /* add separator if appropriate */
4577 510 : if (first_arg)
4578 228 : first_arg = false;
4579 : else
4580 282 : appendStringInfoString(&str, sepstr);
4581 :
4582 : /* call the appropriate type output function, append the result */
4583 510 : appendStringInfoString(&str,
4584 510 : OutputFunctionCall(&foutcache[i], value));
4585 : }
4586 : }
4587 :
4588 234 : result = cstring_to_text_with_len(str.data, str.len);
4589 234 : pfree(str.data);
4590 :
4591 234 : return result;
4592 : }
4593 :
4594 : /*
4595 : * Concatenate all arguments. NULL arguments are ignored.
4596 : */
4597 : Datum
4598 186 : text_concat(PG_FUNCTION_ARGS)
4599 : {
4600 : text *result;
4601 :
4602 186 : result = concat_internal("", 0, fcinfo);
4603 186 : if (result == NULL)
4604 6 : PG_RETURN_NULL();
4605 180 : PG_RETURN_TEXT_P(result);
4606 : }
4607 :
4608 : /*
4609 : * Concatenate all but first argument value with separators. The first
4610 : * parameter is used as the separator. NULL arguments are ignored.
4611 : */
4612 : Datum
4613 84 : text_concat_ws(PG_FUNCTION_ARGS)
4614 : {
4615 : char *sep;
4616 : text *result;
4617 :
4618 : /* return NULL when separator is NULL */
4619 84 : if (PG_ARGISNULL(0))
4620 6 : PG_RETURN_NULL();
4621 78 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
4622 :
4623 78 : result = concat_internal(sep, 1, fcinfo);
4624 78 : if (result == NULL)
4625 6 : PG_RETURN_NULL();
4626 72 : PG_RETURN_TEXT_P(result);
4627 : }
4628 :
4629 : /*
4630 : * Return first n characters in the string. When n is negative,
4631 : * return all but last |n| characters.
4632 : */
4633 : Datum
4634 2148 : text_left(PG_FUNCTION_ARGS)
4635 : {
4636 2148 : int n = PG_GETARG_INT32(1);
4637 :
4638 2148 : if (n < 0)
4639 : {
4640 30 : text *str = PG_GETARG_TEXT_PP(0);
4641 30 : const char *p = VARDATA_ANY(str);
4642 30 : int len = VARSIZE_ANY_EXHDR(str);
4643 : int rlen;
4644 :
4645 30 : n = pg_mbstrlen_with_len(p, len) + n;
4646 30 : rlen = pg_mbcharcliplen(p, len, n);
4647 30 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
4648 : }
4649 : else
4650 2118 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
4651 : }
4652 :
4653 : /*
4654 : * Return last n characters in the string. When n is negative,
4655 : * return all but first |n| characters.
4656 : */
4657 : Datum
4658 66 : text_right(PG_FUNCTION_ARGS)
4659 : {
4660 66 : text *str = PG_GETARG_TEXT_PP(0);
4661 66 : const char *p = VARDATA_ANY(str);
4662 66 : int len = VARSIZE_ANY_EXHDR(str);
4663 66 : int n = PG_GETARG_INT32(1);
4664 : int off;
4665 :
4666 66 : if (n < 0)
4667 30 : n = -n;
4668 : else
4669 36 : n = pg_mbstrlen_with_len(p, len) - n;
4670 66 : off = pg_mbcharcliplen(p, len, n);
4671 :
4672 66 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4673 : }
4674 :
4675 : /*
4676 : * Return reversed string
4677 : */
4678 : Datum
4679 6 : text_reverse(PG_FUNCTION_ARGS)
4680 : {
4681 6 : text *str = PG_GETARG_TEXT_PP(0);
4682 6 : const char *p = VARDATA_ANY(str);
4683 6 : int len = VARSIZE_ANY_EXHDR(str);
4684 6 : const char *endp = p + len;
4685 : text *result;
4686 : char *dst;
4687 :
4688 6 : result = palloc(len + VARHDRSZ);
4689 6 : dst = (char *) VARDATA(result) + len;
4690 6 : SET_VARSIZE(result, len + VARHDRSZ);
4691 :
4692 6 : if (pg_database_encoding_max_length() > 1)
4693 : {
4694 : /* multibyte version */
4695 36 : while (p < endp)
4696 : {
4697 : int sz;
4698 :
4699 30 : sz = pg_mblen(p);
4700 30 : dst -= sz;
4701 30 : memcpy(dst, p, sz);
4702 30 : p += sz;
4703 : }
4704 : }
4705 : else
4706 : {
4707 : /* single byte version */
4708 0 : while (p < endp)
4709 0 : *(--dst) = *p++;
4710 : }
4711 :
4712 6 : PG_RETURN_TEXT_P(result);
4713 : }
4714 :
4715 :
4716 : /*
4717 : * Support macros for text_format()
4718 : */
4719 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4720 :
4721 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4722 : do { \
4723 : if (++(ptr) >= (end_ptr)) \
4724 : ereport(ERROR, \
4725 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4726 : errmsg("unterminated format() type specifier"), \
4727 : errhint("For a single \"%%\" use \"%%%%\"."))); \
4728 : } while (0)
4729 :
4730 : /*
4731 : * Returns a formatted string
4732 : */
4733 : Datum
4734 33178 : text_format(PG_FUNCTION_ARGS)
4735 : {
4736 : text *fmt;
4737 : StringInfoData str;
4738 : const char *cp;
4739 : const char *start_ptr;
4740 : const char *end_ptr;
4741 : text *result;
4742 : int arg;
4743 : bool funcvariadic;
4744 : int nargs;
4745 33178 : Datum *elements = NULL;
4746 33178 : bool *nulls = NULL;
4747 33178 : Oid element_type = InvalidOid;
4748 33178 : Oid prev_type = InvalidOid;
4749 33178 : Oid prev_width_type = InvalidOid;
4750 : FmgrInfo typoutputfinfo;
4751 : FmgrInfo typoutputinfo_width;
4752 :
4753 : /* When format string is null, immediately return null */
4754 33178 : if (PG_ARGISNULL(0))
4755 6 : PG_RETURN_NULL();
4756 :
4757 : /* If argument is marked VARIADIC, expand array into elements */
4758 33172 : if (get_fn_expr_variadic(fcinfo->flinfo))
4759 : {
4760 : ArrayType *arr;
4761 : int16 elmlen;
4762 : bool elmbyval;
4763 : char elmalign;
4764 : int nitems;
4765 :
4766 : /* Should have just the one argument */
4767 : Assert(PG_NARGS() == 2);
4768 :
4769 : /* If argument is NULL, we treat it as zero-length array */
4770 48 : if (PG_ARGISNULL(1))
4771 6 : nitems = 0;
4772 : else
4773 : {
4774 : /*
4775 : * Non-null argument had better be an array. We assume that any
4776 : * call context that could let get_fn_expr_variadic return true
4777 : * will have checked that a VARIADIC-labeled parameter actually is
4778 : * an array. So it should be okay to just Assert that it's an
4779 : * array rather than doing a full-fledged error check.
4780 : */
4781 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
4782 :
4783 : /* OK, safe to fetch the array value */
4784 42 : arr = PG_GETARG_ARRAYTYPE_P(1);
4785 :
4786 : /* Get info about array element type */
4787 42 : element_type = ARR_ELEMTYPE(arr);
4788 42 : get_typlenbyvalalign(element_type,
4789 : &elmlen, &elmbyval, &elmalign);
4790 :
4791 : /* Extract all array elements */
4792 42 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
4793 : &elements, &nulls, &nitems);
4794 : }
4795 :
4796 48 : nargs = nitems + 1;
4797 48 : funcvariadic = true;
4798 : }
4799 : else
4800 : {
4801 : /* Non-variadic case, we'll process the arguments individually */
4802 33124 : nargs = PG_NARGS();
4803 33124 : funcvariadic = false;
4804 : }
4805 :
4806 : /* Setup for main loop. */
4807 33172 : fmt = PG_GETARG_TEXT_PP(0);
4808 33172 : start_ptr = VARDATA_ANY(fmt);
4809 33172 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
4810 33172 : initStringInfo(&str);
4811 33172 : arg = 1; /* next argument position to print */
4812 :
4813 : /* Scan format string, looking for conversion specifiers. */
4814 1012540 : for (cp = start_ptr; cp < end_ptr; cp++)
4815 : {
4816 : int argpos;
4817 : int widthpos;
4818 : int flags;
4819 : int width;
4820 : Datum value;
4821 : bool isNull;
4822 : Oid typid;
4823 :
4824 : /*
4825 : * If it's not the start of a conversion specifier, just copy it to
4826 : * the output buffer.
4827 : */
4828 979428 : if (*cp != '%')
4829 : {
4830 913598 : appendStringInfoCharMacro(&str, *cp);
4831 913616 : continue;
4832 : }
4833 :
4834 65830 : ADVANCE_PARSE_POINTER(cp, end_ptr);
4835 :
4836 : /* Easy case: %% outputs a single % */
4837 65830 : if (*cp == '%')
4838 : {
4839 18 : appendStringInfoCharMacro(&str, *cp);
4840 18 : continue;
4841 : }
4842 :
4843 : /* Parse the optional portions of the format specifier */
4844 65812 : cp = text_format_parse_format(cp, end_ptr,
4845 : &argpos, &widthpos,
4846 : &flags, &width);
4847 :
4848 : /*
4849 : * Next we should see the main conversion specifier. Whether or not
4850 : * an argument position was present, it's known that at least one
4851 : * character remains in the string at this point. Experience suggests
4852 : * that it's worth checking that that character is one of the expected
4853 : * ones before we try to fetch arguments, so as to produce the least
4854 : * confusing response to a mis-formatted specifier.
4855 : */
4856 65788 : if (strchr("sIL", *cp) == NULL)
4857 6 : ereport(ERROR,
4858 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4859 : errmsg("unrecognized format() type specifier \"%.*s\"",
4860 : pg_mblen(cp), cp),
4861 : errhint("For a single \"%%\" use \"%%%%\".")));
4862 :
4863 : /* If indirect width was specified, get its value */
4864 65782 : if (widthpos >= 0)
4865 : {
4866 : /* Collect the specified or next argument position */
4867 42 : if (widthpos > 0)
4868 36 : arg = widthpos;
4869 42 : if (arg >= nargs)
4870 0 : ereport(ERROR,
4871 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4872 : errmsg("too few arguments for format()")));
4873 :
4874 : /* Get the value and type of the selected argument */
4875 42 : if (!funcvariadic)
4876 : {
4877 42 : value = PG_GETARG_DATUM(arg);
4878 42 : isNull = PG_ARGISNULL(arg);
4879 42 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4880 : }
4881 : else
4882 : {
4883 0 : value = elements[arg - 1];
4884 0 : isNull = nulls[arg - 1];
4885 0 : typid = element_type;
4886 : }
4887 42 : if (!OidIsValid(typid))
4888 0 : elog(ERROR, "could not determine data type of format() input");
4889 :
4890 42 : arg++;
4891 :
4892 : /* We can treat NULL width the same as zero */
4893 42 : if (isNull)
4894 6 : width = 0;
4895 36 : else if (typid == INT4OID)
4896 36 : width = DatumGetInt32(value);
4897 0 : else if (typid == INT2OID)
4898 0 : width = DatumGetInt16(value);
4899 : else
4900 : {
4901 : /* For less-usual datatypes, convert to text then to int */
4902 : char *str;
4903 :
4904 0 : if (typid != prev_width_type)
4905 : {
4906 : Oid typoutputfunc;
4907 : bool typIsVarlena;
4908 :
4909 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4910 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
4911 0 : prev_width_type = typid;
4912 : }
4913 :
4914 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
4915 :
4916 : /* pg_strtoint32 will complain about bad data or overflow */
4917 0 : width = pg_strtoint32(str);
4918 :
4919 0 : pfree(str);
4920 : }
4921 : }
4922 :
4923 : /* Collect the specified or next argument position */
4924 65782 : if (argpos > 0)
4925 132 : arg = argpos;
4926 65782 : if (arg >= nargs)
4927 24 : ereport(ERROR,
4928 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4929 : errmsg("too few arguments for format()")));
4930 :
4931 : /* Get the value and type of the selected argument */
4932 65758 : if (!funcvariadic)
4933 : {
4934 64486 : value = PG_GETARG_DATUM(arg);
4935 64486 : isNull = PG_ARGISNULL(arg);
4936 64486 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4937 : }
4938 : else
4939 : {
4940 1272 : value = elements[arg - 1];
4941 1272 : isNull = nulls[arg - 1];
4942 1272 : typid = element_type;
4943 : }
4944 65758 : if (!OidIsValid(typid))
4945 0 : elog(ERROR, "could not determine data type of format() input");
4946 :
4947 65758 : arg++;
4948 :
4949 : /*
4950 : * Get the appropriate typOutput function, reusing previous one if
4951 : * same type as previous argument. That's particularly useful in the
4952 : * variadic-array case, but often saves work even for ordinary calls.
4953 : */
4954 65758 : if (typid != prev_type)
4955 : {
4956 : Oid typoutputfunc;
4957 : bool typIsVarlena;
4958 :
4959 34240 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4960 34240 : fmgr_info(typoutputfunc, &typoutputfinfo);
4961 34240 : prev_type = typid;
4962 : }
4963 :
4964 : /*
4965 : * And now we can format the value.
4966 : */
4967 65758 : switch (*cp)
4968 : {
4969 65758 : case 's':
4970 : case 'I':
4971 : case 'L':
4972 65758 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
4973 : value, isNull,
4974 : flags, width);
4975 65752 : break;
4976 0 : default:
4977 : /* should not get here, because of previous check */
4978 0 : ereport(ERROR,
4979 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4980 : errmsg("unrecognized format() type specifier \"%.*s\"",
4981 : pg_mblen(cp), cp),
4982 : errhint("For a single \"%%\" use \"%%%%\".")));
4983 : break;
4984 : }
4985 : }
4986 :
4987 : /* Don't need deconstruct_array results anymore. */
4988 33112 : if (elements != NULL)
4989 42 : pfree(elements);
4990 33112 : if (nulls != NULL)
4991 42 : pfree(nulls);
4992 :
4993 : /* Generate results. */
4994 33112 : result = cstring_to_text_with_len(str.data, str.len);
4995 33112 : pfree(str.data);
4996 :
4997 33112 : PG_RETURN_TEXT_P(result);
4998 : }
4999 :
5000 : /*
5001 : * Parse contiguous digits as a decimal number.
5002 : *
5003 : * Returns true if some digits could be parsed.
5004 : * The value is returned into *value, and *ptr is advanced to the next
5005 : * character to be parsed.
5006 : *
5007 : * Note parsing invariant: at least one character is known available before
5008 : * string end (end_ptr) at entry, and this is still true at exit.
5009 : */
5010 : static bool
5011 131588 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5012 : {
5013 131588 : bool found = false;
5014 131588 : const char *cp = *ptr;
5015 131588 : int val = 0;
5016 :
5017 131900 : while (*cp >= '0' && *cp <= '9')
5018 : {
5019 318 : int8 digit = (*cp - '0');
5020 :
5021 318 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5022 318 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5023 0 : ereport(ERROR,
5024 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5025 : errmsg("number is out of range")));
5026 318 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5027 312 : found = true;
5028 : }
5029 :
5030 131582 : *ptr = cp;
5031 131582 : *value = val;
5032 :
5033 131582 : return found;
5034 : }
5035 :
5036 : /*
5037 : * Parse a format specifier (generally following the SUS printf spec).
5038 : *
5039 : * We have already advanced over the initial '%', and we are looking for
5040 : * [argpos][flags][width]type (but the type character is not consumed here).
5041 : *
5042 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5043 : * Output parameters:
5044 : * argpos: argument position for value to be printed. -1 means unspecified.
5045 : * widthpos: argument position for width. Zero means the argument position
5046 : * was unspecified (ie, take the next arg) and -1 means no width
5047 : * argument (width was omitted or specified as a constant).
5048 : * flags: bitmask of flags.
5049 : * width: directly-specified width value. Zero means the width was omitted
5050 : * (note it's not necessary to distinguish this case from an explicit
5051 : * zero width value).
5052 : *
5053 : * The function result is the next character position to be parsed, ie, the
5054 : * location where the type character is/should be.
5055 : *
5056 : * Note parsing invariant: at least one character is known available before
5057 : * string end (end_ptr) at entry, and this is still true at exit.
5058 : */
5059 : static const char *
5060 65812 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5061 : int *argpos, int *widthpos,
5062 : int *flags, int *width)
5063 : {
5064 65812 : const char *cp = start_ptr;
5065 : int n;
5066 :
5067 : /* set defaults for output parameters */
5068 65812 : *argpos = -1;
5069 65812 : *widthpos = -1;
5070 65812 : *flags = 0;
5071 65812 : *width = 0;
5072 :
5073 : /* try to identify first number */
5074 65812 : if (text_format_parse_digits(&cp, end_ptr, &n))
5075 : {
5076 174 : if (*cp != '$')
5077 : {
5078 : /* Must be just a width and a type, so we're done */
5079 24 : *width = n;
5080 24 : return cp;
5081 : }
5082 : /* The number was argument position */
5083 150 : *argpos = n;
5084 : /* Explicit 0 for argument index is immediately refused */
5085 150 : if (n == 0)
5086 6 : ereport(ERROR,
5087 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5088 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5089 144 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5090 : }
5091 :
5092 : /* Handle flags (only minus is supported now) */
5093 65806 : while (*cp == '-')
5094 : {
5095 30 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5096 30 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5097 : }
5098 :
5099 65776 : if (*cp == '*')
5100 : {
5101 : /* Handle indirect width */
5102 48 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5103 48 : if (text_format_parse_digits(&cp, end_ptr, &n))
5104 : {
5105 : /* number in this position must be closed by $ */
5106 42 : if (*cp != '$')
5107 0 : ereport(ERROR,
5108 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5109 : errmsg("width argument position must be ended by \"$\"")));
5110 : /* The number was width argument position */
5111 42 : *widthpos = n;
5112 : /* Explicit 0 for argument index is immediately refused */
5113 42 : if (n == 0)
5114 6 : ereport(ERROR,
5115 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5116 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5117 36 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5118 : }
5119 : else
5120 6 : *widthpos = 0; /* width's argument position is unspecified */
5121 : }
5122 : else
5123 : {
5124 : /* Check for direct width specification */
5125 65728 : if (text_format_parse_digits(&cp, end_ptr, &n))
5126 30 : *width = n;
5127 : }
5128 :
5129 : /* cp should now be pointing at type character */
5130 65764 : return cp;
5131 : }
5132 :
5133 : /*
5134 : * Format a %s, %I, or %L conversion
5135 : */
5136 : static void
5137 65758 : text_format_string_conversion(StringInfo buf, char conversion,
5138 : FmgrInfo *typOutputInfo,
5139 : Datum value, bool isNull,
5140 : int flags, int width)
5141 : {
5142 : char *str;
5143 :
5144 : /* Handle NULL arguments before trying to stringify the value. */
5145 65758 : if (isNull)
5146 : {
5147 342 : if (conversion == 's')
5148 270 : text_format_append_string(buf, "", flags, width);
5149 72 : else if (conversion == 'L')
5150 66 : text_format_append_string(buf, "NULL", flags, width);
5151 6 : else if (conversion == 'I')
5152 6 : ereport(ERROR,
5153 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5154 : errmsg("null values cannot be formatted as an SQL identifier")));
5155 336 : return;
5156 : }
5157 :
5158 : /* Stringify. */
5159 65416 : str = OutputFunctionCall(typOutputInfo, value);
5160 :
5161 : /* Escape. */
5162 65416 : if (conversion == 'I')
5163 : {
5164 : /* quote_identifier may or may not allocate a new string. */
5165 4902 : text_format_append_string(buf, quote_identifier(str), flags, width);
5166 : }
5167 60514 : else if (conversion == 'L')
5168 : {
5169 3232 : char *qstr = quote_literal_cstr(str);
5170 :
5171 3232 : text_format_append_string(buf, qstr, flags, width);
5172 : /* quote_literal_cstr() always allocates a new string */
5173 3232 : pfree(qstr);
5174 : }
5175 : else
5176 57282 : text_format_append_string(buf, str, flags, width);
5177 :
5178 : /* Cleanup. */
5179 65416 : pfree(str);
5180 : }
5181 :
5182 : /*
5183 : * Append str to buf, padding as directed by flags/width
5184 : */
5185 : static void
5186 65752 : text_format_append_string(StringInfo buf, const char *str,
5187 : int flags, int width)
5188 : {
5189 65752 : bool align_to_left = false;
5190 : int len;
5191 :
5192 : /* fast path for typical easy case */
5193 65752 : if (width == 0)
5194 : {
5195 65668 : appendStringInfoString(buf, str);
5196 65668 : return;
5197 : }
5198 :
5199 84 : if (width < 0)
5200 : {
5201 : /* Negative width: implicit '-' flag, then take absolute value */
5202 6 : align_to_left = true;
5203 : /* -INT_MIN is undefined */
5204 6 : if (width <= INT_MIN)
5205 0 : ereport(ERROR,
5206 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5207 : errmsg("number is out of range")));
5208 6 : width = -width;
5209 : }
5210 78 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
5211 24 : align_to_left = true;
5212 :
5213 84 : len = pg_mbstrlen(str);
5214 84 : if (align_to_left)
5215 : {
5216 : /* left justify */
5217 30 : appendStringInfoString(buf, str);
5218 30 : if (len < width)
5219 30 : appendStringInfoSpaces(buf, width - len);
5220 : }
5221 : else
5222 : {
5223 : /* right justify */
5224 54 : if (len < width)
5225 54 : appendStringInfoSpaces(buf, width - len);
5226 54 : appendStringInfoString(buf, str);
5227 : }
5228 : }
5229 :
5230 : /*
5231 : * text_format_nv - nonvariadic wrapper for text_format function.
5232 : *
5233 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5234 : * which checks that all built-in functions that share the implementing C
5235 : * function take the same number of arguments.
5236 : */
5237 : Datum
5238 3810 : text_format_nv(PG_FUNCTION_ARGS)
5239 : {
5240 3810 : return text_format(fcinfo);
5241 : }
5242 :
5243 : /*
5244 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
5245 : * for this use case.
5246 : */
5247 : static inline bool
5248 0 : rest_of_char_same(const char *s1, const char *s2, int len)
5249 : {
5250 0 : while (len > 0)
5251 : {
5252 0 : len--;
5253 0 : if (s1[len] != s2[len])
5254 0 : return false;
5255 : }
5256 0 : return true;
5257 : }
5258 :
5259 : /* Expand each Levenshtein distance variant */
5260 : #include "levenshtein.c"
5261 : #define LEVENSHTEIN_LESS_EQUAL
5262 : #include "levenshtein.c"
5263 :
5264 :
5265 : /*
5266 : * The following *ClosestMatch() functions can be used to determine whether a
5267 : * user-provided string resembles any known valid values, which is useful for
5268 : * providing hints in log messages, among other things. Use these functions
5269 : * like so:
5270 : *
5271 : * initClosestMatch(&state, source_string, max_distance);
5272 : *
5273 : * for (int i = 0; i < num_valid_strings; i++)
5274 : * updateClosestMatch(&state, valid_strings[i]);
5275 : *
5276 : * closestMatch = getClosestMatch(&state);
5277 : */
5278 :
5279 : /*
5280 : * Initialize the given state with the source string and maximum Levenshtein
5281 : * distance to consider.
5282 : */
5283 : void
5284 78 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
5285 : {
5286 : Assert(state);
5287 : Assert(max_d >= 0);
5288 :
5289 78 : state->source = source;
5290 78 : state->min_d = -1;
5291 78 : state->max_d = max_d;
5292 78 : state->match = NULL;
5293 78 : }
5294 :
5295 : /*
5296 : * If the candidate string is a closer match than the current one saved (or
5297 : * there is no match saved), save it as the closest match.
5298 : *
5299 : * If the source or candidate string is NULL, empty, or too long, this function
5300 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
5301 : * allowed or more than half the characters are different, no action is taken.
5302 : */
5303 : void
5304 804 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
5305 : {
5306 : int dist;
5307 :
5308 : Assert(state);
5309 :
5310 804 : if (state->source == NULL || state->source[0] == '\0' ||
5311 804 : candidate == NULL || candidate[0] == '\0')
5312 0 : return;
5313 :
5314 : /*
5315 : * To avoid ERROR-ing, we check the lengths here instead of setting
5316 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
5317 : */
5318 804 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
5319 804 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
5320 0 : return;
5321 :
5322 804 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
5323 804 : candidate, strlen(candidate), 1, 1, 1,
5324 : state->max_d, true);
5325 804 : if (dist <= state->max_d &&
5326 62 : dist <= strlen(state->source) / 2 &&
5327 14 : (state->min_d == -1 || dist < state->min_d))
5328 : {
5329 14 : state->min_d = dist;
5330 14 : state->match = candidate;
5331 : }
5332 : }
5333 :
5334 : /*
5335 : * Return the closest match. If no suitable candidates were provided via
5336 : * updateClosestMatch(), return NULL.
5337 : */
5338 : const char *
5339 78 : getClosestMatch(ClosestMatchState *state)
5340 : {
5341 : Assert(state);
5342 :
5343 78 : return state->match;
5344 : }
5345 :
5346 :
5347 : /*
5348 : * Unicode support
5349 : */
5350 :
5351 : static UnicodeNormalizationForm
5352 210 : unicode_norm_form_from_string(const char *formstr)
5353 : {
5354 210 : UnicodeNormalizationForm form = -1;
5355 :
5356 : /*
5357 : * Might as well check this while we're here.
5358 : */
5359 210 : if (GetDatabaseEncoding() != PG_UTF8)
5360 0 : ereport(ERROR,
5361 : (errcode(ERRCODE_SYNTAX_ERROR),
5362 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
5363 :
5364 210 : if (pg_strcasecmp(formstr, "NFC") == 0)
5365 66 : form = UNICODE_NFC;
5366 144 : else if (pg_strcasecmp(formstr, "NFD") == 0)
5367 60 : form = UNICODE_NFD;
5368 84 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
5369 36 : form = UNICODE_NFKC;
5370 48 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
5371 36 : form = UNICODE_NFKD;
5372 : else
5373 12 : ereport(ERROR,
5374 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5375 : errmsg("invalid normalization form: %s", formstr)));
5376 :
5377 198 : return form;
5378 : }
5379 :
5380 : /*
5381 : * Returns version of Unicode used by Postgres in "major.minor" format (the
5382 : * same format as the Unicode version reported by ICU). The third component
5383 : * ("update version") never involves additions to the character repertoire and
5384 : * is unimportant for most purposes.
5385 : *
5386 : * See: https://unicode.org/versions/
5387 : */
5388 : Datum
5389 32 : unicode_version(PG_FUNCTION_ARGS)
5390 : {
5391 32 : PG_RETURN_TEXT_P(cstring_to_text(PG_UNICODE_VERSION));
5392 : }
5393 :
5394 : /*
5395 : * Returns version of Unicode used by ICU, if enabled; otherwise NULL.
5396 : */
5397 : Datum
5398 2 : icu_unicode_version(PG_FUNCTION_ARGS)
5399 : {
5400 : #ifdef USE_ICU
5401 2 : PG_RETURN_TEXT_P(cstring_to_text(U_UNICODE_VERSION));
5402 : #else
5403 : PG_RETURN_NULL();
5404 : #endif
5405 : }
5406 :
5407 : /*
5408 : * Check whether the string contains only assigned Unicode code
5409 : * points. Requires that the database encoding is UTF-8.
5410 : */
5411 : Datum
5412 12 : unicode_assigned(PG_FUNCTION_ARGS)
5413 : {
5414 12 : text *input = PG_GETARG_TEXT_PP(0);
5415 : unsigned char *p;
5416 : int size;
5417 :
5418 12 : if (GetDatabaseEncoding() != PG_UTF8)
5419 0 : ereport(ERROR,
5420 : (errmsg("Unicode categorization can only be performed if server encoding is UTF8")));
5421 :
5422 : /* convert to pg_wchar */
5423 12 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5424 12 : p = (unsigned char *) VARDATA_ANY(input);
5425 48 : for (int i = 0; i < size; i++)
5426 : {
5427 42 : pg_wchar uchar = utf8_to_unicode(p);
5428 42 : int category = unicode_category(uchar);
5429 :
5430 42 : if (category == PG_U_UNASSIGNED)
5431 6 : PG_RETURN_BOOL(false);
5432 :
5433 36 : p += pg_utf_mblen(p);
5434 : }
5435 :
5436 6 : PG_RETURN_BOOL(true);
5437 : }
5438 :
5439 : Datum
5440 72 : unicode_normalize_func(PG_FUNCTION_ARGS)
5441 : {
5442 72 : text *input = PG_GETARG_TEXT_PP(0);
5443 72 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5444 : UnicodeNormalizationForm form;
5445 : int size;
5446 : pg_wchar *input_chars;
5447 : pg_wchar *output_chars;
5448 : unsigned char *p;
5449 : text *result;
5450 : int i;
5451 :
5452 72 : form = unicode_norm_form_from_string(formstr);
5453 :
5454 : /* convert to pg_wchar */
5455 66 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5456 66 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
5457 66 : p = (unsigned char *) VARDATA_ANY(input);
5458 288 : for (i = 0; i < size; i++)
5459 : {
5460 222 : input_chars[i] = utf8_to_unicode(p);
5461 222 : p += pg_utf_mblen(p);
5462 : }
5463 66 : input_chars[i] = (pg_wchar) '\0';
5464 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5465 :
5466 : /* action */
5467 66 : output_chars = unicode_normalize(form, input_chars);
5468 :
5469 : /* convert back to UTF-8 string */
5470 66 : size = 0;
5471 306 : for (pg_wchar *wp = output_chars; *wp; wp++)
5472 : {
5473 : unsigned char buf[4];
5474 :
5475 240 : unicode_to_utf8(*wp, buf);
5476 240 : size += pg_utf_mblen(buf);
5477 : }
5478 :
5479 66 : result = palloc(size + VARHDRSZ);
5480 66 : SET_VARSIZE(result, size + VARHDRSZ);
5481 :
5482 66 : p = (unsigned char *) VARDATA_ANY(result);
5483 306 : for (pg_wchar *wp = output_chars; *wp; wp++)
5484 : {
5485 240 : unicode_to_utf8(*wp, p);
5486 240 : p += pg_utf_mblen(p);
5487 : }
5488 : Assert((char *) p == (char *) result + size + VARHDRSZ);
5489 :
5490 66 : PG_RETURN_TEXT_P(result);
5491 : }
5492 :
5493 : /*
5494 : * Check whether the string is in the specified Unicode normalization form.
5495 : *
5496 : * This is done by converting the string to the specified normal form and then
5497 : * comparing that to the original string. To speed that up, we also apply the
5498 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
5499 : * answer for many strings by just scanning the string once.
5500 : *
5501 : * This function should generally be optimized for the case where the string
5502 : * is in fact normalized. In that case, we'll end up looking at the entire
5503 : * string, so it's probably not worth doing any incremental conversion etc.
5504 : */
5505 : Datum
5506 138 : unicode_is_normalized(PG_FUNCTION_ARGS)
5507 : {
5508 138 : text *input = PG_GETARG_TEXT_PP(0);
5509 138 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5510 : UnicodeNormalizationForm form;
5511 : int size;
5512 : pg_wchar *input_chars;
5513 : pg_wchar *output_chars;
5514 : unsigned char *p;
5515 : int i;
5516 : UnicodeNormalizationQC quickcheck;
5517 : int output_size;
5518 : bool result;
5519 :
5520 138 : form = unicode_norm_form_from_string(formstr);
5521 :
5522 : /* convert to pg_wchar */
5523 132 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5524 132 : input_chars = palloc((size + 1) * sizeof(pg_wchar));
5525 132 : p = (unsigned char *) VARDATA_ANY(input);
5526 504 : for (i = 0; i < size; i++)
5527 : {
5528 372 : input_chars[i] = utf8_to_unicode(p);
5529 372 : p += pg_utf_mblen(p);
5530 : }
5531 132 : input_chars[i] = (pg_wchar) '\0';
5532 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5533 :
5534 : /* quick check (see UAX #15) */
5535 132 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
5536 132 : if (quickcheck == UNICODE_NORM_QC_YES)
5537 42 : PG_RETURN_BOOL(true);
5538 90 : else if (quickcheck == UNICODE_NORM_QC_NO)
5539 12 : PG_RETURN_BOOL(false);
5540 :
5541 : /* normalize and compare with original */
5542 78 : output_chars = unicode_normalize(form, input_chars);
5543 :
5544 78 : output_size = 0;
5545 324 : for (pg_wchar *wp = output_chars; *wp; wp++)
5546 246 : output_size++;
5547 :
5548 114 : result = (size == output_size) &&
5549 36 : (memcmp(input_chars, output_chars, size * sizeof(pg_wchar)) == 0);
5550 :
5551 78 : PG_RETURN_BOOL(result);
5552 : }
5553 :
5554 : /*
5555 : * Check if first n chars are hexadecimal digits
5556 : */
5557 : static bool
5558 156 : isxdigits_n(const char *instr, size_t n)
5559 : {
5560 660 : for (size_t i = 0; i < n; i++)
5561 570 : if (!isxdigit((unsigned char) instr[i]))
5562 66 : return false;
5563 :
5564 90 : return true;
5565 : }
5566 :
5567 : static unsigned int
5568 504 : hexval(unsigned char c)
5569 : {
5570 504 : if (c >= '0' && c <= '9')
5571 384 : return c - '0';
5572 120 : if (c >= 'a' && c <= 'f')
5573 60 : return c - 'a' + 0xA;
5574 60 : if (c >= 'A' && c <= 'F')
5575 60 : return c - 'A' + 0xA;
5576 0 : elog(ERROR, "invalid hexadecimal digit");
5577 : return 0; /* not reached */
5578 : }
5579 :
5580 : /*
5581 : * Translate string with hexadecimal digits to number
5582 : */
5583 : static unsigned int
5584 90 : hexval_n(const char *instr, size_t n)
5585 : {
5586 90 : unsigned int result = 0;
5587 :
5588 594 : for (size_t i = 0; i < n; i++)
5589 504 : result += hexval(instr[i]) << (4 * (n - i - 1));
5590 :
5591 90 : return result;
5592 : }
5593 :
5594 : /*
5595 : * Replaces Unicode escape sequences by Unicode characters
5596 : */
5597 : Datum
5598 66 : unistr(PG_FUNCTION_ARGS)
5599 : {
5600 66 : text *input_text = PG_GETARG_TEXT_PP(0);
5601 : char *instr;
5602 : int len;
5603 : StringInfoData str;
5604 : text *result;
5605 66 : pg_wchar pair_first = 0;
5606 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
5607 :
5608 66 : instr = VARDATA_ANY(input_text);
5609 66 : len = VARSIZE_ANY_EXHDR(input_text);
5610 :
5611 66 : initStringInfo(&str);
5612 :
5613 510 : while (len > 0)
5614 : {
5615 486 : if (instr[0] == '\\')
5616 : {
5617 102 : if (len >= 2 &&
5618 102 : instr[1] == '\\')
5619 : {
5620 6 : if (pair_first)
5621 0 : goto invalid_pair;
5622 6 : appendStringInfoChar(&str, '\\');
5623 6 : instr += 2;
5624 6 : len -= 2;
5625 : }
5626 96 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
5627 66 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
5628 30 : {
5629 : pg_wchar unicode;
5630 42 : int offset = instr[1] == 'u' ? 2 : 1;
5631 :
5632 42 : unicode = hexval_n(instr + offset, 4);
5633 :
5634 42 : if (!is_valid_unicode_codepoint(unicode))
5635 0 : ereport(ERROR,
5636 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5637 : errmsg("invalid Unicode code point: %04X", unicode));
5638 :
5639 42 : if (pair_first)
5640 : {
5641 12 : if (is_utf16_surrogate_second(unicode))
5642 : {
5643 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5644 0 : pair_first = 0;
5645 : }
5646 : else
5647 12 : goto invalid_pair;
5648 : }
5649 30 : else if (is_utf16_surrogate_second(unicode))
5650 0 : goto invalid_pair;
5651 :
5652 30 : if (is_utf16_surrogate_first(unicode))
5653 18 : pair_first = unicode;
5654 : else
5655 : {
5656 12 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5657 12 : appendStringInfoString(&str, cbuf);
5658 : }
5659 :
5660 30 : instr += 4 + offset;
5661 30 : len -= 4 + offset;
5662 : }
5663 54 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
5664 12 : {
5665 : pg_wchar unicode;
5666 :
5667 24 : unicode = hexval_n(instr + 2, 6);
5668 :
5669 24 : if (!is_valid_unicode_codepoint(unicode))
5670 6 : ereport(ERROR,
5671 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5672 : errmsg("invalid Unicode code point: %04X", unicode));
5673 :
5674 18 : if (pair_first)
5675 : {
5676 6 : if (is_utf16_surrogate_second(unicode))
5677 : {
5678 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5679 0 : pair_first = 0;
5680 : }
5681 : else
5682 6 : goto invalid_pair;
5683 : }
5684 12 : else if (is_utf16_surrogate_second(unicode))
5685 0 : goto invalid_pair;
5686 :
5687 12 : if (is_utf16_surrogate_first(unicode))
5688 6 : pair_first = unicode;
5689 : else
5690 : {
5691 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5692 6 : appendStringInfoString(&str, cbuf);
5693 : }
5694 :
5695 12 : instr += 8;
5696 12 : len -= 8;
5697 : }
5698 30 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
5699 12 : {
5700 : pg_wchar unicode;
5701 :
5702 24 : unicode = hexval_n(instr + 2, 8);
5703 :
5704 24 : if (!is_valid_unicode_codepoint(unicode))
5705 6 : ereport(ERROR,
5706 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5707 : errmsg("invalid Unicode code point: %04X", unicode));
5708 :
5709 18 : if (pair_first)
5710 : {
5711 6 : if (is_utf16_surrogate_second(unicode))
5712 : {
5713 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5714 0 : pair_first = 0;
5715 : }
5716 : else
5717 6 : goto invalid_pair;
5718 : }
5719 12 : else if (is_utf16_surrogate_second(unicode))
5720 0 : goto invalid_pair;
5721 :
5722 12 : if (is_utf16_surrogate_first(unicode))
5723 6 : pair_first = unicode;
5724 : else
5725 : {
5726 6 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5727 6 : appendStringInfoString(&str, cbuf);
5728 : }
5729 :
5730 12 : instr += 10;
5731 12 : len -= 10;
5732 : }
5733 : else
5734 6 : ereport(ERROR,
5735 : (errcode(ERRCODE_SYNTAX_ERROR),
5736 : errmsg("invalid Unicode escape"),
5737 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
5738 : }
5739 : else
5740 : {
5741 384 : if (pair_first)
5742 0 : goto invalid_pair;
5743 :
5744 384 : appendStringInfoChar(&str, *instr++);
5745 384 : len--;
5746 : }
5747 : }
5748 :
5749 : /* unfinished surrogate pair? */
5750 24 : if (pair_first)
5751 6 : goto invalid_pair;
5752 :
5753 18 : result = cstring_to_text_with_len(str.data, str.len);
5754 18 : pfree(str.data);
5755 :
5756 18 : PG_RETURN_TEXT_P(result);
5757 :
5758 30 : invalid_pair:
5759 30 : ereport(ERROR,
5760 : (errcode(ERRCODE_SYNTAX_ERROR),
5761 : errmsg("invalid Unicode surrogate pair")));
5762 : PG_RETURN_NULL(); /* keep compiler quiet */
5763 : }
|
