Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * varlena.c
4 : * Functions for the variable-length built-in types.
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/adt/varlena.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <ctype.h>
18 : #include <limits.h>
19 :
20 : #include "access/detoast.h"
21 : #include "access/toast_compression.h"
22 : #include "access/tupmacs.h"
23 : #include "catalog/pg_collation.h"
24 : #include "catalog/pg_type.h"
25 : #include "common/hashfn.h"
26 : #include "common/int.h"
27 : #include "common/unicode_category.h"
28 : #include "common/unicode_norm.h"
29 : #include "common/unicode_version.h"
30 : #include "funcapi.h"
31 : #include "lib/hyperloglog.h"
32 : #include "libpq/pqformat.h"
33 : #include "miscadmin.h"
34 : #include "nodes/execnodes.h"
35 : #include "parser/scansup.h"
36 : #include "port/pg_bswap.h"
37 : #include "regex/regex.h"
38 : #include "utils/builtins.h"
39 : #include "utils/guc.h"
40 : #include "utils/lsyscache.h"
41 : #include "utils/memutils.h"
42 : #include "utils/pg_locale.h"
43 : #include "utils/sortsupport.h"
44 : #include "utils/tuplestore.h"
45 : #include "utils/varlena.h"
46 :
47 : typedef varlena VarString;
48 :
49 : /*
50 : * State for text_position_* functions.
51 : */
52 : typedef struct
53 : {
54 : pg_locale_t locale; /* collation used for substring matching */
55 : bool is_multibyte_char_in_char; /* need to check char boundaries? */
56 : bool greedy; /* find longest possible substring? */
57 :
58 : char *str1; /* haystack string */
59 : char *str2; /* needle string */
60 : int len1; /* string lengths in bytes */
61 : int len2;
62 :
63 : /* Skip table for Boyer-Moore-Horspool search algorithm: */
64 : int skiptablemask; /* mask for ANDing with skiptable subscripts */
65 : int skiptable[256]; /* skip distance for given mismatched char */
66 :
67 : /*
68 : * Note that with nondeterministic collations, the length of the last
69 : * match is not necessarily equal to the length of the "needle" passed in.
70 : */
71 : char *last_match; /* pointer to last match in 'str1' */
72 : int last_match_len; /* length of last match */
73 : int last_match_len_tmp; /* same but for internal use */
74 :
75 : /*
76 : * Sometimes we need to convert the byte position of a match to a
77 : * character position. These store the last position that was converted,
78 : * so that on the next call, we can continue from that point, rather than
79 : * count characters from the very beginning.
80 : */
81 : char *refpoint; /* pointer within original haystack string */
82 : int refpos; /* 0-based character offset of the same point */
83 : } TextPositionState;
84 :
85 : typedef struct
86 : {
87 : char *buf1; /* 1st string, or abbreviation original string
88 : * buf */
89 : char *buf2; /* 2nd string, or abbreviation strxfrm() buf */
90 : int buflen1; /* Allocated length of buf1 */
91 : int buflen2; /* Allocated length of buf2 */
92 : int last_len1; /* Length of last buf1 string/strxfrm() input */
93 : int last_len2; /* Length of last buf2 string/strxfrm() blob */
94 : int last_returned; /* Last comparison result (cache) */
95 : bool cache_blob; /* Does buf2 contain strxfrm() blob, etc? */
96 : bool collate_c;
97 : Oid typid; /* Actual datatype (text/bpchar/name) */
98 : hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
99 : hyperLogLogState full_card; /* Full key cardinality state */
100 : double prop_card; /* Required cardinality proportion */
101 : pg_locale_t locale;
102 : } VarStringSortSupport;
103 :
104 : /*
105 : * Output data for split_text(): we output either to an array or a table.
106 : * tupstore and tupdesc must be set up in advance to output to a table.
107 : */
108 : typedef struct
109 : {
110 : ArrayBuildState *astate;
111 : Tuplestorestate *tupstore;
112 : TupleDesc tupdesc;
113 : } SplitTextOutputData;
114 :
115 : /*
116 : * This should be large enough that most strings will fit, but small enough
117 : * that we feel comfortable putting it on the stack
118 : */
119 : #define TEXTBUFLEN 1024
120 :
121 : #define DatumGetVarStringP(X) ((VarString *) PG_DETOAST_DATUM(X))
122 : #define DatumGetVarStringPP(X) ((VarString *) PG_DETOAST_DATUM_PACKED(X))
123 :
124 : static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup);
125 : static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup);
126 : static int namefastcmp_c(Datum x, Datum y, SortSupport ssup);
127 : static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup);
128 : static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup);
129 : static int varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup);
130 : static Datum varstr_abbrev_convert(Datum original, SortSupport ssup);
131 : static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup);
132 : static int32 text_length(Datum str);
133 : static text *text_catenate(text *t1, text *t2);
134 : static text *text_substring(Datum str,
135 : int32 start,
136 : int32 length,
137 : bool length_not_specified);
138 : static int pg_mbcharcliplen_chars(const char *mbstr, int len, int limit);
139 : static text *text_overlay(text *t1, text *t2, int sp, int sl);
140 : static int text_position(text *t1, text *t2, Oid collid);
141 : static void text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state);
142 : static bool text_position_next(TextPositionState *state);
143 : static char *text_position_next_internal(char *start_ptr, TextPositionState *state);
144 : static char *text_position_get_match_ptr(TextPositionState *state);
145 : static int text_position_get_match_pos(TextPositionState *state);
146 : static void text_position_cleanup(TextPositionState *state);
147 : static void check_collation_set(Oid collid);
148 : static int text_cmp(text *arg1, text *arg2, Oid collid);
149 : static void appendStringInfoText(StringInfo str, const text *t);
150 : static bool split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate);
151 : static void split_text_accum_result(SplitTextOutputData *tstate,
152 : text *field_value,
153 : text *null_string,
154 : Oid collation);
155 : static text *array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
156 : const char *fldsep, const char *null_string);
157 : static StringInfo makeStringAggState(FunctionCallInfo fcinfo);
158 : static bool text_format_parse_digits(const char **ptr, const char *end_ptr,
159 : int *value);
160 : static const char *text_format_parse_format(const char *start_ptr,
161 : const char *end_ptr,
162 : int *argpos, int *widthpos,
163 : int *flags, int *width);
164 : static void text_format_string_conversion(StringInfo buf, char conversion,
165 : FmgrInfo *typOutputInfo,
166 : Datum value, bool isNull,
167 : int flags, int width);
168 : static void text_format_append_string(StringInfo buf, const char *str,
169 : int flags, int width);
170 :
171 :
172 : /*****************************************************************************
173 : * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
174 : *****************************************************************************/
175 :
176 : /*
177 : * cstring_to_text
178 : *
179 : * Create a text value from a null-terminated C string.
180 : *
181 : * The new text value is freshly palloc'd with a full-size VARHDR.
182 : */
183 : text *
184 14111514 : cstring_to_text(const char *s)
185 : {
186 14111514 : return cstring_to_text_with_len(s, strlen(s));
187 : }
188 :
189 : /*
190 : * cstring_to_text_with_len
191 : *
192 : * Same as cstring_to_text except the caller specifies the string length;
193 : * the string need not be null_terminated.
194 : */
195 : text *
196 15641490 : cstring_to_text_with_len(const char *s, int len)
197 : {
198 15641490 : text *result = (text *) palloc(len + VARHDRSZ);
199 :
200 15641490 : SET_VARSIZE(result, len + VARHDRSZ);
201 15641490 : memcpy(VARDATA(result), s, len);
202 :
203 15641490 : return result;
204 : }
205 :
206 : /*
207 : * text_to_cstring
208 : *
209 : * Create a palloc'd, null-terminated C string from a text value.
210 : *
211 : * We support being passed a compressed or toasted text value.
212 : * This is a bit bogus since such values shouldn't really be referred to as
213 : * "text *", but it seems useful for robustness. If we didn't handle that
214 : * case here, we'd need another routine that did, anyway.
215 : */
216 : char *
217 10473558 : text_to_cstring(const text *t)
218 : {
219 : /* must cast away the const, unfortunately */
220 10473558 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
221 10473558 : int len = VARSIZE_ANY_EXHDR(tunpacked);
222 : char *result;
223 :
224 10473558 : result = (char *) palloc(len + 1);
225 10473558 : memcpy(result, VARDATA_ANY(tunpacked), len);
226 10473558 : result[len] = '\0';
227 :
228 10473558 : if (tunpacked != t)
229 27902 : pfree(tunpacked);
230 :
231 10473558 : return result;
232 : }
233 :
234 : /*
235 : * text_to_cstring_buffer
236 : *
237 : * Copy a text value into a caller-supplied buffer of size dst_len.
238 : *
239 : * The text string is truncated if necessary to fit. The result is
240 : * guaranteed null-terminated (unless dst_len == 0).
241 : *
242 : * We support being passed a compressed or toasted text value.
243 : * This is a bit bogus since such values shouldn't really be referred to as
244 : * "text *", but it seems useful for robustness. If we didn't handle that
245 : * case here, we'd need another routine that did, anyway.
246 : */
247 : void
248 727 : text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
249 : {
250 : /* must cast away the const, unfortunately */
251 727 : text *srcunpacked = pg_detoast_datum_packed(unconstify(text *, src));
252 727 : size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
253 :
254 727 : if (dst_len > 0)
255 : {
256 727 : dst_len--;
257 727 : if (dst_len >= src_len)
258 727 : dst_len = src_len;
259 : else /* ensure truncation is encoding-safe */
260 0 : dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
261 727 : memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
262 727 : dst[dst_len] = '\0';
263 : }
264 :
265 727 : if (srcunpacked != src)
266 0 : pfree(srcunpacked);
267 727 : }
268 :
269 :
270 : /*****************************************************************************
271 : * USER I/O ROUTINES *
272 : *****************************************************************************/
273 :
274 : /*
275 : * textin - converts cstring to internal representation
276 : */
277 : Datum
278 12109415 : textin(PG_FUNCTION_ARGS)
279 : {
280 12109415 : char *inputText = PG_GETARG_CSTRING(0);
281 :
282 12109415 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
283 : }
284 :
285 : /*
286 : * textout - converts internal representation to cstring
287 : */
288 : Datum
289 4801392 : textout(PG_FUNCTION_ARGS)
290 : {
291 4801392 : Datum txt = PG_GETARG_DATUM(0);
292 :
293 4801392 : PG_RETURN_CSTRING(TextDatumGetCString(txt));
294 : }
295 :
296 : /*
297 : * textrecv - converts external binary format to text
298 : */
299 : Datum
300 27 : textrecv(PG_FUNCTION_ARGS)
301 : {
302 27 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
303 : text *result;
304 : char *str;
305 : int nbytes;
306 :
307 27 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
308 :
309 27 : result = cstring_to_text_with_len(str, nbytes);
310 27 : pfree(str);
311 27 : PG_RETURN_TEXT_P(result);
312 : }
313 :
314 : /*
315 : * textsend - converts text to binary format
316 : */
317 : Datum
318 2396 : textsend(PG_FUNCTION_ARGS)
319 : {
320 2396 : text *t = PG_GETARG_TEXT_PP(0);
321 : StringInfoData buf;
322 :
323 2396 : pq_begintypsend(&buf);
324 2396 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
325 2396 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
326 : }
327 :
328 :
329 : /*
330 : * unknownin - converts cstring to internal representation
331 : */
332 : Datum
333 0 : unknownin(PG_FUNCTION_ARGS)
334 : {
335 0 : char *str = PG_GETARG_CSTRING(0);
336 :
337 : /* representation is same as cstring */
338 0 : PG_RETURN_CSTRING(pstrdup(str));
339 : }
340 :
341 : /*
342 : * unknownout - converts internal representation to cstring
343 : */
344 : Datum
345 619 : unknownout(PG_FUNCTION_ARGS)
346 : {
347 : /* representation is same as cstring */
348 619 : char *str = PG_GETARG_CSTRING(0);
349 :
350 619 : PG_RETURN_CSTRING(pstrdup(str));
351 : }
352 :
353 : /*
354 : * unknownrecv - converts external binary format to unknown
355 : */
356 : Datum
357 0 : unknownrecv(PG_FUNCTION_ARGS)
358 : {
359 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
360 : char *str;
361 : int nbytes;
362 :
363 0 : str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
364 : /* representation is same as cstring */
365 0 : PG_RETURN_CSTRING(str);
366 : }
367 :
368 : /*
369 : * unknownsend - converts unknown to binary format
370 : */
371 : Datum
372 0 : unknownsend(PG_FUNCTION_ARGS)
373 : {
374 : /* representation is same as cstring */
375 0 : char *str = PG_GETARG_CSTRING(0);
376 : StringInfoData buf;
377 :
378 0 : pq_begintypsend(&buf);
379 0 : pq_sendtext(&buf, str, strlen(str));
380 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
381 : }
382 :
383 :
384 : /* ========== PUBLIC ROUTINES ========== */
385 :
386 : /*
387 : * textlen -
388 : * returns the logical length of a text*
389 : * (which is less than the VARSIZE of the text*)
390 : */
391 : Datum
392 286144 : textlen(PG_FUNCTION_ARGS)
393 : {
394 286144 : Datum str = PG_GETARG_DATUM(0);
395 :
396 : /* try to avoid decompressing argument */
397 286144 : PG_RETURN_INT32(text_length(str));
398 : }
399 :
400 : /*
401 : * text_length -
402 : * Does the real work for textlen()
403 : *
404 : * This is broken out so it can be called directly by other string processing
405 : * functions. Note that the argument is passed as a Datum, to indicate that
406 : * it may still be in compressed form. We can avoid decompressing it at all
407 : * in some cases.
408 : */
409 : static int32
410 286154 : text_length(Datum str)
411 : {
412 : /* fastpath when max encoding length is one */
413 286154 : if (pg_database_encoding_max_length() == 1)
414 10 : return (toast_raw_datum_size(str) - VARHDRSZ);
415 : else
416 : {
417 286144 : text *t = DatumGetTextPP(str);
418 :
419 286144 : return (pg_mbstrlen_with_len(VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t)));
420 : }
421 : }
422 :
423 : /*
424 : * textoctetlen -
425 : * returns the physical length of a text*
426 : * (which is less than the VARSIZE of the text*)
427 : */
428 : Datum
429 45 : textoctetlen(PG_FUNCTION_ARGS)
430 : {
431 45 : Datum str = PG_GETARG_DATUM(0);
432 :
433 : /* We need not detoast the input at all */
434 45 : PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
435 : }
436 :
437 : /*
438 : * textcat -
439 : * takes two text* and returns a text* that is the concatenation of
440 : * the two.
441 : *
442 : * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
443 : * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
444 : * Allocate space for output in all cases.
445 : * XXX - thomas 1997-07-10
446 : */
447 : Datum
448 1307221 : textcat(PG_FUNCTION_ARGS)
449 : {
450 1307221 : text *t1 = PG_GETARG_TEXT_PP(0);
451 1307221 : text *t2 = PG_GETARG_TEXT_PP(1);
452 :
453 1307221 : PG_RETURN_TEXT_P(text_catenate(t1, t2));
454 : }
455 :
456 : /*
457 : * text_catenate
458 : * Guts of textcat(), broken out so it can be used by other functions
459 : *
460 : * Arguments can be in short-header form, but not compressed or out-of-line
461 : */
462 : static text *
463 1307277 : text_catenate(text *t1, text *t2)
464 : {
465 : text *result;
466 : int len1,
467 : len2,
468 : len;
469 : char *ptr;
470 :
471 1307277 : len1 = VARSIZE_ANY_EXHDR(t1);
472 1307277 : len2 = VARSIZE_ANY_EXHDR(t2);
473 :
474 : /* paranoia ... probably should throw error instead? */
475 1307277 : if (len1 < 0)
476 0 : len1 = 0;
477 1307277 : if (len2 < 0)
478 0 : len2 = 0;
479 :
480 1307277 : len = len1 + len2 + VARHDRSZ;
481 1307277 : result = (text *) palloc(len);
482 :
483 : /* Set size of result string... */
484 1307277 : SET_VARSIZE(result, len);
485 :
486 : /* Fill data field of result string... */
487 1307277 : ptr = VARDATA(result);
488 1307277 : if (len1 > 0)
489 1306747 : memcpy(ptr, VARDATA_ANY(t1), len1);
490 1307277 : if (len2 > 0)
491 1307138 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
492 :
493 1307277 : return result;
494 : }
495 :
496 : /*
497 : * charlen_to_bytelen()
498 : * Compute the number of bytes occupied by n characters starting at *p
499 : *
500 : * The caller shall ensure there are n complete characters. Callers achieve
501 : * this by deriving "n" from regmatch_t findings from searching a wchar array.
502 : * pg_mb2wchar_with_len() skips any trailing incomplete character, so regex
503 : * matches will end no later than the last complete character. (The string
504 : * need not be null-terminated.)
505 : */
506 : static int
507 11233 : charlen_to_bytelen(const char *p, int n)
508 : {
509 11233 : if (pg_database_encoding_max_length() == 1)
510 : {
511 : /* Optimization for single-byte encodings */
512 90 : return n;
513 : }
514 : else
515 : {
516 : const char *s;
517 :
518 3076502 : for (s = p; n > 0; n--)
519 3065359 : s += pg_mblen_unbounded(s); /* caller verified encoding */
520 :
521 11143 : return s - p;
522 : }
523 : }
524 :
525 : /*
526 : * text_substr()
527 : * Return a substring starting at the specified position.
528 : * - thomas 1997-12-31
529 : *
530 : * Input:
531 : * - string
532 : * - starting position (is one-based)
533 : * - string length
534 : *
535 : * If the starting position is zero or less, then return from the start of the string
536 : * adjusting the length to be consistent with the "negative start" per SQL.
537 : * If the length is less than zero, return the remaining string.
538 : *
539 : * Added multibyte support.
540 : * - Tatsuo Ishii 1998-4-21
541 : * Changed behavior if starting position is less than one to conform to SQL behavior.
542 : * Formerly returned the entire string; now returns a portion.
543 : * - Thomas Lockhart 1998-12-10
544 : * Now uses faster TOAST-slicing interface
545 : * - John Gray 2002-02-22
546 : * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
547 : * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
548 : * error; if E < 1, return '', not entire string). Fixed MB related bug when
549 : * S > LC and < LC + 4 sometimes garbage characters are returned.
550 : * - Joe Conway 2002-08-10
551 : */
552 : Datum
553 368764 : text_substr(PG_FUNCTION_ARGS)
554 : {
555 368764 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
556 : PG_GETARG_INT32(1),
557 : PG_GETARG_INT32(2),
558 : false));
559 : }
560 :
561 : /*
562 : * text_substr_no_len -
563 : * Wrapper to avoid opr_sanity failure due to
564 : * one function accepting a different number of args.
565 : */
566 : Datum
567 24 : text_substr_no_len(PG_FUNCTION_ARGS)
568 : {
569 24 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0),
570 : PG_GETARG_INT32(1),
571 : -1, true));
572 : }
573 :
574 : /*
575 : * text_substring -
576 : * Does the real work for text_substr() and text_substr_no_len()
577 : *
578 : * This is broken out so it can be called directly by other string processing
579 : * functions. Note that the argument is passed as a Datum, to indicate that
580 : * it may still be in compressed/toasted form. We can avoid detoasting all
581 : * of it in some cases.
582 : *
583 : * The result is always a freshly palloc'd datum.
584 : */
585 : static text *
586 395532 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
587 : {
588 395532 : int32 eml = pg_database_encoding_max_length();
589 395532 : int32 S = start; /* start position */
590 : int32 S1; /* adjusted start position */
591 : int32 L1; /* adjusted substring length */
592 : int32 E; /* end position, exclusive */
593 :
594 : /*
595 : * SQL99 says S can be zero or negative (which we don't document), but we
596 : * still must fetch from the start of the string.
597 : * https://www.postgresql.org/message-id/170905442373.643.11536838320909376197%40wrigleys.postgresql.org
598 : */
599 395532 : S1 = Max(S, 1);
600 :
601 : /* life is easy if the encoding max length is 1 */
602 395532 : if (eml == 1)
603 : {
604 11 : if (length_not_specified) /* special case - get length to end of
605 : * string */
606 0 : L1 = -1;
607 11 : else if (length < 0)
608 : {
609 : /* SQL99 says to throw an error for E < S, i.e., negative length */
610 0 : ereport(ERROR,
611 : (errcode(ERRCODE_SUBSTRING_ERROR),
612 : errmsg("negative substring length not allowed")));
613 : L1 = -1; /* silence stupider compilers */
614 : }
615 11 : else if (pg_add_s32_overflow(S, length, &E))
616 : {
617 : /*
618 : * L could be large enough for S + L to overflow, in which case
619 : * the substring must run to end of string.
620 : */
621 0 : L1 = -1;
622 : }
623 : else
624 : {
625 : /*
626 : * A zero or negative value for the end position can happen if the
627 : * start was negative or one. SQL99 says to return a zero-length
628 : * string.
629 : */
630 11 : if (E < 1)
631 0 : return cstring_to_text("");
632 :
633 11 : L1 = E - S1;
634 : }
635 :
636 : /*
637 : * If the start position is past the end of the string, SQL99 says to
638 : * return a zero-length string -- DatumGetTextPSlice() will do that
639 : * for us. We need only convert S1 to zero-based starting position.
640 : */
641 11 : return DatumGetTextPSlice(str, S1 - 1, L1);
642 : }
643 395521 : else if (eml > 1)
644 : {
645 : /*
646 : * When encoding max length is > 1, we can't get LC without
647 : * detoasting, so we'll grab a conservatively large slice now and go
648 : * back later to do the right thing
649 : */
650 : int32 slice_start;
651 : int32 slice_size;
652 : int32 slice_strlen;
653 : int32 slice_len;
654 : text *slice;
655 : int32 E1;
656 : int32 i;
657 : char *p;
658 : char *s;
659 : text *ret;
660 :
661 : /*
662 : * We need to start at position zero because there is no way to know
663 : * in advance which byte offset corresponds to the supplied start
664 : * position.
665 : */
666 395521 : slice_start = 0;
667 :
668 395521 : if (length_not_specified) /* special case - get length to end of
669 : * string */
670 52 : E = slice_size = L1 = -1;
671 395469 : else if (length < 0)
672 : {
673 : /* SQL99 says to throw an error for E < S, i.e., negative length */
674 8 : ereport(ERROR,
675 : (errcode(ERRCODE_SUBSTRING_ERROR),
676 : errmsg("negative substring length not allowed")));
677 : E = slice_size = L1 = -1; /* silence stupider compilers */
678 : }
679 395461 : else if (pg_add_s32_overflow(S, length, &E))
680 : {
681 : /*
682 : * L could be large enough for S + L to overflow, in which case
683 : * the substring must run to end of string.
684 : */
685 5 : slice_size = L1 = -1;
686 : }
687 : else
688 : {
689 : /*
690 : * Ending at position 1, exclusive, obviously yields an empty
691 : * string. A zero or negative value can happen if the start was
692 : * negative or one. SQL99 says to return a zero-length string.
693 : */
694 395456 : if (E <= 1)
695 8 : return cstring_to_text("");
696 :
697 : /*
698 : * if E is past the end of the string, the tuple toaster will
699 : * truncate the length for us
700 : */
701 395448 : L1 = E - S1;
702 :
703 : /*
704 : * Total slice size in bytes can't be any longer than the
705 : * inclusive end position times the encoding max length. If that
706 : * overflows, we can just use -1.
707 : */
708 395448 : if (pg_mul_s32_overflow(E - 1, eml, &slice_size))
709 5 : slice_size = -1;
710 : }
711 :
712 : /*
713 : * If we're working with an untoasted source, no need to do an extra
714 : * copying step.
715 : */
716 790922 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
717 395417 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
718 240 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
719 : else
720 395265 : slice = (text *) DatumGetPointer(str);
721 :
722 : /* see if we got back an empty string */
723 395505 : slice_len = VARSIZE_ANY_EXHDR(slice);
724 395505 : if (slice_len == 0)
725 : {
726 0 : if (slice != (text *) DatumGetPointer(str))
727 0 : pfree(slice);
728 0 : return cstring_to_text("");
729 : }
730 :
731 : /*
732 : * Now we can get the actual length of the slice in MB characters,
733 : * stopping at the end of the substring. Continuing beyond the
734 : * substring end could find an incomplete character attributable
735 : * solely to DatumGetTextPSlice() chopping in the middle of a
736 : * character, and it would be superfluous work at best.
737 : */
738 395497 : slice_strlen =
739 395505 : (slice_size == -1 ?
740 395505 : pg_mbstrlen_with_len(VARDATA_ANY(slice), slice_len) :
741 395443 : pg_mbcharcliplen_chars(VARDATA_ANY(slice), slice_len, E - 1));
742 :
743 : /*
744 : * Check that the start position wasn't > slice_strlen. If so, SQL99
745 : * says to return a zero-length string.
746 : */
747 395497 : if (S1 > slice_strlen)
748 : {
749 25 : if (slice != (text *) DatumGetPointer(str))
750 4 : pfree(slice);
751 25 : return cstring_to_text("");
752 : }
753 :
754 : /*
755 : * Adjust L1 and E1 now that we know the slice string length. Again
756 : * remember that S1 is one based, and slice_start is zero based.
757 : */
758 395472 : if (L1 > -1)
759 395428 : E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
760 : else
761 44 : E1 = slice_start + 1 + slice_strlen;
762 :
763 : /*
764 : * Find the start position in the slice; remember S1 is not zero based
765 : */
766 395472 : p = VARDATA_ANY(slice);
767 4238358 : for (i = 0; i < S1 - 1; i++)
768 3842886 : p += pg_mblen_unbounded(p);
769 :
770 : /* hang onto a pointer to our start position */
771 395472 : s = p;
772 :
773 : /*
774 : * Count the actual bytes used by the substring of the requested
775 : * length.
776 : */
777 6229185 : for (i = S1; i < E1; i++)
778 5833713 : p += pg_mblen_unbounded(p);
779 :
780 395472 : ret = (text *) palloc(VARHDRSZ + (p - s));
781 395472 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
782 395472 : memcpy(VARDATA(ret), s, (p - s));
783 :
784 395472 : if (slice != (text *) DatumGetPointer(str))
785 232 : pfree(slice);
786 :
787 395472 : return ret;
788 : }
789 : else
790 0 : elog(ERROR, "invalid backend encoding: encoding max length < 1");
791 :
792 : /* not reached: suppress compiler warning */
793 : return NULL;
794 : }
795 :
796 : /*
797 : * pg_mbcharcliplen_chars -
798 : * Mirror pg_mbcharcliplen(), except return value unit is chars, not bytes.
799 : *
800 : * This mirrors all the dubious historical behavior, so it's static to
801 : * discourage proliferation. The assertions are specific to the one caller.
802 : */
803 : static int
804 395443 : pg_mbcharcliplen_chars(const char *mbstr, int len, int limit)
805 : {
806 395443 : int nch = 0;
807 : int l;
808 :
809 : Assert(len > 0);
810 : Assert(limit > 0);
811 : Assert(pg_database_encoding_max_length() > 1);
812 :
813 8108759 : while (len > 0 && *mbstr)
814 : {
815 8108424 : l = pg_mblen_with_len(mbstr, len);
816 8108416 : nch++;
817 8108416 : if (nch == limit)
818 395100 : break;
819 7713316 : len -= l;
820 7713316 : mbstr += l;
821 : }
822 395435 : return nch;
823 : }
824 :
825 : /*
826 : * textoverlay
827 : * Replace specified substring of first string with second
828 : *
829 : * The SQL standard defines OVERLAY() in terms of substring and concatenation.
830 : * This code is a direct implementation of what the standard says.
831 : */
832 : Datum
833 18 : textoverlay(PG_FUNCTION_ARGS)
834 : {
835 18 : text *t1 = PG_GETARG_TEXT_PP(0);
836 18 : text *t2 = PG_GETARG_TEXT_PP(1);
837 18 : int sp = PG_GETARG_INT32(2); /* substring start position */
838 18 : int sl = PG_GETARG_INT32(3); /* substring length */
839 :
840 18 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
841 : }
842 :
843 : Datum
844 10 : textoverlay_no_len(PG_FUNCTION_ARGS)
845 : {
846 10 : text *t1 = PG_GETARG_TEXT_PP(0);
847 10 : text *t2 = PG_GETARG_TEXT_PP(1);
848 10 : int sp = PG_GETARG_INT32(2); /* substring start position */
849 : int sl;
850 :
851 10 : sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
852 10 : PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
853 : }
854 :
855 : static text *
856 28 : text_overlay(text *t1, text *t2, int sp, int sl)
857 : {
858 : text *result;
859 : text *s1;
860 : text *s2;
861 : int sp_pl_sl;
862 :
863 : /*
864 : * Check for possible integer-overflow cases. For negative sp, throw a
865 : * "substring length" error because that's what should be expected
866 : * according to the spec's definition of OVERLAY().
867 : */
868 28 : if (sp <= 0)
869 0 : ereport(ERROR,
870 : (errcode(ERRCODE_SUBSTRING_ERROR),
871 : errmsg("negative substring length not allowed")));
872 28 : if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
873 0 : ereport(ERROR,
874 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
875 : errmsg("integer out of range")));
876 :
877 28 : s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
878 28 : s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
879 28 : result = text_catenate(s1, t2);
880 28 : result = text_catenate(result, s2);
881 :
882 28 : return result;
883 : }
884 :
885 : /*
886 : * textpos -
887 : * Return the position of the specified substring.
888 : * Implements the SQL POSITION() function.
889 : * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
890 : * - thomas 1997-07-27
891 : */
892 : Datum
893 87 : textpos(PG_FUNCTION_ARGS)
894 : {
895 87 : text *str = PG_GETARG_TEXT_PP(0);
896 87 : text *search_str = PG_GETARG_TEXT_PP(1);
897 :
898 87 : PG_RETURN_INT32((int32) text_position(str, search_str, PG_GET_COLLATION()));
899 : }
900 :
901 : /*
902 : * text_position -
903 : * Does the real work for textpos()
904 : *
905 : * Inputs:
906 : * t1 - string to be searched
907 : * t2 - pattern to match within t1
908 : * Result:
909 : * Character index of the first matched char, starting from 1,
910 : * or 0 if no match.
911 : *
912 : * This is broken out so it can be called directly by other string processing
913 : * functions.
914 : */
915 : static int
916 87 : text_position(text *t1, text *t2, Oid collid)
917 : {
918 : TextPositionState state;
919 : int result;
920 :
921 87 : check_collation_set(collid);
922 :
923 : /* Empty needle always matches at position 1 */
924 87 : if (VARSIZE_ANY_EXHDR(t2) < 1)
925 10 : return 1;
926 :
927 : /* Otherwise, can't match if haystack is shorter than needle */
928 77 : if (VARSIZE_ANY_EXHDR(t1) < VARSIZE_ANY_EXHDR(t2) &&
929 13 : pg_newlocale_from_collation(collid)->deterministic)
930 13 : return 0;
931 :
932 64 : text_position_setup(t1, t2, collid, &state);
933 : /* don't need greedy mode here */
934 64 : state.greedy = false;
935 :
936 64 : if (!text_position_next(&state))
937 14 : result = 0;
938 : else
939 50 : result = text_position_get_match_pos(&state);
940 64 : text_position_cleanup(&state);
941 64 : return result;
942 : }
943 :
944 :
945 : /*
946 : * text_position_setup, text_position_next, text_position_cleanup -
947 : * Component steps of text_position()
948 : *
949 : * These are broken out so that a string can be efficiently searched for
950 : * multiple occurrences of the same pattern. text_position_next may be
951 : * called multiple times, and it advances to the next match on each call.
952 : * text_position_get_match_ptr() and text_position_get_match_pos() return
953 : * a pointer or 1-based character position of the last match, respectively.
954 : *
955 : * The "state" variable is normally just a local variable in the caller.
956 : *
957 : * NOTE: text_position_next skips over the matched portion. For example,
958 : * searching for "xx" in "xxx" returns only one match, not two.
959 : */
960 :
961 : static void
962 1087 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
963 : {
964 1087 : int len1 = VARSIZE_ANY_EXHDR(t1);
965 1087 : int len2 = VARSIZE_ANY_EXHDR(t2);
966 :
967 1087 : check_collation_set(collid);
968 :
969 1087 : state->locale = pg_newlocale_from_collation(collid);
970 :
971 : /*
972 : * Most callers need greedy mode, but some might want to unset this to
973 : * optimize.
974 : */
975 1087 : state->greedy = true;
976 :
977 : Assert(len2 > 0);
978 :
979 : /*
980 : * Even with a multi-byte encoding, we perform the search using the raw
981 : * byte sequence, ignoring multibyte issues. For UTF-8, that works fine,
982 : * because in UTF-8 the byte sequence of one character cannot contain
983 : * another character. For other multi-byte encodings, we do the search
984 : * initially as a simple byte search, ignoring multibyte issues, but
985 : * verify afterwards that the match we found is at a character boundary,
986 : * and continue the search if it was a false match.
987 : */
988 1087 : if (pg_database_encoding_max_length() == 1)
989 54 : state->is_multibyte_char_in_char = false;
990 1033 : else if (GetDatabaseEncoding() == PG_UTF8)
991 1033 : state->is_multibyte_char_in_char = false;
992 : else
993 0 : state->is_multibyte_char_in_char = true;
994 :
995 1087 : state->str1 = VARDATA_ANY(t1);
996 1087 : state->str2 = VARDATA_ANY(t2);
997 1087 : state->len1 = len1;
998 1087 : state->len2 = len2;
999 1087 : state->last_match = NULL;
1000 1087 : state->refpoint = state->str1;
1001 1087 : state->refpos = 0;
1002 :
1003 : /*
1004 : * Prepare the skip table for Boyer-Moore-Horspool searching. In these
1005 : * notes we use the terminology that the "haystack" is the string to be
1006 : * searched (t1) and the "needle" is the pattern being sought (t2).
1007 : *
1008 : * If the needle is empty or bigger than the haystack then there is no
1009 : * point in wasting cycles initializing the table. We also choose not to
1010 : * use B-M-H for needles of length 1, since the skip table can't possibly
1011 : * save anything in that case.
1012 : *
1013 : * (With nondeterministic collations, the search is already
1014 : * multibyte-aware, so we don't need this.)
1015 : */
1016 1087 : if (len1 >= len2 && len2 > 1 && state->locale->deterministic)
1017 : {
1018 849 : int searchlength = len1 - len2;
1019 : int skiptablemask;
1020 : int last;
1021 : int i;
1022 849 : const char *str2 = state->str2;
1023 :
1024 : /*
1025 : * First we must determine how much of the skip table to use. The
1026 : * declaration of TextPositionState allows up to 256 elements, but for
1027 : * short search problems we don't really want to have to initialize so
1028 : * many elements --- it would take too long in comparison to the
1029 : * actual search time. So we choose a useful skip table size based on
1030 : * the haystack length minus the needle length. The closer the needle
1031 : * length is to the haystack length the less useful skipping becomes.
1032 : *
1033 : * Note: since we use bit-masking to select table elements, the skip
1034 : * table size MUST be a power of 2, and so the mask must be 2^N-1.
1035 : */
1036 849 : if (searchlength < 16)
1037 80 : skiptablemask = 3;
1038 769 : else if (searchlength < 64)
1039 24 : skiptablemask = 7;
1040 745 : else if (searchlength < 128)
1041 15 : skiptablemask = 15;
1042 730 : else if (searchlength < 512)
1043 182 : skiptablemask = 31;
1044 548 : else if (searchlength < 2048)
1045 406 : skiptablemask = 63;
1046 142 : else if (searchlength < 4096)
1047 100 : skiptablemask = 127;
1048 : else
1049 42 : skiptablemask = 255;
1050 849 : state->skiptablemask = skiptablemask;
1051 :
1052 : /*
1053 : * Initialize the skip table. We set all elements to the needle
1054 : * length, since this is the correct skip distance for any character
1055 : * not found in the needle.
1056 : */
1057 56961 : for (i = 0; i <= skiptablemask; i++)
1058 56112 : state->skiptable[i] = len2;
1059 :
1060 : /*
1061 : * Now examine the needle. For each character except the last one,
1062 : * set the corresponding table element to the appropriate skip
1063 : * distance. Note that when two characters share the same skip table
1064 : * entry, the one later in the needle must determine the skip
1065 : * distance.
1066 : */
1067 849 : last = len2 - 1;
1068 :
1069 10426 : for (i = 0; i < last; i++)
1070 9577 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1071 : }
1072 1087 : }
1073 :
1074 : /*
1075 : * Advance to the next match, starting from the end of the previous match
1076 : * (or the beginning of the string, on first call). Returns true if a match
1077 : * is found.
1078 : *
1079 : * Note that this refuses to match an empty-string needle. Most callers
1080 : * will have handled that case specially and we'll never see it here.
1081 : */
1082 : static bool
1083 5237 : text_position_next(TextPositionState *state)
1084 : {
1085 5237 : int needle_len = state->len2;
1086 : char *start_ptr;
1087 : char *matchptr;
1088 :
1089 5237 : if (needle_len <= 0)
1090 0 : return false; /* result for empty pattern */
1091 :
1092 : /* Start from the point right after the previous match. */
1093 5237 : if (state->last_match)
1094 4140 : start_ptr = state->last_match + state->last_match_len;
1095 : else
1096 1097 : start_ptr = state->str1;
1097 :
1098 5237 : retry:
1099 5237 : matchptr = text_position_next_internal(start_ptr, state);
1100 :
1101 5237 : if (!matchptr)
1102 1027 : return false;
1103 :
1104 : /*
1105 : * Found a match for the byte sequence. If this is a multibyte encoding,
1106 : * where one character's byte sequence can appear inside a longer
1107 : * multi-byte character, we need to verify that the match was at a
1108 : * character boundary, not in the middle of a multi-byte character.
1109 : */
1110 4210 : if (state->is_multibyte_char_in_char && state->locale->deterministic)
1111 : {
1112 0 : const char *haystack_end = state->str1 + state->len1;
1113 :
1114 : /* Walk one character at a time, until we reach the match. */
1115 :
1116 : /* the search should never move backwards. */
1117 : Assert(state->refpoint <= matchptr);
1118 :
1119 0 : while (state->refpoint < matchptr)
1120 : {
1121 : /* step to next character. */
1122 0 : state->refpoint += pg_mblen_range(state->refpoint, haystack_end);
1123 0 : state->refpos++;
1124 :
1125 : /*
1126 : * If we stepped over the match's start position, then it was a
1127 : * false positive, where the byte sequence appeared in the middle
1128 : * of a multi-byte character. Skip it, and continue the search at
1129 : * the next character boundary.
1130 : */
1131 0 : if (state->refpoint > matchptr)
1132 : {
1133 0 : start_ptr = state->refpoint;
1134 0 : goto retry;
1135 : }
1136 : }
1137 : }
1138 :
1139 4210 : state->last_match = matchptr;
1140 4210 : state->last_match_len = state->last_match_len_tmp;
1141 4210 : return true;
1142 : }
1143 :
1144 : /*
1145 : * Subroutine of text_position_next(). This searches for the raw byte
1146 : * sequence, ignoring any multi-byte encoding issues. Returns the first
1147 : * match starting at 'start_ptr', or NULL if no match is found.
1148 : */
1149 : static char *
1150 5237 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1151 : {
1152 5237 : int haystack_len = state->len1;
1153 5237 : int needle_len = state->len2;
1154 5237 : int skiptablemask = state->skiptablemask;
1155 5237 : const char *haystack = state->str1;
1156 5237 : const char *needle = state->str2;
1157 5237 : const char *haystack_end = &haystack[haystack_len];
1158 : const char *hptr;
1159 :
1160 : Assert(start_ptr >= haystack && start_ptr <= haystack_end);
1161 : Assert(needle_len > 0);
1162 :
1163 5237 : state->last_match_len_tmp = needle_len;
1164 :
1165 5237 : if (!state->locale->deterministic)
1166 : {
1167 : /*
1168 : * With a nondeterministic collation, we have to use an unoptimized
1169 : * route. We walk through the haystack and see if at each position
1170 : * there is a substring of the remaining string that is equal to the
1171 : * needle under the given collation.
1172 : *
1173 : * Note, the found substring could have a different length than the
1174 : * needle. Callers that want to skip over the found string need to
1175 : * read the length of the found substring from last_match_len rather
1176 : * than just using the length of their needle.
1177 : *
1178 : * Most callers will require "greedy" semantics, meaning that we need
1179 : * to find the longest such substring, not the shortest. For callers
1180 : * that don't need greedy semantics, we can finish on the first match.
1181 : *
1182 : * This loop depends on the assumption that the needle is nonempty and
1183 : * any matching substring must also be nonempty. (Even if the
1184 : * collation would accept an empty match, returning one would send
1185 : * callers that search for successive matches into an infinite loop.)
1186 : */
1187 176 : const char *result_hptr = NULL;
1188 :
1189 176 : hptr = start_ptr;
1190 482 : while (hptr < haystack_end)
1191 : {
1192 : const char *test_end;
1193 :
1194 : /*
1195 : * First check the common case that there is a match in the
1196 : * haystack of exactly the length of the needle.
1197 : */
1198 403 : if (!state->greedy &&
1199 72 : haystack_end - hptr >= needle_len &&
1200 36 : pg_strncoll(hptr, needle_len, needle, needle_len, state->locale) == 0)
1201 8 : return (char *) hptr;
1202 :
1203 : /*
1204 : * Else check if any of the non-empty substrings starting at hptr
1205 : * compare equal to the needle.
1206 : */
1207 395 : test_end = hptr;
1208 : do
1209 : {
1210 1583 : test_end += pg_mblen_range(test_end, haystack_end);
1211 1583 : if (pg_strncoll(hptr, (test_end - hptr), needle, needle_len, state->locale) == 0)
1212 : {
1213 97 : state->last_match_len_tmp = (test_end - hptr);
1214 97 : result_hptr = hptr;
1215 97 : if (!state->greedy)
1216 0 : break;
1217 : }
1218 1583 : } while (test_end < haystack_end);
1219 :
1220 395 : if (result_hptr)
1221 89 : break;
1222 :
1223 306 : hptr += pg_mblen_range(hptr, haystack_end);
1224 : }
1225 :
1226 168 : return (char *) result_hptr;
1227 : }
1228 5061 : else if (needle_len == 1)
1229 : {
1230 : /* No point in using B-M-H for a one-character needle */
1231 553 : char nchar = *needle;
1232 :
1233 553 : hptr = start_ptr;
1234 4110 : while (hptr < haystack_end)
1235 : {
1236 3988 : if (*hptr == nchar)
1237 431 : return (char *) hptr;
1238 3557 : hptr++;
1239 : }
1240 : }
1241 : else
1242 : {
1243 4508 : const char *needle_last = &needle[needle_len - 1];
1244 :
1245 : /* Start at startpos plus the length of the needle */
1246 4508 : hptr = start_ptr + needle_len - 1;
1247 110090 : while (hptr < haystack_end)
1248 : {
1249 : /* Match the needle scanning *backward* */
1250 : const char *nptr;
1251 : const char *p;
1252 :
1253 109264 : nptr = needle_last;
1254 109264 : p = hptr;
1255 163933 : while (*nptr == *p)
1256 : {
1257 : /* Matched it all? If so, return 1-based position */
1258 58351 : if (nptr == needle)
1259 3682 : return (char *) p;
1260 54669 : nptr--, p--;
1261 : }
1262 :
1263 : /*
1264 : * No match, so use the haystack char at hptr to decide how far to
1265 : * advance. If the needle had any occurrence of that character
1266 : * (or more precisely, one sharing the same skiptable entry)
1267 : * before its last character, then we advance far enough to align
1268 : * the last such needle character with that haystack position.
1269 : * Otherwise we can advance by the whole needle length.
1270 : */
1271 105582 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1272 : }
1273 : }
1274 :
1275 948 : return 0; /* not found */
1276 : }
1277 :
1278 : /*
1279 : * Return a pointer to the current match.
1280 : *
1281 : * The returned pointer points into the original haystack string.
1282 : */
1283 : static char *
1284 4135 : text_position_get_match_ptr(TextPositionState *state)
1285 : {
1286 4135 : return state->last_match;
1287 : }
1288 :
1289 : /*
1290 : * Return the offset of the current match.
1291 : *
1292 : * The offset is in characters, 1-based.
1293 : */
1294 : static int
1295 50 : text_position_get_match_pos(TextPositionState *state)
1296 : {
1297 : /* Convert the byte position to char position. */
1298 100 : state->refpos += pg_mbstrlen_with_len(state->refpoint,
1299 50 : state->last_match - state->refpoint);
1300 50 : state->refpoint = state->last_match;
1301 50 : return state->refpos + 1;
1302 : }
1303 :
1304 : /*
1305 : * Reset search state to the initial state installed by text_position_setup.
1306 : *
1307 : * The next call to text_position_next will search from the beginning
1308 : * of the string.
1309 : */
1310 : static void
1311 10 : text_position_reset(TextPositionState *state)
1312 : {
1313 10 : state->last_match = NULL;
1314 10 : state->refpoint = state->str1;
1315 10 : state->refpos = 0;
1316 10 : }
1317 :
1318 : static void
1319 1087 : text_position_cleanup(TextPositionState *state)
1320 : {
1321 : /* no cleanup needed */
1322 1087 : }
1323 :
1324 :
1325 : static void
1326 10388581 : check_collation_set(Oid collid)
1327 : {
1328 10388581 : if (!OidIsValid(collid))
1329 : {
1330 : /*
1331 : * This typically means that the parser could not resolve a conflict
1332 : * of implicit collations, so report it that way.
1333 : */
1334 20 : ereport(ERROR,
1335 : (errcode(ERRCODE_INDETERMINATE_COLLATION),
1336 : errmsg("could not determine which collation to use for string comparison"),
1337 : errhint("Use the COLLATE clause to set the collation explicitly.")));
1338 : }
1339 10388561 : }
1340 :
1341 : /*
1342 : * varstr_cmp()
1343 : *
1344 : * Comparison function for text strings with given lengths, using the
1345 : * appropriate locale. Returns an integer less than, equal to, or greater than
1346 : * zero, indicating whether arg1 is less than, equal to, or greater than arg2.
1347 : *
1348 : * Note: many functions that depend on this are marked leakproof; therefore,
1349 : * avoid reporting the actual contents of the input when throwing errors.
1350 : * All errors herein should be things that can't happen except on corrupt
1351 : * data, anyway; otherwise we will have trouble with indexing strings that
1352 : * would cause them.
1353 : */
1354 : int
1355 5874063 : varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
1356 : {
1357 : int result;
1358 : pg_locale_t mylocale;
1359 :
1360 5874063 : check_collation_set(collid);
1361 :
1362 5874051 : mylocale = pg_newlocale_from_collation(collid);
1363 :
1364 5874051 : if (mylocale->collate_is_c)
1365 : {
1366 2051053 : result = memcmp(arg1, arg2, Min(len1, len2));
1367 2051053 : if ((result == 0) && (len1 != len2))
1368 83876 : result = (len1 < len2) ? -1 : 1;
1369 : }
1370 : else
1371 : {
1372 : /*
1373 : * memcmp() can't tell us which of two unequal strings sorts first,
1374 : * but it's a cheap way to tell if they're equal. Testing shows that
1375 : * memcmp() followed by strcoll() is only trivially slower than
1376 : * strcoll() by itself, so we don't lose much if this doesn't work out
1377 : * very often, and if it does - for example, because there are many
1378 : * equal strings in the input - then we win big by avoiding expensive
1379 : * collation-aware comparisons.
1380 : */
1381 3822998 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1382 1058529 : return 0;
1383 :
1384 2764469 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1385 :
1386 : /* Break tie if necessary. */
1387 2764469 : if (result == 0 && mylocale->deterministic)
1388 : {
1389 0 : result = memcmp(arg1, arg2, Min(len1, len2));
1390 0 : if ((result == 0) && (len1 != len2))
1391 0 : result = (len1 < len2) ? -1 : 1;
1392 : }
1393 : }
1394 :
1395 4815522 : return result;
1396 : }
1397 :
1398 : /* text_cmp()
1399 : * Internal comparison function for text strings.
1400 : * Returns -1, 0 or 1
1401 : */
1402 : static int
1403 4676648 : text_cmp(text *arg1, text *arg2, Oid collid)
1404 : {
1405 : char *a1p,
1406 : *a2p;
1407 : int len1,
1408 : len2;
1409 :
1410 4676648 : a1p = VARDATA_ANY(arg1);
1411 4676648 : a2p = VARDATA_ANY(arg2);
1412 :
1413 4676648 : len1 = VARSIZE_ANY_EXHDR(arg1);
1414 4676648 : len2 = VARSIZE_ANY_EXHDR(arg2);
1415 :
1416 4676648 : return varstr_cmp(a1p, len1, a2p, len2, collid);
1417 : }
1418 :
1419 : /*
1420 : * Comparison functions for text strings.
1421 : *
1422 : * Note: btree indexes need these routines not to leak memory; therefore,
1423 : * be careful to free working copies of toasted datums. Most places don't
1424 : * need to be so careful.
1425 : */
1426 :
1427 : Datum
1428 4038528 : texteq(PG_FUNCTION_ARGS)
1429 : {
1430 4038528 : Oid collid = PG_GET_COLLATION();
1431 4038528 : pg_locale_t mylocale = 0;
1432 : bool result;
1433 :
1434 4038528 : check_collation_set(collid);
1435 :
1436 4038528 : mylocale = pg_newlocale_from_collation(collid);
1437 :
1438 4038528 : if (mylocale->deterministic)
1439 : {
1440 4035554 : Datum arg1 = PG_GETARG_DATUM(0);
1441 4035554 : Datum arg2 = PG_GETARG_DATUM(1);
1442 : Size len1,
1443 : len2;
1444 :
1445 : /*
1446 : * Since we only care about equality or not-equality, we can avoid all
1447 : * the expense of strcoll() here, and just do bitwise comparison. In
1448 : * fact, we don't even have to do a bitwise comparison if we can show
1449 : * the lengths of the strings are unequal; which might save us from
1450 : * having to detoast one or both values.
1451 : */
1452 4035554 : len1 = toast_raw_datum_size(arg1);
1453 4035554 : len2 = toast_raw_datum_size(arg2);
1454 4035554 : if (len1 != len2)
1455 1918630 : result = false;
1456 : else
1457 : {
1458 2116924 : text *targ1 = DatumGetTextPP(arg1);
1459 2116924 : text *targ2 = DatumGetTextPP(arg2);
1460 :
1461 2116924 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1462 : len1 - VARHDRSZ) == 0);
1463 :
1464 2116924 : PG_FREE_IF_COPY(targ1, 0);
1465 2116924 : PG_FREE_IF_COPY(targ2, 1);
1466 : }
1467 : }
1468 : else
1469 : {
1470 2974 : text *arg1 = PG_GETARG_TEXT_PP(0);
1471 2974 : text *arg2 = PG_GETARG_TEXT_PP(1);
1472 :
1473 2974 : result = (text_cmp(arg1, arg2, collid) == 0);
1474 :
1475 2974 : PG_FREE_IF_COPY(arg1, 0);
1476 2974 : PG_FREE_IF_COPY(arg2, 1);
1477 : }
1478 :
1479 4038528 : PG_RETURN_BOOL(result);
1480 : }
1481 :
1482 : Datum
1483 207766 : textne(PG_FUNCTION_ARGS)
1484 : {
1485 207766 : Oid collid = PG_GET_COLLATION();
1486 : pg_locale_t mylocale;
1487 : bool result;
1488 :
1489 207766 : check_collation_set(collid);
1490 :
1491 207766 : mylocale = pg_newlocale_from_collation(collid);
1492 :
1493 207766 : if (mylocale->deterministic)
1494 : {
1495 207750 : Datum arg1 = PG_GETARG_DATUM(0);
1496 207750 : Datum arg2 = PG_GETARG_DATUM(1);
1497 : Size len1,
1498 : len2;
1499 :
1500 : /* See comment in texteq() */
1501 207750 : len1 = toast_raw_datum_size(arg1);
1502 207750 : len2 = toast_raw_datum_size(arg2);
1503 207750 : if (len1 != len2)
1504 11681 : result = true;
1505 : else
1506 : {
1507 196069 : text *targ1 = DatumGetTextPP(arg1);
1508 196069 : text *targ2 = DatumGetTextPP(arg2);
1509 :
1510 196069 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1511 : len1 - VARHDRSZ) != 0);
1512 :
1513 196069 : PG_FREE_IF_COPY(targ1, 0);
1514 196069 : PG_FREE_IF_COPY(targ2, 1);
1515 : }
1516 : }
1517 : else
1518 : {
1519 16 : text *arg1 = PG_GETARG_TEXT_PP(0);
1520 16 : text *arg2 = PG_GETARG_TEXT_PP(1);
1521 :
1522 16 : result = (text_cmp(arg1, arg2, collid) != 0);
1523 :
1524 16 : PG_FREE_IF_COPY(arg1, 0);
1525 16 : PG_FREE_IF_COPY(arg2, 1);
1526 : }
1527 :
1528 207766 : PG_RETURN_BOOL(result);
1529 : }
1530 :
1531 : Datum
1532 231433 : text_lt(PG_FUNCTION_ARGS)
1533 : {
1534 231433 : text *arg1 = PG_GETARG_TEXT_PP(0);
1535 231433 : text *arg2 = PG_GETARG_TEXT_PP(1);
1536 : bool result;
1537 :
1538 231433 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1539 :
1540 231421 : PG_FREE_IF_COPY(arg1, 0);
1541 231421 : PG_FREE_IF_COPY(arg2, 1);
1542 :
1543 231421 : PG_RETURN_BOOL(result);
1544 : }
1545 :
1546 : Datum
1547 213625 : text_le(PG_FUNCTION_ARGS)
1548 : {
1549 213625 : text *arg1 = PG_GETARG_TEXT_PP(0);
1550 213625 : text *arg2 = PG_GETARG_TEXT_PP(1);
1551 : bool result;
1552 :
1553 213625 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1554 :
1555 213625 : PG_FREE_IF_COPY(arg1, 0);
1556 213625 : PG_FREE_IF_COPY(arg2, 1);
1557 :
1558 213625 : PG_RETURN_BOOL(result);
1559 : }
1560 :
1561 : Datum
1562 216151 : text_gt(PG_FUNCTION_ARGS)
1563 : {
1564 216151 : text *arg1 = PG_GETARG_TEXT_PP(0);
1565 216151 : text *arg2 = PG_GETARG_TEXT_PP(1);
1566 : bool result;
1567 :
1568 216151 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1569 :
1570 216151 : PG_FREE_IF_COPY(arg1, 0);
1571 216151 : PG_FREE_IF_COPY(arg2, 1);
1572 :
1573 216151 : PG_RETURN_BOOL(result);
1574 : }
1575 :
1576 : Datum
1577 116415 : text_ge(PG_FUNCTION_ARGS)
1578 : {
1579 116415 : text *arg1 = PG_GETARG_TEXT_PP(0);
1580 116415 : text *arg2 = PG_GETARG_TEXT_PP(1);
1581 : bool result;
1582 :
1583 116415 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1584 :
1585 116415 : PG_FREE_IF_COPY(arg1, 0);
1586 116415 : PG_FREE_IF_COPY(arg2, 1);
1587 :
1588 116415 : PG_RETURN_BOOL(result);
1589 : }
1590 :
1591 : Datum
1592 25276 : text_starts_with(PG_FUNCTION_ARGS)
1593 : {
1594 25276 : Datum arg1 = PG_GETARG_DATUM(0);
1595 25276 : Datum arg2 = PG_GETARG_DATUM(1);
1596 25276 : Oid collid = PG_GET_COLLATION();
1597 : pg_locale_t mylocale;
1598 : bool result;
1599 : Size len1,
1600 : len2;
1601 :
1602 25276 : check_collation_set(collid);
1603 :
1604 25276 : mylocale = pg_newlocale_from_collation(collid);
1605 :
1606 25276 : if (!mylocale->deterministic)
1607 0 : ereport(ERROR,
1608 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1609 : errmsg("nondeterministic collations are not supported for substring searches")));
1610 :
1611 25276 : len1 = toast_raw_datum_size(arg1);
1612 25276 : len2 = toast_raw_datum_size(arg2);
1613 25276 : if (len2 > len1)
1614 0 : result = false;
1615 : else
1616 : {
1617 25276 : text *targ1 = text_substring(arg1, 1, len2, false);
1618 25276 : text *targ2 = DatumGetTextPP(arg2);
1619 :
1620 25276 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1621 : VARSIZE_ANY_EXHDR(targ2)) == 0);
1622 :
1623 25276 : PG_FREE_IF_COPY(targ1, 0);
1624 25276 : PG_FREE_IF_COPY(targ2, 1);
1625 : }
1626 :
1627 25276 : PG_RETURN_BOOL(result);
1628 : }
1629 :
1630 : Datum
1631 3699587 : bttextcmp(PG_FUNCTION_ARGS)
1632 : {
1633 3699587 : text *arg1 = PG_GETARG_TEXT_PP(0);
1634 3699587 : text *arg2 = PG_GETARG_TEXT_PP(1);
1635 : int32 result;
1636 :
1637 3699587 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1638 :
1639 3699587 : PG_FREE_IF_COPY(arg1, 0);
1640 3699587 : PG_FREE_IF_COPY(arg2, 1);
1641 :
1642 3699587 : PG_RETURN_INT32(result);
1643 : }
1644 :
1645 : Datum
1646 50499 : bttextsortsupport(PG_FUNCTION_ARGS)
1647 : {
1648 50499 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1649 50499 : Oid collid = ssup->ssup_collation;
1650 : MemoryContext oldcontext;
1651 :
1652 50499 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1653 :
1654 : /* Use generic string SortSupport */
1655 50499 : varstr_sortsupport(ssup, TEXTOID, collid);
1656 :
1657 50491 : MemoryContextSwitchTo(oldcontext);
1658 :
1659 50491 : PG_RETURN_VOID();
1660 : }
1661 :
1662 : /*
1663 : * Generic sortsupport interface for character type's operator classes.
1664 : * Includes locale support, and support for BpChar semantics (i.e. removing
1665 : * trailing spaces before comparison).
1666 : *
1667 : * Relies on the assumption that text, VarChar, and BpChar all have the
1668 : * same representation.
1669 : */
1670 : void
1671 85695 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1672 : {
1673 85695 : bool abbreviate = ssup->abbreviate;
1674 85695 : bool collate_c = false;
1675 : VarStringSortSupport *sss;
1676 : pg_locale_t locale;
1677 :
1678 85695 : check_collation_set(collid);
1679 :
1680 85687 : locale = pg_newlocale_from_collation(collid);
1681 :
1682 : /*
1683 : * If possible, set ssup->comparator to a function which can be used to
1684 : * directly compare two datums. If we can do this, we'll avoid the
1685 : * overhead of a trip through the fmgr layer for every comparison, which
1686 : * can be substantial.
1687 : *
1688 : * Most typically, we'll set the comparator to varlenafastcmp_locale,
1689 : * which uses strcoll() to perform comparisons. We use that for the
1690 : * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1691 : * LC_COLLATE = C, we can make things quite a bit faster with
1692 : * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1693 : * memcmp() rather than strcoll().
1694 : */
1695 85687 : if (locale->collate_is_c)
1696 : {
1697 58253 : if (typid == BPCHAROID)
1698 157 : ssup->comparator = bpcharfastcmp_c;
1699 58096 : else if (typid == NAMEOID)
1700 : {
1701 34518 : ssup->comparator = namefastcmp_c;
1702 : /* Not supporting abbreviation with type NAME, for now */
1703 34518 : abbreviate = false;
1704 : }
1705 : else
1706 23578 : ssup->comparator = varstrfastcmp_c;
1707 :
1708 58253 : collate_c = true;
1709 : }
1710 : else
1711 : {
1712 : /*
1713 : * We use varlenafastcmp_locale except for type NAME.
1714 : */
1715 27434 : if (typid == NAMEOID)
1716 : {
1717 0 : ssup->comparator = namefastcmp_locale;
1718 : /* Not supporting abbreviation with type NAME, for now */
1719 0 : abbreviate = false;
1720 : }
1721 : else
1722 27434 : ssup->comparator = varlenafastcmp_locale;
1723 :
1724 : /*
1725 : * Unfortunately, it seems that abbreviation for non-C collations is
1726 : * broken on many common platforms; see pg_strxfrm_enabled().
1727 : *
1728 : * Even apart from the risk of broken locales, it's possible that
1729 : * there are platforms where the use of abbreviated keys should be
1730 : * disabled at compile time. For example, macOS's strxfrm()
1731 : * implementation is known to not effectively concentrate a
1732 : * significant amount of entropy from the original string in earlier
1733 : * transformed blobs. It's possible that other supported platforms
1734 : * are similarly encumbered. So, if we ever get past disabling this
1735 : * categorically, we may still want or need to disable it for
1736 : * particular platforms.
1737 : */
1738 27434 : if (!pg_strxfrm_enabled(locale))
1739 26904 : abbreviate = false;
1740 : }
1741 :
1742 : /*
1743 : * If we're using abbreviated keys, or if we're using a locale-aware
1744 : * comparison, we need to initialize a VarStringSortSupport object. Both
1745 : * cases will make use of the temporary buffers we initialize here for
1746 : * scratch space (and to detect requirement for BpChar semantics from
1747 : * caller), and the abbreviation case requires additional state.
1748 : */
1749 85687 : if (abbreviate || !collate_c)
1750 : {
1751 40266 : sss = palloc_object(VarStringSortSupport);
1752 40266 : sss->buf1 = palloc(TEXTBUFLEN);
1753 40266 : sss->buflen1 = TEXTBUFLEN;
1754 40266 : sss->buf2 = palloc(TEXTBUFLEN);
1755 40266 : sss->buflen2 = TEXTBUFLEN;
1756 : /* Start with invalid values */
1757 40266 : sss->last_len1 = -1;
1758 40266 : sss->last_len2 = -1;
1759 : /* Initialize */
1760 40266 : sss->last_returned = 0;
1761 40266 : if (collate_c)
1762 12832 : sss->locale = NULL;
1763 : else
1764 27434 : sss->locale = locale;
1765 :
1766 : /*
1767 : * To avoid somehow confusing a strxfrm() blob and an original string,
1768 : * constantly keep track of the variety of data that buf1 and buf2
1769 : * currently contain.
1770 : *
1771 : * Comparisons may be interleaved with conversion calls. Frequently,
1772 : * conversions and comparisons are batched into two distinct phases,
1773 : * but the correctness of caching cannot hinge upon this. For
1774 : * comparison caching, buffer state is only trusted if cache_blob is
1775 : * found set to false, whereas strxfrm() caching only trusts the state
1776 : * when cache_blob is found set to true.
1777 : *
1778 : * Arbitrarily initialize cache_blob to true.
1779 : */
1780 40266 : sss->cache_blob = true;
1781 40266 : sss->collate_c = collate_c;
1782 40266 : sss->typid = typid;
1783 40266 : ssup->ssup_extra = sss;
1784 :
1785 : /*
1786 : * If possible, plan to use the abbreviated keys optimization. The
1787 : * core code may switch back to authoritative comparator should
1788 : * abbreviation be aborted.
1789 : */
1790 40266 : if (abbreviate)
1791 : {
1792 13230 : sss->prop_card = 0.20;
1793 13230 : initHyperLogLog(&sss->abbr_card, 10);
1794 13230 : initHyperLogLog(&sss->full_card, 10);
1795 13230 : ssup->abbrev_full_comparator = ssup->comparator;
1796 13230 : ssup->comparator = ssup_datum_unsigned_cmp;
1797 13230 : ssup->abbrev_converter = varstr_abbrev_convert;
1798 13230 : ssup->abbrev_abort = varstr_abbrev_abort;
1799 : }
1800 : }
1801 85687 : }
1802 :
1803 : /*
1804 : * sortsupport comparison func (for C locale case)
1805 : */
1806 : static int
1807 24267588 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
1808 : {
1809 24267588 : VarString *arg1 = DatumGetVarStringPP(x);
1810 24267588 : VarString *arg2 = DatumGetVarStringPP(y);
1811 : char *a1p,
1812 : *a2p;
1813 : int len1,
1814 : len2,
1815 : result;
1816 :
1817 24267588 : a1p = VARDATA_ANY(arg1);
1818 24267588 : a2p = VARDATA_ANY(arg2);
1819 :
1820 24267588 : len1 = VARSIZE_ANY_EXHDR(arg1);
1821 24267588 : len2 = VARSIZE_ANY_EXHDR(arg2);
1822 :
1823 24267588 : result = memcmp(a1p, a2p, Min(len1, len2));
1824 24267588 : if ((result == 0) && (len1 != len2))
1825 719913 : result = (len1 < len2) ? -1 : 1;
1826 :
1827 : /* We can't afford to leak memory here. */
1828 24267588 : if (PointerGetDatum(arg1) != x)
1829 0 : pfree(arg1);
1830 24267588 : if (PointerGetDatum(arg2) != y)
1831 0 : pfree(arg2);
1832 :
1833 24267588 : return result;
1834 : }
1835 :
1836 : /*
1837 : * sortsupport comparison func (for BpChar C locale case)
1838 : *
1839 : * BpChar outsources its sortsupport to this module. Specialization for the
1840 : * varstr_sortsupport BpChar case, modeled on
1841 : * internal_bpchar_pattern_compare().
1842 : */
1843 : static int
1844 31161 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
1845 : {
1846 31161 : BpChar *arg1 = DatumGetBpCharPP(x);
1847 31161 : BpChar *arg2 = DatumGetBpCharPP(y);
1848 : char *a1p,
1849 : *a2p;
1850 : int len1,
1851 : len2,
1852 : result;
1853 :
1854 31161 : a1p = VARDATA_ANY(arg1);
1855 31161 : a2p = VARDATA_ANY(arg2);
1856 :
1857 31161 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
1858 31161 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
1859 :
1860 31161 : result = memcmp(a1p, a2p, Min(len1, len2));
1861 31161 : if ((result == 0) && (len1 != len2))
1862 2 : result = (len1 < len2) ? -1 : 1;
1863 :
1864 : /* We can't afford to leak memory here. */
1865 31161 : if (PointerGetDatum(arg1) != x)
1866 0 : pfree(arg1);
1867 31161 : if (PointerGetDatum(arg2) != y)
1868 0 : pfree(arg2);
1869 :
1870 31161 : return result;
1871 : }
1872 :
1873 : /*
1874 : * sortsupport comparison func (for NAME C locale case)
1875 : */
1876 : static int
1877 23681676 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
1878 : {
1879 23681676 : Name arg1 = DatumGetName(x);
1880 23681676 : Name arg2 = DatumGetName(y);
1881 :
1882 23681676 : return strncmp(NameStr(*arg1), NameStr(*arg2), NAMEDATALEN);
1883 : }
1884 :
1885 : /*
1886 : * sortsupport comparison func (for locale case with all varlena types)
1887 : */
1888 : static int
1889 26344316 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
1890 : {
1891 26344316 : VarString *arg1 = DatumGetVarStringPP(x);
1892 26344316 : VarString *arg2 = DatumGetVarStringPP(y);
1893 : char *a1p,
1894 : *a2p;
1895 : int len1,
1896 : len2,
1897 : result;
1898 :
1899 26344316 : a1p = VARDATA_ANY(arg1);
1900 26344316 : a2p = VARDATA_ANY(arg2);
1901 :
1902 26344316 : len1 = VARSIZE_ANY_EXHDR(arg1);
1903 26344316 : len2 = VARSIZE_ANY_EXHDR(arg2);
1904 :
1905 26344316 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
1906 :
1907 : /* We can't afford to leak memory here. */
1908 26344316 : if (PointerGetDatum(arg1) != x)
1909 0 : pfree(arg1);
1910 26344316 : if (PointerGetDatum(arg2) != y)
1911 0 : pfree(arg2);
1912 :
1913 26344316 : return result;
1914 : }
1915 :
1916 : /*
1917 : * sortsupport comparison func (for locale case with NAME type)
1918 : */
1919 : static int
1920 0 : namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
1921 : {
1922 0 : Name arg1 = DatumGetName(x);
1923 0 : Name arg2 = DatumGetName(y);
1924 :
1925 0 : return varstrfastcmp_locale(NameStr(*arg1), strlen(NameStr(*arg1)),
1926 0 : NameStr(*arg2), strlen(NameStr(*arg2)),
1927 : ssup);
1928 : }
1929 :
1930 : /*
1931 : * sortsupport comparison func for locale cases
1932 : */
1933 : static int
1934 26344316 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
1935 : {
1936 26344316 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
1937 : int result;
1938 : bool arg1_match;
1939 :
1940 : /* Fast pre-check for equality, as discussed in varstr_cmp() */
1941 26344316 : if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
1942 : {
1943 : /*
1944 : * No change in buf1 or buf2 contents, so avoid changing last_len1 or
1945 : * last_len2. Existing contents of buffers might still be used by
1946 : * next call.
1947 : *
1948 : * It's fine to allow the comparison of BpChar padding bytes here,
1949 : * even though that implies that the memcmp() will usually be
1950 : * performed for BpChar callers (though multibyte characters could
1951 : * still prevent that from occurring). The memcmp() is still very
1952 : * cheap, and BpChar's funny semantics have us remove trailing spaces
1953 : * (not limited to padding), so we need make no distinction between
1954 : * padding space characters and "real" space characters.
1955 : */
1956 6876694 : return 0;
1957 : }
1958 :
1959 19467622 : if (sss->typid == BPCHAROID)
1960 : {
1961 : /* Get true number of bytes, ignoring trailing spaces */
1962 18723 : len1 = bpchartruelen(a1p, len1);
1963 18723 : len2 = bpchartruelen(a2p, len2);
1964 : }
1965 :
1966 19467622 : if (len1 >= sss->buflen1)
1967 : {
1968 7 : sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
1969 7 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
1970 : }
1971 19467622 : if (len2 >= sss->buflen2)
1972 : {
1973 4 : sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
1974 4 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
1975 : }
1976 :
1977 : /*
1978 : * We're likely to be asked to compare the same strings repeatedly, and
1979 : * memcmp() is so much cheaper than strcoll() that it pays to try to cache
1980 : * comparisons, even though in general there is no reason to think that
1981 : * that will work out (every string datum may be unique). Caching does
1982 : * not slow things down measurably when it doesn't work out, and can speed
1983 : * things up by rather a lot when it does. In part, this is because the
1984 : * memcmp() compares data from cachelines that are needed in L1 cache even
1985 : * when the last comparison's result cannot be reused.
1986 : */
1987 19467622 : arg1_match = true;
1988 19467622 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
1989 : {
1990 17980530 : arg1_match = false;
1991 17980530 : memcpy(sss->buf1, a1p, len1);
1992 17980530 : sss->buf1[len1] = '\0';
1993 17980530 : sss->last_len1 = len1;
1994 : }
1995 :
1996 : /*
1997 : * If we're comparing the same two strings as last time, we can return the
1998 : * same answer without calling strcoll() again. This is more likely than
1999 : * it seems (at least with moderate to low cardinality sets), because
2000 : * quicksort compares the same pivot against many values.
2001 : */
2002 19467622 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2003 : {
2004 2987814 : memcpy(sss->buf2, a2p, len2);
2005 2987814 : sss->buf2[len2] = '\0';
2006 2987814 : sss->last_len2 = len2;
2007 : }
2008 16479808 : else if (arg1_match && !sss->cache_blob)
2009 : {
2010 : /* Use result cached following last actual strcoll() call */
2011 1177029 : return sss->last_returned;
2012 : }
2013 :
2014 18290593 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
2015 :
2016 : /* Break tie if necessary. */
2017 18290593 : if (result == 0 && sss->locale->deterministic)
2018 0 : result = strcmp(sss->buf1, sss->buf2);
2019 :
2020 : /* Cache result, perhaps saving an expensive strcoll() call next time */
2021 18290593 : sss->cache_blob = false;
2022 18290593 : sss->last_returned = result;
2023 18290593 : return result;
2024 : }
2025 :
2026 : /*
2027 : * Conversion routine for sortsupport. Converts original to abbreviated key
2028 : * representation. Our encoding strategy is simple -- pack the first 8 bytes
2029 : * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2030 : * stored in reverse order), and treat it as an unsigned integer. When the "C"
2031 : * locale is used just memcpy() from original instead.
2032 : */
2033 : static Datum
2034 498605 : varstr_abbrev_convert(Datum original, SortSupport ssup)
2035 : {
2036 498605 : const size_t max_prefix_bytes = sizeof(Datum);
2037 498605 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2038 498605 : VarString *authoritative = DatumGetVarStringPP(original);
2039 498605 : char *authoritative_data = VARDATA_ANY(authoritative);
2040 :
2041 : /* working state */
2042 : Datum res;
2043 : char *pres;
2044 : int len;
2045 : uint32 hash;
2046 :
2047 498605 : pres = (char *) &res;
2048 : /* memset(), so any non-overwritten bytes are NUL */
2049 498605 : memset(pres, 0, max_prefix_bytes);
2050 498605 : len = VARSIZE_ANY_EXHDR(authoritative);
2051 :
2052 : /* Get number of bytes, ignoring trailing spaces */
2053 498605 : if (sss->typid == BPCHAROID)
2054 540 : len = bpchartruelen(authoritative_data, len);
2055 :
2056 : /*
2057 : * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2058 : * abbreviate keys. The full comparator for the C locale is also
2059 : * memcmp(). This should be faster than strxfrm().
2060 : */
2061 498605 : if (sss->collate_c)
2062 497383 : memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
2063 : else
2064 : {
2065 : Size bsize;
2066 :
2067 : /*
2068 : * We're not using the C collation, so fall back on strxfrm or ICU
2069 : * analogs.
2070 : */
2071 :
2072 : /* By convention, we use buffer 1 to store and NUL-terminate */
2073 1222 : if (len >= sss->buflen1)
2074 : {
2075 0 : sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2076 0 : sss->buf1 = repalloc(sss->buf1, sss->buflen1);
2077 : }
2078 :
2079 : /* Might be able to reuse strxfrm() blob from last call */
2080 1222 : if (sss->last_len1 == len && sss->cache_blob &&
2081 610 : memcmp(sss->buf1, authoritative_data, len) == 0)
2082 : {
2083 112 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, sss->last_len2));
2084 : /* No change affecting cardinality, so no hashing required */
2085 112 : goto done;
2086 : }
2087 :
2088 1110 : memcpy(sss->buf1, authoritative_data, len);
2089 :
2090 : /*
2091 : * pg_strxfrm() and pg_strxfrm_prefix expect NUL-terminated strings.
2092 : */
2093 1110 : sss->buf1[len] = '\0';
2094 1110 : sss->last_len1 = len;
2095 :
2096 1110 : if (pg_strxfrm_prefix_enabled(sss->locale))
2097 : {
2098 1110 : if (sss->buflen2 < max_prefix_bytes)
2099 : {
2100 0 : sss->buflen2 = Max(max_prefix_bytes,
2101 : Min(sss->buflen2 * 2, MaxAllocSize));
2102 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2103 : }
2104 :
2105 1110 : bsize = pg_strxfrm_prefix(sss->buf2, sss->buf1,
2106 : max_prefix_bytes, sss->locale);
2107 1110 : sss->last_len2 = bsize;
2108 : }
2109 : else
2110 : {
2111 : /*
2112 : * Loop: Call pg_strxfrm(), possibly enlarge buffer, and try
2113 : * again. The pg_strxfrm() function leaves the result buffer
2114 : * content undefined if the result did not fit, so we need to
2115 : * retry until everything fits, even though we only need the first
2116 : * few bytes in the end.
2117 : */
2118 : for (;;)
2119 : {
2120 0 : bsize = pg_strxfrm(sss->buf2, sss->buf1, sss->buflen2,
2121 : sss->locale);
2122 :
2123 0 : sss->last_len2 = bsize;
2124 0 : if (bsize < sss->buflen2)
2125 0 : break;
2126 :
2127 : /*
2128 : * Grow buffer and retry.
2129 : */
2130 0 : sss->buflen2 = Max(bsize + 1,
2131 : Min(sss->buflen2 * 2, MaxAllocSize));
2132 0 : sss->buf2 = repalloc(sss->buf2, sss->buflen2);
2133 : }
2134 : }
2135 :
2136 : /*
2137 : * Every Datum byte is always compared. This is safe because the
2138 : * strxfrm() blob is itself NUL terminated, leaving no danger of
2139 : * misinterpreting any NUL bytes not intended to be interpreted as
2140 : * logically representing termination.
2141 : */
2142 1110 : memcpy(pres, sss->buf2, Min(max_prefix_bytes, bsize));
2143 : }
2144 :
2145 : /*
2146 : * Maintain approximate cardinality of both abbreviated keys and original,
2147 : * authoritative keys using HyperLogLog. Used as cheap insurance against
2148 : * the worst case, where we do many string transformations for no saving
2149 : * in full strcoll()-based comparisons. These statistics are used by
2150 : * varstr_abbrev_abort().
2151 : *
2152 : * First, Hash key proper, or a significant fraction of it. Mix in length
2153 : * in order to compensate for cases where differences are past
2154 : * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2155 : */
2156 498493 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2157 : Min(len, PG_CACHE_LINE_SIZE)));
2158 :
2159 498493 : if (len > PG_CACHE_LINE_SIZE)
2160 97 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2161 :
2162 498493 : addHyperLogLog(&sss->full_card, hash);
2163 :
2164 : /* Hash abbreviated key */
2165 : {
2166 : uint32 tmp;
2167 :
2168 498493 : tmp = DatumGetUInt32(res) ^ (uint32) (DatumGetUInt64(res) >> 32);
2169 498493 : hash = DatumGetUInt32(hash_uint32(tmp));
2170 : }
2171 :
2172 498493 : addHyperLogLog(&sss->abbr_card, hash);
2173 :
2174 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2175 498493 : sss->cache_blob = true;
2176 498605 : done:
2177 :
2178 : /*
2179 : * Byteswap on little-endian machines.
2180 : *
2181 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
2182 : * 3-way comparator) works correctly on all platforms. If we didn't do
2183 : * this, the comparator would have to call memcmp() with a pair of
2184 : * pointers to the first byte of each abbreviated key, which is slower.
2185 : */
2186 498605 : res = DatumBigEndianToNative(res);
2187 :
2188 : /* Don't leak memory here */
2189 498605 : if (PointerGetDatum(authoritative) != original)
2190 1 : pfree(authoritative);
2191 :
2192 498605 : return res;
2193 : }
2194 :
2195 : /*
2196 : * Callback for estimating effectiveness of abbreviated key optimization, using
2197 : * heuristic rules. Returns value indicating if the abbreviation optimization
2198 : * should be aborted, based on its projected effectiveness.
2199 : */
2200 : static bool
2201 1438 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2202 : {
2203 1438 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2204 : double abbrev_distinct,
2205 : key_distinct;
2206 :
2207 : Assert(ssup->abbreviate);
2208 :
2209 : /* Have a little patience */
2210 1438 : if (memtupcount < 100)
2211 830 : return false;
2212 :
2213 608 : abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2214 608 : key_distinct = estimateHyperLogLog(&sss->full_card);
2215 :
2216 : /*
2217 : * Clamp cardinality estimates to at least one distinct value. While
2218 : * NULLs are generally disregarded, if only NULL values were seen so far,
2219 : * that might misrepresent costs if we failed to clamp.
2220 : */
2221 608 : if (abbrev_distinct < 1.0)
2222 0 : abbrev_distinct = 1.0;
2223 :
2224 608 : if (key_distinct < 1.0)
2225 0 : key_distinct = 1.0;
2226 :
2227 : /*
2228 : * In the worst case all abbreviated keys are identical, while at the same
2229 : * time there are differences within full key strings not captured in
2230 : * abbreviations.
2231 : */
2232 608 : if (trace_sort)
2233 : {
2234 0 : double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2235 :
2236 0 : elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2237 : "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2238 : memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2239 : sss->prop_card);
2240 : }
2241 :
2242 : /*
2243 : * If the number of distinct abbreviated keys approximately matches the
2244 : * number of distinct authoritative original keys, that's reason enough to
2245 : * proceed. We can win even with a very low cardinality set if most
2246 : * tie-breakers only memcmp(). This is by far the most important
2247 : * consideration.
2248 : *
2249 : * While comparisons that are resolved at the abbreviated key level are
2250 : * considerably cheaper than tie-breakers resolved with memcmp(), both of
2251 : * those two outcomes are so much cheaper than a full strcoll() once
2252 : * sorting is underway that it doesn't seem worth it to weigh abbreviated
2253 : * cardinality against the overall size of the set in order to more
2254 : * accurately model costs. Assume that an abbreviated comparison, and an
2255 : * abbreviated comparison with a cheap memcmp()-based authoritative
2256 : * resolution are equivalent.
2257 : */
2258 608 : if (abbrev_distinct > key_distinct * sss->prop_card)
2259 : {
2260 : /*
2261 : * When we have exceeded 10,000 tuples, decay required cardinality
2262 : * aggressively for next call.
2263 : *
2264 : * This is useful because the number of comparisons required on
2265 : * average increases at a linearithmic rate, and at roughly 10,000
2266 : * tuples that factor will start to dominate over the linear costs of
2267 : * string transformation (this is a conservative estimate). The decay
2268 : * rate is chosen to be a little less aggressive than halving -- which
2269 : * (since we're called at points at which memtupcount has doubled)
2270 : * would never see the cost model actually abort past the first call
2271 : * following a decay. This decay rate is mostly a precaution against
2272 : * a sudden, violent swing in how well abbreviated cardinality tracks
2273 : * full key cardinality. The decay also serves to prevent a marginal
2274 : * case from being aborted too late, when too much has already been
2275 : * invested in string transformation.
2276 : *
2277 : * It's possible for sets of several million distinct strings with
2278 : * mere tens of thousands of distinct abbreviated keys to still
2279 : * benefit very significantly. This will generally occur provided
2280 : * each abbreviated key is a proxy for a roughly uniform number of the
2281 : * set's full keys. If it isn't so, we hope to catch that early and
2282 : * abort. If it isn't caught early, by the time the problem is
2283 : * apparent it's probably not worth aborting.
2284 : */
2285 608 : if (memtupcount > 10000)
2286 2 : sss->prop_card *= 0.65;
2287 :
2288 608 : return false;
2289 : }
2290 :
2291 : /*
2292 : * Abort abbreviation strategy.
2293 : *
2294 : * The worst case, where all abbreviated keys are identical while all
2295 : * original strings differ will typically only see a regression of about
2296 : * 10% in execution time for small to medium sized lists of strings.
2297 : * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2298 : * often expect very large improvements, particularly with sets of strings
2299 : * of moderately high to high abbreviated cardinality. There is little to
2300 : * lose but much to gain, which our strategy reflects.
2301 : */
2302 0 : if (trace_sort)
2303 0 : elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2304 : "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2305 : memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2306 :
2307 0 : return true;
2308 : }
2309 :
2310 : /*
2311 : * Generic equalimage support function for character type's operator classes.
2312 : * Disables the use of deduplication with nondeterministic collations.
2313 : */
2314 : Datum
2315 5350 : btvarstrequalimage(PG_FUNCTION_ARGS)
2316 : {
2317 : #ifdef NOT_USED
2318 : Oid opcintype = PG_GETARG_OID(0);
2319 : #endif
2320 5350 : Oid collid = PG_GET_COLLATION();
2321 : pg_locale_t locale;
2322 :
2323 5350 : check_collation_set(collid);
2324 :
2325 5350 : locale = pg_newlocale_from_collation(collid);
2326 :
2327 5350 : PG_RETURN_BOOL(locale->deterministic);
2328 : }
2329 :
2330 : Datum
2331 145885 : text_larger(PG_FUNCTION_ARGS)
2332 : {
2333 145885 : text *arg1 = PG_GETARG_TEXT_PP(0);
2334 145885 : text *arg2 = PG_GETARG_TEXT_PP(1);
2335 : text *result;
2336 :
2337 145885 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2338 :
2339 145885 : PG_RETURN_TEXT_P(result);
2340 : }
2341 :
2342 : Datum
2343 50562 : text_smaller(PG_FUNCTION_ARGS)
2344 : {
2345 50562 : text *arg1 = PG_GETARG_TEXT_PP(0);
2346 50562 : text *arg2 = PG_GETARG_TEXT_PP(1);
2347 : text *result;
2348 :
2349 50562 : result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2350 :
2351 50562 : PG_RETURN_TEXT_P(result);
2352 : }
2353 :
2354 :
2355 : /*
2356 : * Cross-type comparison functions for types text and name.
2357 : */
2358 :
2359 : Datum
2360 144983 : nameeqtext(PG_FUNCTION_ARGS)
2361 : {
2362 144983 : Name arg1 = PG_GETARG_NAME(0);
2363 144983 : text *arg2 = PG_GETARG_TEXT_PP(1);
2364 144983 : size_t len1 = strlen(NameStr(*arg1));
2365 144983 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2366 144983 : Oid collid = PG_GET_COLLATION();
2367 : bool result;
2368 :
2369 144983 : check_collation_set(collid);
2370 :
2371 144983 : if (collid == C_COLLATION_OID)
2372 171970 : result = (len1 == len2 &&
2373 83258 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2374 : else
2375 56271 : result = (varstr_cmp(NameStr(*arg1), len1,
2376 56271 : VARDATA_ANY(arg2), len2,
2377 : collid) == 0);
2378 :
2379 144983 : PG_FREE_IF_COPY(arg2, 1);
2380 :
2381 144983 : PG_RETURN_BOOL(result);
2382 : }
2383 :
2384 : Datum
2385 5722 : texteqname(PG_FUNCTION_ARGS)
2386 : {
2387 5722 : text *arg1 = PG_GETARG_TEXT_PP(0);
2388 5722 : Name arg2 = PG_GETARG_NAME(1);
2389 5722 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2390 5722 : size_t len2 = strlen(NameStr(*arg2));
2391 5722 : Oid collid = PG_GET_COLLATION();
2392 : bool result;
2393 :
2394 5722 : check_collation_set(collid);
2395 :
2396 5722 : if (collid == C_COLLATION_OID)
2397 378 : result = (len1 == len2 &&
2398 121 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2399 : else
2400 5465 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2401 5465 : NameStr(*arg2), len2,
2402 : collid) == 0);
2403 :
2404 5722 : PG_FREE_IF_COPY(arg1, 0);
2405 :
2406 5722 : PG_RETURN_BOOL(result);
2407 : }
2408 :
2409 : Datum
2410 12 : namenetext(PG_FUNCTION_ARGS)
2411 : {
2412 12 : Name arg1 = PG_GETARG_NAME(0);
2413 12 : text *arg2 = PG_GETARG_TEXT_PP(1);
2414 12 : size_t len1 = strlen(NameStr(*arg1));
2415 12 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2416 12 : Oid collid = PG_GET_COLLATION();
2417 : bool result;
2418 :
2419 12 : check_collation_set(collid);
2420 :
2421 12 : if (collid == C_COLLATION_OID)
2422 0 : result = !(len1 == len2 &&
2423 0 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2424 : else
2425 12 : result = !(varstr_cmp(NameStr(*arg1), len1,
2426 12 : VARDATA_ANY(arg2), len2,
2427 : collid) == 0);
2428 :
2429 12 : PG_FREE_IF_COPY(arg2, 1);
2430 :
2431 12 : PG_RETURN_BOOL(result);
2432 : }
2433 :
2434 : Datum
2435 12 : textnename(PG_FUNCTION_ARGS)
2436 : {
2437 12 : text *arg1 = PG_GETARG_TEXT_PP(0);
2438 12 : Name arg2 = PG_GETARG_NAME(1);
2439 12 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2440 12 : size_t len2 = strlen(NameStr(*arg2));
2441 12 : Oid collid = PG_GET_COLLATION();
2442 : bool result;
2443 :
2444 12 : check_collation_set(collid);
2445 :
2446 12 : if (collid == C_COLLATION_OID)
2447 0 : result = !(len1 == len2 &&
2448 0 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2449 : else
2450 12 : result = !(varstr_cmp(VARDATA_ANY(arg1), len1,
2451 12 : NameStr(*arg2), len2,
2452 : collid) == 0);
2453 :
2454 12 : PG_FREE_IF_COPY(arg1, 0);
2455 :
2456 12 : PG_RETURN_BOOL(result);
2457 : }
2458 :
2459 : Datum
2460 89322 : btnametextcmp(PG_FUNCTION_ARGS)
2461 : {
2462 89322 : Name arg1 = PG_GETARG_NAME(0);
2463 89322 : text *arg2 = PG_GETARG_TEXT_PP(1);
2464 : int32 result;
2465 :
2466 89322 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2467 89322 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2468 : PG_GET_COLLATION());
2469 :
2470 89322 : PG_FREE_IF_COPY(arg2, 1);
2471 :
2472 89322 : PG_RETURN_INT32(result);
2473 : }
2474 :
2475 : Datum
2476 22 : bttextnamecmp(PG_FUNCTION_ARGS)
2477 : {
2478 22 : text *arg1 = PG_GETARG_TEXT_PP(0);
2479 22 : Name arg2 = PG_GETARG_NAME(1);
2480 : int32 result;
2481 :
2482 22 : result = varstr_cmp(VARDATA_ANY(arg1), VARSIZE_ANY_EXHDR(arg1),
2483 22 : NameStr(*arg2), strlen(NameStr(*arg2)),
2484 : PG_GET_COLLATION());
2485 :
2486 22 : PG_FREE_IF_COPY(arg1, 0);
2487 :
2488 22 : PG_RETURN_INT32(result);
2489 : }
2490 :
2491 : #define CmpCall(cmpfunc) \
2492 : DatumGetInt32(DirectFunctionCall2Coll(cmpfunc, \
2493 : PG_GET_COLLATION(), \
2494 : PG_GETARG_DATUM(0), \
2495 : PG_GETARG_DATUM(1)))
2496 :
2497 : Datum
2498 42760 : namelttext(PG_FUNCTION_ARGS)
2499 : {
2500 42760 : PG_RETURN_BOOL(CmpCall(btnametextcmp) < 0);
2501 : }
2502 :
2503 : Datum
2504 0 : nameletext(PG_FUNCTION_ARGS)
2505 : {
2506 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) <= 0);
2507 : }
2508 :
2509 : Datum
2510 0 : namegttext(PG_FUNCTION_ARGS)
2511 : {
2512 0 : PG_RETURN_BOOL(CmpCall(btnametextcmp) > 0);
2513 : }
2514 :
2515 : Datum
2516 38338 : namegetext(PG_FUNCTION_ARGS)
2517 : {
2518 38338 : PG_RETURN_BOOL(CmpCall(btnametextcmp) >= 0);
2519 : }
2520 :
2521 : Datum
2522 0 : textltname(PG_FUNCTION_ARGS)
2523 : {
2524 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) < 0);
2525 : }
2526 :
2527 : Datum
2528 0 : textlename(PG_FUNCTION_ARGS)
2529 : {
2530 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) <= 0);
2531 : }
2532 :
2533 : Datum
2534 0 : textgtname(PG_FUNCTION_ARGS)
2535 : {
2536 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) > 0);
2537 : }
2538 :
2539 : Datum
2540 0 : textgename(PG_FUNCTION_ARGS)
2541 : {
2542 0 : PG_RETURN_BOOL(CmpCall(bttextnamecmp) >= 0);
2543 : }
2544 :
2545 : #undef CmpCall
2546 :
2547 :
2548 : /*
2549 : * The following operators support character-by-character comparison
2550 : * of text datums, to allow building indexes suitable for LIKE clauses.
2551 : * Note that the regular texteq/textne comparison operators, and regular
2552 : * support functions 1 and 2 with "C" collation are assumed to be
2553 : * compatible with these!
2554 : */
2555 :
2556 : static int
2557 107164 : internal_text_pattern_compare(text *arg1, text *arg2)
2558 : {
2559 : int result;
2560 : int len1,
2561 : len2;
2562 :
2563 107164 : len1 = VARSIZE_ANY_EXHDR(arg1);
2564 107164 : len2 = VARSIZE_ANY_EXHDR(arg2);
2565 :
2566 107164 : result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2567 107164 : if (result != 0)
2568 107076 : return result;
2569 88 : else if (len1 < len2)
2570 0 : return -1;
2571 88 : else if (len1 > len2)
2572 56 : return 1;
2573 : else
2574 32 : return 0;
2575 : }
2576 :
2577 :
2578 : Datum
2579 32112 : text_pattern_lt(PG_FUNCTION_ARGS)
2580 : {
2581 32112 : text *arg1 = PG_GETARG_TEXT_PP(0);
2582 32112 : text *arg2 = PG_GETARG_TEXT_PP(1);
2583 : int result;
2584 :
2585 32112 : result = internal_text_pattern_compare(arg1, arg2);
2586 :
2587 32112 : PG_FREE_IF_COPY(arg1, 0);
2588 32112 : PG_FREE_IF_COPY(arg2, 1);
2589 :
2590 32112 : PG_RETURN_BOOL(result < 0);
2591 : }
2592 :
2593 :
2594 : Datum
2595 25006 : text_pattern_le(PG_FUNCTION_ARGS)
2596 : {
2597 25006 : text *arg1 = PG_GETARG_TEXT_PP(0);
2598 25006 : text *arg2 = PG_GETARG_TEXT_PP(1);
2599 : int result;
2600 :
2601 25006 : result = internal_text_pattern_compare(arg1, arg2);
2602 :
2603 25006 : PG_FREE_IF_COPY(arg1, 0);
2604 25006 : PG_FREE_IF_COPY(arg2, 1);
2605 :
2606 25006 : PG_RETURN_BOOL(result <= 0);
2607 : }
2608 :
2609 :
2610 : Datum
2611 25022 : text_pattern_ge(PG_FUNCTION_ARGS)
2612 : {
2613 25022 : text *arg1 = PG_GETARG_TEXT_PP(0);
2614 25022 : text *arg2 = PG_GETARG_TEXT_PP(1);
2615 : int result;
2616 :
2617 25022 : result = internal_text_pattern_compare(arg1, arg2);
2618 :
2619 25022 : PG_FREE_IF_COPY(arg1, 0);
2620 25022 : PG_FREE_IF_COPY(arg2, 1);
2621 :
2622 25022 : PG_RETURN_BOOL(result >= 0);
2623 : }
2624 :
2625 :
2626 : Datum
2627 25006 : text_pattern_gt(PG_FUNCTION_ARGS)
2628 : {
2629 25006 : text *arg1 = PG_GETARG_TEXT_PP(0);
2630 25006 : text *arg2 = PG_GETARG_TEXT_PP(1);
2631 : int result;
2632 :
2633 25006 : result = internal_text_pattern_compare(arg1, arg2);
2634 :
2635 25006 : PG_FREE_IF_COPY(arg1, 0);
2636 25006 : PG_FREE_IF_COPY(arg2, 1);
2637 :
2638 25006 : PG_RETURN_BOOL(result > 0);
2639 : }
2640 :
2641 :
2642 : Datum
2643 18 : bttext_pattern_cmp(PG_FUNCTION_ARGS)
2644 : {
2645 18 : text *arg1 = PG_GETARG_TEXT_PP(0);
2646 18 : text *arg2 = PG_GETARG_TEXT_PP(1);
2647 : int result;
2648 :
2649 18 : result = internal_text_pattern_compare(arg1, arg2);
2650 :
2651 18 : PG_FREE_IF_COPY(arg1, 0);
2652 18 : PG_FREE_IF_COPY(arg2, 1);
2653 :
2654 18 : PG_RETURN_INT32(result);
2655 : }
2656 :
2657 :
2658 : Datum
2659 77 : bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
2660 : {
2661 77 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
2662 : MemoryContext oldcontext;
2663 :
2664 77 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2665 :
2666 : /* Use generic string SortSupport, forcing "C" collation */
2667 77 : varstr_sortsupport(ssup, TEXTOID, C_COLLATION_OID);
2668 :
2669 77 : MemoryContextSwitchTo(oldcontext);
2670 :
2671 77 : PG_RETURN_VOID();
2672 : }
2673 :
2674 :
2675 : /* text_name()
2676 : * Converts a text type to a Name type.
2677 : */
2678 : Datum
2679 20735 : text_name(PG_FUNCTION_ARGS)
2680 : {
2681 20735 : text *s = PG_GETARG_TEXT_PP(0);
2682 : Name result;
2683 : int len;
2684 :
2685 20735 : len = VARSIZE_ANY_EXHDR(s);
2686 :
2687 : /* Truncate oversize input */
2688 20735 : if (len >= NAMEDATALEN)
2689 5 : len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
2690 :
2691 : /* We use palloc0 here to ensure result is zero-padded */
2692 20735 : result = (Name) palloc0(NAMEDATALEN);
2693 20735 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
2694 :
2695 20735 : PG_RETURN_NAME(result);
2696 : }
2697 :
2698 : /* name_text()
2699 : * Converts a Name type to a text type.
2700 : */
2701 : Datum
2702 421921 : name_text(PG_FUNCTION_ARGS)
2703 : {
2704 421921 : Name s = PG_GETARG_NAME(0);
2705 :
2706 421921 : PG_RETURN_TEXT_P(cstring_to_text(NameStr(*s)));
2707 : }
2708 :
2709 :
2710 : /*
2711 : * textToQualifiedNameList - convert a text object to list of names
2712 : *
2713 : * This implements the input parsing needed by nextval() and other
2714 : * functions that take a text parameter representing a qualified name.
2715 : * We split the name at dots, downcase if not double-quoted, and
2716 : * truncate names if they're too long.
2717 : */
2718 : List *
2719 2885 : textToQualifiedNameList(text *textval)
2720 : {
2721 : char *rawname;
2722 2885 : List *result = NIL;
2723 : List *namelist;
2724 : ListCell *l;
2725 :
2726 : /* Convert to C string (handles possible detoasting). */
2727 : /* Note we rely on being able to modify rawname below. */
2728 2885 : rawname = text_to_cstring(textval);
2729 :
2730 2885 : if (!SplitIdentifierString(rawname, '.', &namelist))
2731 0 : ereport(ERROR,
2732 : (errcode(ERRCODE_INVALID_NAME),
2733 : errmsg("invalid name syntax")));
2734 :
2735 2885 : if (namelist == NIL)
2736 0 : ereport(ERROR,
2737 : (errcode(ERRCODE_INVALID_NAME),
2738 : errmsg("invalid name syntax")));
2739 :
2740 5847 : foreach(l, namelist)
2741 : {
2742 2962 : char *curname = (char *) lfirst(l);
2743 :
2744 2962 : result = lappend(result, makeString(pstrdup(curname)));
2745 : }
2746 :
2747 2885 : pfree(rawname);
2748 2885 : list_free(namelist);
2749 :
2750 2885 : return result;
2751 : }
2752 :
2753 : /*
2754 : * SplitIdentifierString --- parse a string containing identifiers
2755 : *
2756 : * This is the guts of textToQualifiedNameList, and is exported for use in
2757 : * other situations such as parsing GUC variables. In the GUC case, it's
2758 : * important to avoid memory leaks, so the API is designed to minimize the
2759 : * amount of stuff that needs to be allocated and freed.
2760 : *
2761 : * Inputs:
2762 : * rawstring: the input string; must be overwritable! On return, it's
2763 : * been modified to contain the separated identifiers.
2764 : * separator: the separator punctuation expected between identifiers
2765 : * (typically '.' or ','). Whitespace may also appear around
2766 : * identifiers.
2767 : * Outputs:
2768 : * namelist: filled with a palloc'd list of pointers to identifiers within
2769 : * rawstring. Caller should list_free() this even on error return.
2770 : *
2771 : * Returns true if okay, false if there is a syntax error in the string.
2772 : *
2773 : * Note that an empty string is considered okay here, though not in
2774 : * textToQualifiedNameList.
2775 : */
2776 : bool
2777 211911 : SplitIdentifierString(char *rawstring, char separator,
2778 : List **namelist)
2779 : {
2780 211911 : char *nextp = rawstring;
2781 211911 : bool done = false;
2782 :
2783 211911 : *namelist = NIL;
2784 :
2785 211915 : while (scanner_isspace(*nextp))
2786 4 : nextp++; /* skip leading whitespace */
2787 :
2788 211911 : if (*nextp == '\0')
2789 17331 : return true; /* empty string represents empty list */
2790 :
2791 : /* At the top of the loop, we are at start of a new identifier. */
2792 : do
2793 : {
2794 : char *curname;
2795 : char *endp;
2796 :
2797 362001 : if (*nextp == '"')
2798 : {
2799 : /* Quoted name --- collapse quote-quote pairs, no downcasing */
2800 24914 : curname = nextp + 1;
2801 : for (;;)
2802 : {
2803 24916 : endp = strchr(nextp + 1, '"');
2804 24915 : if (endp == NULL)
2805 0 : return false; /* mismatched quotes */
2806 24915 : if (endp[1] != '"')
2807 24914 : break; /* found end of quoted name */
2808 : /* Collapse adjacent quotes into one quote, and look again */
2809 1 : memmove(endp, endp + 1, strlen(endp));
2810 1 : nextp = endp;
2811 : }
2812 : /* endp now points at the terminating quote */
2813 24914 : nextp = endp + 1;
2814 : }
2815 : else
2816 : {
2817 : /* Unquoted name --- extends to separator or whitespace */
2818 : char *downname;
2819 : int len;
2820 :
2821 337087 : curname = nextp;
2822 3094322 : while (*nextp && *nextp != separator &&
2823 2757236 : !scanner_isspace(*nextp))
2824 2757235 : nextp++;
2825 337087 : endp = nextp;
2826 337087 : if (curname == nextp)
2827 0 : return false; /* empty unquoted name not allowed */
2828 :
2829 : /*
2830 : * Downcase the identifier, using same code as main lexer does.
2831 : *
2832 : * XXX because we want to overwrite the input in-place, we cannot
2833 : * support a downcasing transformation that increases the string
2834 : * length. This is not a problem given the current implementation
2835 : * of downcase_truncate_identifier, but we'll probably have to do
2836 : * something about this someday.
2837 : */
2838 337087 : len = endp - curname;
2839 337087 : downname = downcase_truncate_identifier(curname, len, false);
2840 : Assert(strlen(downname) <= len);
2841 337087 : strncpy(curname, downname, len); /* strncpy is required here */
2842 337087 : pfree(downname);
2843 : }
2844 :
2845 362002 : while (scanner_isspace(*nextp))
2846 1 : nextp++; /* skip trailing whitespace */
2847 :
2848 362001 : if (*nextp == separator)
2849 : {
2850 167421 : nextp++;
2851 320362 : while (scanner_isspace(*nextp))
2852 152941 : nextp++; /* skip leading whitespace for next */
2853 : /* we expect another name, so done remains false */
2854 : }
2855 194580 : else if (*nextp == '\0')
2856 194579 : done = true;
2857 : else
2858 1 : return false; /* invalid syntax */
2859 :
2860 : /* Now safe to overwrite separator with a null */
2861 362000 : *endp = '\0';
2862 :
2863 : /* Truncate name if it's overlength */
2864 362000 : truncate_identifier(curname, strlen(curname), false);
2865 :
2866 : /*
2867 : * Finished isolating current name --- add it to list
2868 : */
2869 362000 : *namelist = lappend(*namelist, curname);
2870 :
2871 : /* Loop back if we didn't reach end of string */
2872 362000 : } while (!done);
2873 :
2874 194579 : return true;
2875 : }
2876 :
2877 :
2878 : /*
2879 : * SplitDirectoriesString --- parse a string containing file/directory names
2880 : *
2881 : * This works fine on file names too; the function name is historical.
2882 : *
2883 : * This is similar to SplitIdentifierString, except that the parsing
2884 : * rules are meant to handle pathnames instead of identifiers: there is
2885 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
2886 : * and we apply canonicalize_path() to each extracted string. Because of the
2887 : * last, the returned strings are separately palloc'd rather than being
2888 : * pointers into rawstring --- but we still scribble on rawstring.
2889 : *
2890 : * Inputs:
2891 : * rawstring: the input string; must be modifiable!
2892 : * separator: the separator punctuation expected between directories
2893 : * (typically ',' or ';'). Whitespace may also appear around
2894 : * directories.
2895 : * Outputs:
2896 : * namelist: filled with a palloc'd list of directory names.
2897 : * Caller should list_free_deep() this even on error return.
2898 : *
2899 : * Returns true if okay, false if there is a syntax error in the string.
2900 : *
2901 : * Note that an empty string is considered okay here.
2902 : */
2903 : bool
2904 996 : SplitDirectoriesString(char *rawstring, char separator,
2905 : List **namelist)
2906 : {
2907 996 : char *nextp = rawstring;
2908 996 : bool done = false;
2909 :
2910 996 : *namelist = NIL;
2911 :
2912 996 : while (scanner_isspace(*nextp))
2913 0 : nextp++; /* skip leading whitespace */
2914 :
2915 996 : if (*nextp == '\0')
2916 1 : return true; /* empty string represents empty list */
2917 :
2918 : /* At the top of the loop, we are at start of a new directory. */
2919 : do
2920 : {
2921 : char *curname;
2922 : char *endp;
2923 :
2924 1000 : if (*nextp == '"')
2925 : {
2926 : /* Quoted name --- collapse quote-quote pairs */
2927 0 : curname = nextp + 1;
2928 : for (;;)
2929 : {
2930 0 : endp = strchr(nextp + 1, '"');
2931 0 : if (endp == NULL)
2932 0 : return false; /* mismatched quotes */
2933 0 : if (endp[1] != '"')
2934 0 : break; /* found end of quoted name */
2935 : /* Collapse adjacent quotes into one quote, and look again */
2936 0 : memmove(endp, endp + 1, strlen(endp));
2937 0 : nextp = endp;
2938 : }
2939 : /* endp now points at the terminating quote */
2940 0 : nextp = endp + 1;
2941 : }
2942 : else
2943 : {
2944 : /* Unquoted name --- extends to separator or end of string */
2945 1000 : curname = endp = nextp;
2946 16726 : while (*nextp && *nextp != separator)
2947 : {
2948 : /* trailing whitespace should not be included in name */
2949 15726 : if (!scanner_isspace(*nextp))
2950 15726 : endp = nextp + 1;
2951 15726 : nextp++;
2952 : }
2953 1000 : if (curname == endp)
2954 0 : return false; /* empty unquoted name not allowed */
2955 : }
2956 :
2957 1000 : while (scanner_isspace(*nextp))
2958 0 : nextp++; /* skip trailing whitespace */
2959 :
2960 1000 : if (*nextp == separator)
2961 : {
2962 5 : nextp++;
2963 8 : while (scanner_isspace(*nextp))
2964 3 : nextp++; /* skip leading whitespace for next */
2965 : /* we expect another name, so done remains false */
2966 : }
2967 995 : else if (*nextp == '\0')
2968 995 : done = true;
2969 : else
2970 0 : return false; /* invalid syntax */
2971 :
2972 : /* Now safe to overwrite separator with a null */
2973 1000 : *endp = '\0';
2974 :
2975 : /* Truncate path if it's overlength */
2976 1000 : if (strlen(curname) >= MAXPGPATH)
2977 0 : curname[MAXPGPATH - 1] = '\0';
2978 :
2979 : /*
2980 : * Finished isolating current name --- add it to list
2981 : */
2982 1000 : curname = pstrdup(curname);
2983 1000 : canonicalize_path(curname);
2984 1000 : *namelist = lappend(*namelist, curname);
2985 :
2986 : /* Loop back if we didn't reach end of string */
2987 1000 : } while (!done);
2988 :
2989 995 : return true;
2990 : }
2991 :
2992 :
2993 : /*
2994 : * SplitGUCList --- parse a string containing identifiers or file names
2995 : *
2996 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
2997 : * presuming whether the elements will be taken as identifiers or file names.
2998 : * We assume the input has already been through flatten_set_variable_args(),
2999 : * so that we need never downcase (if appropriate, that was done already).
3000 : * Nor do we ever truncate, since we don't know the correct max length.
3001 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
3002 : * because any embedded whitespace should have led to double-quoting).
3003 : * Otherwise the API is identical to SplitIdentifierString.
3004 : *
3005 : * XXX it's annoying to have so many copies of this string-splitting logic.
3006 : * However, it's not clear that having one function with a bunch of option
3007 : * flags would be much better.
3008 : *
3009 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
3010 : * Be sure to update that if you have to change this.
3011 : *
3012 : * Inputs:
3013 : * rawstring: the input string; must be overwritable! On return, it's
3014 : * been modified to contain the separated identifiers.
3015 : * separator: the separator punctuation expected between identifiers
3016 : * (typically '.' or ','). Whitespace may also appear around
3017 : * identifiers.
3018 : * Outputs:
3019 : * namelist: filled with a palloc'd list of pointers to identifiers within
3020 : * rawstring. Caller should list_free() this even on error return.
3021 : *
3022 : * Returns true if okay, false if there is a syntax error in the string.
3023 : */
3024 : bool
3025 3843 : SplitGUCList(char *rawstring, char separator,
3026 : List **namelist)
3027 : {
3028 3843 : char *nextp = rawstring;
3029 3843 : bool done = false;
3030 :
3031 3843 : *namelist = NIL;
3032 :
3033 3843 : while (scanner_isspace(*nextp))
3034 0 : nextp++; /* skip leading whitespace */
3035 :
3036 3843 : if (*nextp == '\0')
3037 2135 : return true; /* empty string represents empty list */
3038 :
3039 : /* At the top of the loop, we are at start of a new identifier. */
3040 : do
3041 : {
3042 : char *curname;
3043 : char *endp;
3044 :
3045 1788 : if (*nextp == '"')
3046 : {
3047 : /* Quoted name --- collapse quote-quote pairs */
3048 16 : curname = nextp + 1;
3049 : for (;;)
3050 : {
3051 24 : endp = strchr(nextp + 1, '"');
3052 20 : if (endp == NULL)
3053 0 : return false; /* mismatched quotes */
3054 20 : if (endp[1] != '"')
3055 16 : break; /* found end of quoted name */
3056 : /* Collapse adjacent quotes into one quote, and look again */
3057 4 : memmove(endp, endp + 1, strlen(endp));
3058 4 : nextp = endp;
3059 : }
3060 : /* endp now points at the terminating quote */
3061 16 : nextp = endp + 1;
3062 : }
3063 : else
3064 : {
3065 : /* Unquoted name --- extends to separator or whitespace */
3066 1772 : curname = nextp;
3067 14505 : while (*nextp && *nextp != separator &&
3068 12733 : !scanner_isspace(*nextp))
3069 12733 : nextp++;
3070 1772 : endp = nextp;
3071 1772 : if (curname == nextp)
3072 0 : return false; /* empty unquoted name not allowed */
3073 : }
3074 :
3075 1788 : while (scanner_isspace(*nextp))
3076 0 : nextp++; /* skip trailing whitespace */
3077 :
3078 1788 : if (*nextp == separator)
3079 : {
3080 80 : nextp++;
3081 156 : while (scanner_isspace(*nextp))
3082 76 : nextp++; /* skip leading whitespace for next */
3083 : /* we expect another name, so done remains false */
3084 : }
3085 1708 : else if (*nextp == '\0')
3086 1708 : done = true;
3087 : else
3088 0 : return false; /* invalid syntax */
3089 :
3090 : /* Now safe to overwrite separator with a null */
3091 1788 : *endp = '\0';
3092 :
3093 : /*
3094 : * Finished isolating current name --- add it to list
3095 : */
3096 1788 : *namelist = lappend(*namelist, curname);
3097 :
3098 : /* Loop back if we didn't reach end of string */
3099 1788 : } while (!done);
3100 :
3101 1708 : return true;
3102 : }
3103 :
3104 : /*
3105 : * appendStringInfoText
3106 : *
3107 : * Append a text to str.
3108 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3109 : */
3110 : static void
3111 1125529 : appendStringInfoText(StringInfo str, const text *t)
3112 : {
3113 1125529 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
3114 1125529 : }
3115 :
3116 : /*
3117 : * replace_text
3118 : * replace all occurrences of 'old_sub_str' in 'orig_str'
3119 : * with 'new_sub_str' to form 'new_str'
3120 : *
3121 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3122 : * otherwise returns 'new_str'
3123 : */
3124 : Datum
3125 828 : replace_text(PG_FUNCTION_ARGS)
3126 : {
3127 828 : text *src_text = PG_GETARG_TEXT_PP(0);
3128 828 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
3129 828 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
3130 : int src_text_len;
3131 : int from_sub_text_len;
3132 : TextPositionState state;
3133 : text *ret_text;
3134 : int chunk_len;
3135 : char *curr_ptr;
3136 : char *start_ptr;
3137 : StringInfoData str;
3138 : bool found;
3139 :
3140 828 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
3141 828 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
3142 :
3143 : /* Return unmodified source string if empty source or pattern */
3144 828 : if (src_text_len < 1 || from_sub_text_len < 1)
3145 : {
3146 0 : PG_RETURN_TEXT_P(src_text);
3147 : }
3148 :
3149 828 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
3150 :
3151 828 : found = text_position_next(&state);
3152 :
3153 : /* When the from_sub_text is not found, there is nothing to do. */
3154 828 : if (!found)
3155 : {
3156 170 : text_position_cleanup(&state);
3157 170 : PG_RETURN_TEXT_P(src_text);
3158 : }
3159 658 : curr_ptr = text_position_get_match_ptr(&state);
3160 658 : start_ptr = VARDATA_ANY(src_text);
3161 :
3162 658 : initStringInfo(&str);
3163 :
3164 : do
3165 : {
3166 3690 : CHECK_FOR_INTERRUPTS();
3167 :
3168 : /* copy the data skipped over by last text_position_next() */
3169 3690 : chunk_len = curr_ptr - start_ptr;
3170 3690 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3171 :
3172 3690 : appendStringInfoText(&str, to_sub_text);
3173 :
3174 3690 : start_ptr = curr_ptr + state.last_match_len;
3175 :
3176 3690 : found = text_position_next(&state);
3177 3690 : if (found)
3178 3032 : curr_ptr = text_position_get_match_ptr(&state);
3179 : }
3180 3690 : while (found);
3181 :
3182 : /* copy trailing data */
3183 658 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3184 658 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3185 :
3186 658 : text_position_cleanup(&state);
3187 :
3188 658 : ret_text = cstring_to_text_with_len(str.data, str.len);
3189 658 : pfree(str.data);
3190 :
3191 658 : PG_RETURN_TEXT_P(ret_text);
3192 : }
3193 :
3194 : /*
3195 : * check_replace_text_has_escape
3196 : *
3197 : * Returns 0 if text contains no backslashes that need processing.
3198 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
3199 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
3200 : */
3201 : static int
3202 12172 : check_replace_text_has_escape(const text *replace_text)
3203 : {
3204 12172 : int result = 0;
3205 12172 : const char *p = VARDATA_ANY(replace_text);
3206 12172 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3207 :
3208 24372 : while (p < p_end)
3209 : {
3210 : /* Find next escape char, if any. */
3211 11609 : p = memchr(p, '\\', p_end - p);
3212 11609 : if (p == NULL)
3213 11062 : break;
3214 547 : p++;
3215 : /* Note: a backslash at the end doesn't require extra processing. */
3216 547 : if (p < p_end)
3217 : {
3218 547 : if (*p >= '1' && *p <= '9')
3219 519 : return 2; /* Found a submatch specifier, so done */
3220 28 : result = 1; /* Found some other sequence, keep looking */
3221 28 : p++;
3222 : }
3223 : }
3224 11653 : return result;
3225 : }
3226 :
3227 : /*
3228 : * appendStringInfoRegexpSubstr
3229 : *
3230 : * Append replace_text to str, substituting regexp back references for
3231 : * \n escapes. start_ptr is the start of the match in the source string,
3232 : * at logical character position data_pos.
3233 : */
3234 : static void
3235 169 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
3236 : regmatch_t *pmatch,
3237 : char *start_ptr, int data_pos)
3238 : {
3239 169 : const char *p = VARDATA_ANY(replace_text);
3240 169 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3241 :
3242 402 : while (p < p_end)
3243 : {
3244 360 : const char *chunk_start = p;
3245 : int so;
3246 : int eo;
3247 :
3248 : /* Find next escape char, if any. */
3249 360 : p = memchr(p, '\\', p_end - p);
3250 360 : if (p == NULL)
3251 122 : p = p_end;
3252 :
3253 : /* Copy the text we just scanned over, if any. */
3254 360 : if (p > chunk_start)
3255 225 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3256 :
3257 : /* Done if at end of string, else advance over escape char. */
3258 360 : if (p >= p_end)
3259 122 : break;
3260 238 : p++;
3261 :
3262 238 : if (p >= p_end)
3263 : {
3264 : /* Escape at very end of input. Treat same as unexpected char */
3265 5 : appendStringInfoChar(str, '\\');
3266 5 : break;
3267 : }
3268 :
3269 233 : if (*p >= '1' && *p <= '9')
3270 189 : {
3271 : /* Use the back reference of regexp. */
3272 189 : int idx = *p - '0';
3273 :
3274 189 : so = pmatch[idx].rm_so;
3275 189 : eo = pmatch[idx].rm_eo;
3276 189 : p++;
3277 : }
3278 44 : else if (*p == '&')
3279 : {
3280 : /* Use the entire matched string. */
3281 15 : so = pmatch[0].rm_so;
3282 15 : eo = pmatch[0].rm_eo;
3283 15 : p++;
3284 : }
3285 29 : else if (*p == '\\')
3286 : {
3287 : /* \\ means transfer one \ to output. */
3288 24 : appendStringInfoChar(str, '\\');
3289 24 : p++;
3290 24 : continue;
3291 : }
3292 : else
3293 : {
3294 : /*
3295 : * If escape char is not followed by any expected char, just treat
3296 : * it as ordinary data to copy. (XXX would it be better to throw
3297 : * an error?)
3298 : */
3299 5 : appendStringInfoChar(str, '\\');
3300 5 : continue;
3301 : }
3302 :
3303 204 : if (so >= 0 && eo >= 0)
3304 : {
3305 : /*
3306 : * Copy the text that is back reference of regexp. Note so and eo
3307 : * are counted in characters not bytes.
3308 : */
3309 : char *chunk_start;
3310 : int chunk_len;
3311 :
3312 : Assert(so >= data_pos);
3313 204 : chunk_start = start_ptr;
3314 204 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3315 204 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3316 204 : appendBinaryStringInfo(str, chunk_start, chunk_len);
3317 : }
3318 : }
3319 169 : }
3320 :
3321 : /*
3322 : * replace_text_regexp
3323 : *
3324 : * replace substring(s) in src_text that match pattern with replace_text.
3325 : * The replace_text can contain backslash markers to substitute
3326 : * (parts of) the matched text.
3327 : *
3328 : * cflags: regexp compile flags.
3329 : * collation: collation to use.
3330 : * search_start: the character (not byte) offset in src_text at which to
3331 : * begin searching.
3332 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
3333 : */
3334 : text *
3335 12172 : replace_text_regexp(text *src_text, text *pattern_text,
3336 : text *replace_text,
3337 : int cflags, Oid collation,
3338 : int search_start, int n)
3339 : {
3340 : text *ret_text;
3341 : regex_t *re;
3342 12172 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3343 12172 : int nmatches = 0;
3344 : StringInfoData buf;
3345 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
3346 12172 : int nmatch = lengthof(pmatch);
3347 : pg_wchar *data;
3348 : size_t data_len;
3349 : int data_pos;
3350 : char *start_ptr;
3351 : int escape_status;
3352 :
3353 12172 : initStringInfo(&buf);
3354 :
3355 : /* Convert data string to wide characters. */
3356 12172 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3357 12172 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3358 :
3359 : /* Check whether replace_text has escapes, especially regexp submatches. */
3360 12172 : escape_status = check_replace_text_has_escape(replace_text);
3361 :
3362 : /* If no regexp submatches, we can use REG_NOSUB. */
3363 12172 : if (escape_status < 2)
3364 : {
3365 11653 : cflags |= REG_NOSUB;
3366 : /* Also tell pg_regexec we only want the whole-match location. */
3367 11653 : nmatch = 1;
3368 : }
3369 :
3370 : /* Prepare the regexp. */
3371 12172 : re = RE_compile_and_cache(pattern_text, cflags, collation);
3372 :
3373 : /* start_ptr points to the data_pos'th character of src_text */
3374 12172 : start_ptr = (char *) VARDATA_ANY(src_text);
3375 12172 : data_pos = 0;
3376 :
3377 16313 : while (search_start <= data_len)
3378 : {
3379 : int regexec_result;
3380 :
3381 16308 : CHECK_FOR_INTERRUPTS();
3382 :
3383 16308 : regexec_result = pg_regexec(re,
3384 : data,
3385 : data_len,
3386 : search_start,
3387 : NULL, /* no details */
3388 : nmatch,
3389 : pmatch,
3390 : 0);
3391 :
3392 16308 : if (regexec_result == REG_NOMATCH)
3393 10788 : break;
3394 :
3395 5520 : if (regexec_result != REG_OKAY)
3396 : {
3397 : char errMsg[100];
3398 :
3399 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3400 0 : ereport(ERROR,
3401 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3402 : errmsg("regular expression failed: %s", errMsg)));
3403 : }
3404 :
3405 : /*
3406 : * Count matches, and decide whether to replace this match.
3407 : */
3408 5520 : nmatches++;
3409 5520 : if (n > 0 && nmatches != n)
3410 : {
3411 : /*
3412 : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
3413 : * we treat the matched text as if it weren't matched, and copy it
3414 : * to the output later.)
3415 : */
3416 50 : search_start = pmatch[0].rm_eo;
3417 50 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3418 0 : search_start++;
3419 50 : continue;
3420 : }
3421 :
3422 : /*
3423 : * Copy the text to the left of the match position. Note we are given
3424 : * character not byte indexes.
3425 : */
3426 5470 : if (pmatch[0].rm_so - data_pos > 0)
3427 : {
3428 : int chunk_len;
3429 :
3430 5355 : chunk_len = charlen_to_bytelen(start_ptr,
3431 5355 : pmatch[0].rm_so - data_pos);
3432 5355 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3433 :
3434 : /*
3435 : * Advance start_ptr over that text, to avoid multiple rescans of
3436 : * it if the replace_text contains multiple back-references.
3437 : */
3438 5355 : start_ptr += chunk_len;
3439 5355 : data_pos = pmatch[0].rm_so;
3440 : }
3441 :
3442 : /*
3443 : * Copy the replace_text, processing escapes if any are present.
3444 : */
3445 5470 : if (escape_status > 0)
3446 169 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3447 : start_ptr, data_pos);
3448 : else
3449 5301 : appendStringInfoText(&buf, replace_text);
3450 :
3451 : /* Advance start_ptr and data_pos over the matched text. */
3452 10940 : start_ptr += charlen_to_bytelen(start_ptr,
3453 5470 : pmatch[0].rm_eo - data_pos);
3454 5470 : data_pos = pmatch[0].rm_eo;
3455 :
3456 : /*
3457 : * If we only want to replace one occurrence, we're done.
3458 : */
3459 5470 : if (n > 0)
3460 1379 : break;
3461 :
3462 : /*
3463 : * Advance search position. Normally we start the next search at the
3464 : * end of the previous match; but if the match was of zero length, we
3465 : * have to advance by one character, or we'd just find the same match
3466 : * again.
3467 : */
3468 4091 : search_start = data_pos;
3469 4091 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3470 10 : search_start++;
3471 : }
3472 :
3473 : /*
3474 : * Copy the text to the right of the last match.
3475 : */
3476 12172 : if (data_pos < data_len)
3477 : {
3478 : int chunk_len;
3479 :
3480 11578 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3481 11578 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3482 : }
3483 :
3484 12172 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
3485 12172 : pfree(buf.data);
3486 12172 : pfree(data);
3487 :
3488 12172 : return ret_text;
3489 : }
3490 :
3491 : /*
3492 : * split_part
3493 : * parse input string based on provided field separator
3494 : * return N'th item (1 based, negative counts from end)
3495 : */
3496 : Datum
3497 116 : split_part(PG_FUNCTION_ARGS)
3498 : {
3499 116 : text *inputstring = PG_GETARG_TEXT_PP(0);
3500 116 : text *fldsep = PG_GETARG_TEXT_PP(1);
3501 116 : int fldnum = PG_GETARG_INT32(2);
3502 : int inputstring_len;
3503 : int fldsep_len;
3504 : TextPositionState state;
3505 : char *start_ptr;
3506 : char *end_ptr;
3507 : text *result_text;
3508 : bool found;
3509 :
3510 : /* field number is 1 based */
3511 116 : if (fldnum == 0)
3512 4 : ereport(ERROR,
3513 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3514 : errmsg("field position must not be zero")));
3515 :
3516 112 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3517 112 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3518 :
3519 : /* return empty string for empty input string */
3520 112 : if (inputstring_len < 1)
3521 10 : PG_RETURN_TEXT_P(cstring_to_text(""));
3522 :
3523 : /* handle empty field separator */
3524 102 : if (fldsep_len < 1)
3525 : {
3526 : /* if first or last field, return input string, else empty string */
3527 20 : if (fldnum == 1 || fldnum == -1)
3528 10 : PG_RETURN_TEXT_P(inputstring);
3529 : else
3530 10 : PG_RETURN_TEXT_P(cstring_to_text(""));
3531 : }
3532 :
3533 : /* find the first field separator */
3534 82 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
3535 :
3536 82 : found = text_position_next(&state);
3537 :
3538 : /* special case if fldsep not found at all */
3539 82 : if (!found)
3540 : {
3541 18 : text_position_cleanup(&state);
3542 : /* if first or last field, return input string, else empty string */
3543 18 : if (fldnum == 1 || fldnum == -1)
3544 9 : PG_RETURN_TEXT_P(inputstring);
3545 : else
3546 9 : PG_RETURN_TEXT_P(cstring_to_text(""));
3547 : }
3548 :
3549 : /*
3550 : * take care of a negative field number (i.e. count from the right) by
3551 : * converting to a positive field number; we need total number of fields
3552 : */
3553 64 : if (fldnum < 0)
3554 : {
3555 : /* we found a fldsep, so there are at least two fields */
3556 32 : int numfields = 2;
3557 :
3558 42 : while (text_position_next(&state))
3559 10 : numfields++;
3560 :
3561 : /* special case of last field does not require an extra pass */
3562 32 : if (fldnum == -1)
3563 : {
3564 17 : start_ptr = text_position_get_match_ptr(&state) + state.last_match_len;
3565 17 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
3566 17 : text_position_cleanup(&state);
3567 17 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
3568 : end_ptr - start_ptr));
3569 : }
3570 :
3571 : /* else, convert fldnum to positive notation */
3572 15 : fldnum += numfields + 1;
3573 :
3574 : /* if nonexistent field, return empty string */
3575 15 : if (fldnum <= 0)
3576 : {
3577 5 : text_position_cleanup(&state);
3578 5 : PG_RETURN_TEXT_P(cstring_to_text(""));
3579 : }
3580 :
3581 : /* reset to pointing at first match, but now with positive fldnum */
3582 10 : text_position_reset(&state);
3583 10 : found = text_position_next(&state);
3584 : Assert(found);
3585 : }
3586 :
3587 : /* identify bounds of first field */
3588 42 : start_ptr = VARDATA_ANY(inputstring);
3589 42 : end_ptr = text_position_get_match_ptr(&state);
3590 :
3591 79 : while (found && --fldnum > 0)
3592 : {
3593 : /* identify bounds of next field */
3594 37 : start_ptr = end_ptr + state.last_match_len;
3595 37 : found = text_position_next(&state);
3596 37 : if (found)
3597 15 : end_ptr = text_position_get_match_ptr(&state);
3598 : }
3599 :
3600 42 : text_position_cleanup(&state);
3601 :
3602 42 : if (fldnum > 0)
3603 : {
3604 : /* N'th field separator not found */
3605 : /* if last field requested, return it, else empty string */
3606 22 : if (fldnum == 1)
3607 : {
3608 17 : int last_len = start_ptr - VARDATA_ANY(inputstring);
3609 :
3610 17 : result_text = cstring_to_text_with_len(start_ptr,
3611 : inputstring_len - last_len);
3612 : }
3613 : else
3614 5 : result_text = cstring_to_text("");
3615 : }
3616 : else
3617 : {
3618 : /* non-last field requested */
3619 20 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
3620 : }
3621 :
3622 42 : PG_RETURN_TEXT_P(result_text);
3623 : }
3624 :
3625 : /*
3626 : * Convenience function to return true when two text params are equal.
3627 : */
3628 : static bool
3629 306 : text_isequal(text *txt1, text *txt2, Oid collid)
3630 : {
3631 306 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
3632 : collid,
3633 : PointerGetDatum(txt1),
3634 : PointerGetDatum(txt2)));
3635 : }
3636 :
3637 : /*
3638 : * text_to_array
3639 : * parse input string and return text array of elements,
3640 : * based on provided field separator
3641 : */
3642 : Datum
3643 136 : text_to_array(PG_FUNCTION_ARGS)
3644 : {
3645 : SplitTextOutputData tstate;
3646 :
3647 : /* For array output, tstate should start as all zeroes */
3648 136 : memset(&tstate, 0, sizeof(tstate));
3649 :
3650 136 : if (!split_text(fcinfo, &tstate))
3651 5 : PG_RETURN_NULL();
3652 :
3653 131 : if (tstate.astate == NULL)
3654 5 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
3655 :
3656 126 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
3657 : CurrentMemoryContext));
3658 : }
3659 :
3660 : /*
3661 : * text_to_array_null
3662 : * parse input string and return text array of elements,
3663 : * based on provided field separator and null string
3664 : *
3665 : * This is a separate entry point only to prevent the regression tests from
3666 : * complaining about different argument sets for the same internal function.
3667 : */
3668 : Datum
3669 50 : text_to_array_null(PG_FUNCTION_ARGS)
3670 : {
3671 50 : return text_to_array(fcinfo);
3672 : }
3673 :
3674 : /*
3675 : * text_to_table
3676 : * parse input string and return table of elements,
3677 : * based on provided field separator
3678 : */
3679 : Datum
3680 56 : text_to_table(PG_FUNCTION_ARGS)
3681 : {
3682 56 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
3683 : SplitTextOutputData tstate;
3684 :
3685 56 : tstate.astate = NULL;
3686 56 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
3687 56 : tstate.tupstore = rsi->setResult;
3688 56 : tstate.tupdesc = rsi->setDesc;
3689 :
3690 56 : (void) split_text(fcinfo, &tstate);
3691 :
3692 56 : return (Datum) 0;
3693 : }
3694 :
3695 : /*
3696 : * text_to_table_null
3697 : * parse input string and return table of elements,
3698 : * based on provided field separator and null string
3699 : *
3700 : * This is a separate entry point only to prevent the regression tests from
3701 : * complaining about different argument sets for the same internal function.
3702 : */
3703 : Datum
3704 16 : text_to_table_null(PG_FUNCTION_ARGS)
3705 : {
3706 16 : return text_to_table(fcinfo);
3707 : }
3708 :
3709 : /*
3710 : * Common code for text_to_array, text_to_array_null, text_to_table
3711 : * and text_to_table_null functions.
3712 : *
3713 : * These are not strict so we have to test for null inputs explicitly.
3714 : * Returns false if result is to be null, else returns true.
3715 : *
3716 : * Note that if the result is valid but empty (zero elements), we return
3717 : * without changing *tstate --- caller must handle that case, too.
3718 : */
3719 : static bool
3720 192 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
3721 : {
3722 : text *inputstring;
3723 : text *fldsep;
3724 : text *null_string;
3725 192 : Oid collation = PG_GET_COLLATION();
3726 : int inputstring_len;
3727 : int fldsep_len;
3728 : char *start_ptr;
3729 : text *result_text;
3730 :
3731 : /* when input string is NULL, then result is NULL too */
3732 192 : if (PG_ARGISNULL(0))
3733 9 : return false;
3734 :
3735 183 : inputstring = PG_GETARG_TEXT_PP(0);
3736 :
3737 : /* fldsep can be NULL */
3738 183 : if (!PG_ARGISNULL(1))
3739 159 : fldsep = PG_GETARG_TEXT_PP(1);
3740 : else
3741 24 : fldsep = NULL;
3742 :
3743 : /* null_string can be NULL or omitted */
3744 183 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
3745 66 : null_string = PG_GETARG_TEXT_PP(2);
3746 : else
3747 117 : null_string = NULL;
3748 :
3749 183 : if (fldsep != NULL)
3750 : {
3751 : /*
3752 : * Normal case with non-null fldsep. Use the text_position machinery
3753 : * to search for occurrences of fldsep.
3754 : */
3755 : TextPositionState state;
3756 :
3757 159 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3758 159 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3759 :
3760 : /* return empty set for empty input string */
3761 159 : if (inputstring_len < 1)
3762 46 : return true;
3763 :
3764 : /* empty field separator: return input string as a one-element set */
3765 150 : if (fldsep_len < 1)
3766 : {
3767 37 : split_text_accum_result(tstate, inputstring,
3768 : null_string, collation);
3769 37 : return true;
3770 : }
3771 :
3772 113 : text_position_setup(inputstring, fldsep, collation, &state);
3773 :
3774 113 : start_ptr = VARDATA_ANY(inputstring);
3775 :
3776 : for (;;)
3777 371 : {
3778 : bool found;
3779 : char *end_ptr;
3780 : int chunk_len;
3781 :
3782 484 : CHECK_FOR_INTERRUPTS();
3783 :
3784 484 : found = text_position_next(&state);
3785 484 : if (!found)
3786 : {
3787 : /* fetch last field */
3788 113 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
3789 113 : end_ptr = NULL; /* not used, but some compilers complain */
3790 : }
3791 : else
3792 : {
3793 : /* fetch non-last field */
3794 371 : end_ptr = text_position_get_match_ptr(&state);
3795 371 : chunk_len = end_ptr - start_ptr;
3796 : }
3797 :
3798 : /* build a temp text datum to pass to split_text_accum_result */
3799 484 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3800 :
3801 : /* stash away this field */
3802 484 : split_text_accum_result(tstate, result_text,
3803 : null_string, collation);
3804 :
3805 484 : pfree(result_text);
3806 :
3807 484 : if (!found)
3808 113 : break;
3809 :
3810 371 : start_ptr = end_ptr + state.last_match_len;
3811 : }
3812 :
3813 113 : text_position_cleanup(&state);
3814 : }
3815 : else
3816 : {
3817 : const char *end_ptr;
3818 :
3819 : /*
3820 : * When fldsep is NULL, each character in the input string becomes a
3821 : * separate element in the result set. The separator is effectively
3822 : * the space between characters.
3823 : */
3824 24 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3825 :
3826 24 : start_ptr = VARDATA_ANY(inputstring);
3827 24 : end_ptr = start_ptr + inputstring_len;
3828 :
3829 204 : while (inputstring_len > 0)
3830 : {
3831 180 : int chunk_len = pg_mblen_range(start_ptr, end_ptr);
3832 :
3833 180 : CHECK_FOR_INTERRUPTS();
3834 :
3835 : /* build a temp text datum to pass to split_text_accum_result */
3836 180 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3837 :
3838 : /* stash away this field */
3839 180 : split_text_accum_result(tstate, result_text,
3840 : null_string, collation);
3841 :
3842 180 : pfree(result_text);
3843 :
3844 180 : start_ptr += chunk_len;
3845 180 : inputstring_len -= chunk_len;
3846 : }
3847 : }
3848 :
3849 137 : return true;
3850 : }
3851 :
3852 : /*
3853 : * Add text item to result set (table or array).
3854 : *
3855 : * This is also responsible for checking to see if the item matches
3856 : * the null_string, in which case we should emit NULL instead.
3857 : */
3858 : static void
3859 701 : split_text_accum_result(SplitTextOutputData *tstate,
3860 : text *field_value,
3861 : text *null_string,
3862 : Oid collation)
3863 : {
3864 701 : bool is_null = false;
3865 :
3866 701 : if (null_string && text_isequal(field_value, null_string, collation))
3867 56 : is_null = true;
3868 :
3869 701 : if (tstate->tupstore)
3870 : {
3871 : Datum values[1];
3872 : bool nulls[1];
3873 :
3874 152 : values[0] = PointerGetDatum(field_value);
3875 152 : nulls[0] = is_null;
3876 :
3877 152 : tuplestore_putvalues(tstate->tupstore,
3878 : tstate->tupdesc,
3879 : values,
3880 : nulls);
3881 : }
3882 : else
3883 : {
3884 549 : tstate->astate = accumArrayResult(tstate->astate,
3885 : PointerGetDatum(field_value),
3886 : is_null,
3887 : TEXTOID,
3888 : CurrentMemoryContext);
3889 : }
3890 701 : }
3891 :
3892 : /*
3893 : * array_to_text
3894 : * concatenate Cstring representation of input array elements
3895 : * using provided field separator
3896 : */
3897 : Datum
3898 44343 : array_to_text(PG_FUNCTION_ARGS)
3899 : {
3900 44343 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
3901 44343 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3902 :
3903 44343 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
3904 : }
3905 :
3906 : /*
3907 : * array_to_text_null
3908 : * concatenate Cstring representation of input array elements
3909 : * using provided field separator and null string
3910 : *
3911 : * This version is not strict so we have to test for null inputs explicitly.
3912 : */
3913 : Datum
3914 8 : array_to_text_null(PG_FUNCTION_ARGS)
3915 : {
3916 : ArrayType *v;
3917 : char *fldsep;
3918 : char *null_string;
3919 :
3920 : /* returns NULL when first or second parameter is NULL */
3921 8 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
3922 0 : PG_RETURN_NULL();
3923 :
3924 8 : v = PG_GETARG_ARRAYTYPE_P(0);
3925 8 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3926 :
3927 : /* NULL null string is passed through as a null pointer */
3928 8 : if (!PG_ARGISNULL(2))
3929 4 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
3930 : else
3931 4 : null_string = NULL;
3932 :
3933 8 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
3934 : }
3935 :
3936 : /*
3937 : * common code for array_to_text and array_to_text_null functions
3938 : */
3939 : static text *
3940 44363 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
3941 : const char *fldsep, const char *null_string)
3942 : {
3943 : text *result;
3944 : int nitems,
3945 : *dims,
3946 : ndims;
3947 : Oid element_type;
3948 : int typlen;
3949 : bool typbyval;
3950 : char typalign;
3951 : uint8 typalignby;
3952 : StringInfoData buf;
3953 44363 : bool printed = false;
3954 : char *p;
3955 : bits8 *bitmap;
3956 : int bitmask;
3957 : int i;
3958 : ArrayMetaState *my_extra;
3959 :
3960 44363 : ndims = ARR_NDIM(v);
3961 44363 : dims = ARR_DIMS(v);
3962 44363 : nitems = ArrayGetNItems(ndims, dims);
3963 :
3964 : /* if there are no elements, return an empty string */
3965 44363 : if (nitems == 0)
3966 27712 : return cstring_to_text_with_len("", 0);
3967 :
3968 16651 : element_type = ARR_ELEMTYPE(v);
3969 16651 : initStringInfo(&buf);
3970 :
3971 : /*
3972 : * We arrange to look up info about element type, including its output
3973 : * conversion proc, only once per series of calls, assuming the element
3974 : * type doesn't change underneath us.
3975 : */
3976 16651 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3977 16651 : if (my_extra == NULL)
3978 : {
3979 884 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
3980 : sizeof(ArrayMetaState));
3981 884 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3982 884 : my_extra->element_type = ~element_type;
3983 : }
3984 :
3985 16651 : if (my_extra->element_type != element_type)
3986 : {
3987 : /*
3988 : * Get info about element type, including its output conversion proc
3989 : */
3990 884 : get_type_io_data(element_type, IOFunc_output,
3991 : &my_extra->typlen, &my_extra->typbyval,
3992 : &my_extra->typalign, &my_extra->typdelim,
3993 : &my_extra->typioparam, &my_extra->typiofunc);
3994 884 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
3995 884 : fcinfo->flinfo->fn_mcxt);
3996 884 : my_extra->element_type = element_type;
3997 : }
3998 16651 : typlen = my_extra->typlen;
3999 16651 : typbyval = my_extra->typbyval;
4000 16651 : typalign = my_extra->typalign;
4001 16651 : typalignby = typalign_to_alignby(typalign);
4002 :
4003 16651 : p = ARR_DATA_PTR(v);
4004 16651 : bitmap = ARR_NULLBITMAP(v);
4005 16651 : bitmask = 1;
4006 :
4007 56832 : for (i = 0; i < nitems; i++)
4008 : {
4009 : Datum itemvalue;
4010 : char *value;
4011 :
4012 : /* Get source element, checking for NULL */
4013 40181 : if (bitmap && (*bitmap & bitmask) == 0)
4014 : {
4015 : /* if null_string is NULL, we just ignore null elements */
4016 12 : if (null_string != NULL)
4017 : {
4018 4 : if (printed)
4019 4 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
4020 : else
4021 0 : appendStringInfoString(&buf, null_string);
4022 4 : printed = true;
4023 : }
4024 : }
4025 : else
4026 : {
4027 40169 : itemvalue = fetch_att(p, typbyval, typlen);
4028 :
4029 40169 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
4030 :
4031 40169 : if (printed)
4032 23518 : appendStringInfo(&buf, "%s%s", fldsep, value);
4033 : else
4034 16651 : appendStringInfoString(&buf, value);
4035 40169 : printed = true;
4036 :
4037 40169 : p = att_addlength_pointer(p, typlen, p);
4038 40169 : p = (char *) att_nominal_alignby(p, typalignby);
4039 : }
4040 :
4041 : /* advance bitmap pointer if any */
4042 40181 : if (bitmap)
4043 : {
4044 72 : bitmask <<= 1;
4045 72 : if (bitmask == 0x100)
4046 : {
4047 0 : bitmap++;
4048 0 : bitmask = 1;
4049 : }
4050 : }
4051 : }
4052 :
4053 16651 : result = cstring_to_text_with_len(buf.data, buf.len);
4054 16651 : pfree(buf.data);
4055 :
4056 16651 : return result;
4057 : }
4058 :
4059 : /*
4060 : * Workhorse for to_bin, to_oct, and to_hex. Note that base must be > 1 and <=
4061 : * 16.
4062 : */
4063 : static inline text *
4064 20389 : convert_to_base(uint64 value, int base)
4065 : {
4066 20389 : const char *digits = "0123456789abcdef";
4067 :
4068 : /* We size the buffer for to_bin's longest possible return value. */
4069 : char buf[sizeof(uint64) * BITS_PER_BYTE];
4070 20389 : char *const end = buf + sizeof(buf);
4071 20389 : char *ptr = end;
4072 :
4073 : Assert(base > 1);
4074 : Assert(base <= 16);
4075 :
4076 : do
4077 : {
4078 40307 : *--ptr = digits[value % base];
4079 40307 : value /= base;
4080 40307 : } while (ptr > buf && value);
4081 :
4082 20389 : return cstring_to_text_with_len(ptr, end - ptr);
4083 : }
4084 :
4085 : /*
4086 : * Convert an integer to a string containing a base-2 (binary) representation
4087 : * of the number.
4088 : */
4089 : Datum
4090 10 : to_bin32(PG_FUNCTION_ARGS)
4091 : {
4092 10 : uint64 value = (uint32) PG_GETARG_INT32(0);
4093 :
4094 10 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4095 : }
4096 : Datum
4097 10 : to_bin64(PG_FUNCTION_ARGS)
4098 : {
4099 10 : uint64 value = (uint64) PG_GETARG_INT64(0);
4100 :
4101 10 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4102 : }
4103 :
4104 : /*
4105 : * Convert an integer to a string containing a base-8 (oct) representation of
4106 : * the number.
4107 : */
4108 : Datum
4109 10 : to_oct32(PG_FUNCTION_ARGS)
4110 : {
4111 10 : uint64 value = (uint32) PG_GETARG_INT32(0);
4112 :
4113 10 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4114 : }
4115 : Datum
4116 10 : to_oct64(PG_FUNCTION_ARGS)
4117 : {
4118 10 : uint64 value = (uint64) PG_GETARG_INT64(0);
4119 :
4120 10 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4121 : }
4122 :
4123 : /*
4124 : * Convert an integer to a string containing a base-16 (hex) representation of
4125 : * the number.
4126 : */
4127 : Datum
4128 20339 : to_hex32(PG_FUNCTION_ARGS)
4129 : {
4130 20339 : uint64 value = (uint32) PG_GETARG_INT32(0);
4131 :
4132 20339 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4133 : }
4134 : Datum
4135 10 : to_hex64(PG_FUNCTION_ARGS)
4136 : {
4137 10 : uint64 value = (uint64) PG_GETARG_INT64(0);
4138 :
4139 10 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4140 : }
4141 :
4142 : /*
4143 : * Return the size of a datum, possibly compressed
4144 : *
4145 : * Works on any data type
4146 : */
4147 : Datum
4148 71 : pg_column_size(PG_FUNCTION_ARGS)
4149 : {
4150 71 : Datum value = PG_GETARG_DATUM(0);
4151 : int32 result;
4152 : int typlen;
4153 :
4154 : /* On first call, get the input type's typlen, and save at *fn_extra */
4155 71 : if (fcinfo->flinfo->fn_extra == NULL)
4156 : {
4157 : /* Lookup the datatype of the supplied argument */
4158 71 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4159 :
4160 71 : typlen = get_typlen(argtypeid);
4161 71 : if (typlen == 0) /* should not happen */
4162 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4163 :
4164 71 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4165 : sizeof(int));
4166 71 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4167 : }
4168 : else
4169 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4170 :
4171 71 : if (typlen == -1)
4172 : {
4173 : /* varlena type, possibly toasted */
4174 71 : result = toast_datum_size(value);
4175 : }
4176 0 : else if (typlen == -2)
4177 : {
4178 : /* cstring */
4179 0 : result = strlen(DatumGetCString(value)) + 1;
4180 : }
4181 : else
4182 : {
4183 : /* ordinary fixed-width type */
4184 0 : result = typlen;
4185 : }
4186 :
4187 71 : PG_RETURN_INT32(result);
4188 : }
4189 :
4190 : /*
4191 : * Return the compression method stored in the compressed attribute. Return
4192 : * NULL for non varlena type or uncompressed data.
4193 : */
4194 : Datum
4195 128 : pg_column_compression(PG_FUNCTION_ARGS)
4196 : {
4197 : int typlen;
4198 : char *result;
4199 : ToastCompressionId cmid;
4200 :
4201 : /* On first call, get the input type's typlen, and save at *fn_extra */
4202 128 : if (fcinfo->flinfo->fn_extra == NULL)
4203 : {
4204 : /* Lookup the datatype of the supplied argument */
4205 104 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4206 :
4207 104 : typlen = get_typlen(argtypeid);
4208 104 : if (typlen == 0) /* should not happen */
4209 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4210 :
4211 104 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4212 : sizeof(int));
4213 104 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4214 : }
4215 : else
4216 24 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4217 :
4218 128 : if (typlen != -1)
4219 0 : PG_RETURN_NULL();
4220 :
4221 : /* get the compression method id stored in the compressed varlena */
4222 128 : cmid = toast_get_compression_id((varlena *)
4223 128 : DatumGetPointer(PG_GETARG_DATUM(0)));
4224 128 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
4225 28 : PG_RETURN_NULL();
4226 :
4227 : /* convert compression method id to compression method name */
4228 100 : switch (cmid)
4229 : {
4230 56 : case TOAST_PGLZ_COMPRESSION_ID:
4231 56 : result = "pglz";
4232 56 : break;
4233 44 : case TOAST_LZ4_COMPRESSION_ID:
4234 44 : result = "lz4";
4235 44 : break;
4236 0 : default:
4237 0 : elog(ERROR, "invalid compression method id %d", cmid);
4238 : }
4239 :
4240 100 : PG_RETURN_TEXT_P(cstring_to_text(result));
4241 : }
4242 :
4243 : /*
4244 : * Return the chunk_id of the on-disk TOASTed value. Return NULL if the value
4245 : * is un-TOASTed or not on-disk.
4246 : */
4247 : Datum
4248 34 : pg_column_toast_chunk_id(PG_FUNCTION_ARGS)
4249 : {
4250 : int typlen;
4251 : varlena *attr;
4252 : varatt_external toast_pointer;
4253 :
4254 : /* On first call, get the input type's typlen, and save at *fn_extra */
4255 34 : if (fcinfo->flinfo->fn_extra == NULL)
4256 : {
4257 : /* Lookup the datatype of the supplied argument */
4258 26 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4259 :
4260 26 : typlen = get_typlen(argtypeid);
4261 26 : if (typlen == 0) /* should not happen */
4262 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4263 :
4264 26 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4265 : sizeof(int));
4266 26 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4267 : }
4268 : else
4269 8 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4270 :
4271 34 : if (typlen != -1)
4272 0 : PG_RETURN_NULL();
4273 :
4274 34 : attr = (varlena *) DatumGetPointer(PG_GETARG_DATUM(0));
4275 :
4276 34 : if (!VARATT_IS_EXTERNAL_ONDISK(attr))
4277 8 : PG_RETURN_NULL();
4278 :
4279 26 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
4280 :
4281 26 : PG_RETURN_OID(toast_pointer.va_valueid);
4282 : }
4283 :
4284 : /*
4285 : * string_agg - Concatenates values and returns string.
4286 : *
4287 : * Syntax: string_agg(value text, delimiter text) RETURNS text
4288 : *
4289 : * Note: Any NULL values are ignored. The first-call delimiter isn't
4290 : * actually used at all, and on subsequent calls the delimiter precedes
4291 : * the associated value.
4292 : */
4293 :
4294 : /* subroutine to initialize state */
4295 : static StringInfo
4296 1530 : makeStringAggState(FunctionCallInfo fcinfo)
4297 : {
4298 : StringInfo state;
4299 : MemoryContext aggcontext;
4300 : MemoryContext oldcontext;
4301 :
4302 1530 : if (!AggCheckCallContext(fcinfo, &aggcontext))
4303 : {
4304 : /* cannot be called directly because of internal-type argument */
4305 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
4306 : }
4307 :
4308 : /*
4309 : * Create state in aggregate context. It'll stay there across subsequent
4310 : * calls.
4311 : */
4312 1530 : oldcontext = MemoryContextSwitchTo(aggcontext);
4313 1530 : state = makeStringInfo();
4314 1530 : MemoryContextSwitchTo(oldcontext);
4315 :
4316 1530 : return state;
4317 : }
4318 :
4319 : Datum
4320 568301 : string_agg_transfn(PG_FUNCTION_ARGS)
4321 : {
4322 : StringInfo state;
4323 :
4324 568301 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4325 :
4326 : /* Append the value unless null, preceding it with the delimiter. */
4327 568301 : if (!PG_ARGISNULL(1))
4328 : {
4329 558269 : text *value = PG_GETARG_TEXT_PP(1);
4330 558269 : bool isfirst = false;
4331 :
4332 : /*
4333 : * You might think we can just throw away the first delimiter, however
4334 : * we must keep it as we may be a parallel worker doing partial
4335 : * aggregation building a state to send to the main process. We need
4336 : * to keep the delimiter of every aggregation so that the combine
4337 : * function can properly join up the strings of two separately
4338 : * partially aggregated results. The first delimiter is only stripped
4339 : * off in the final function. To know how much to strip off the front
4340 : * of the string, we store the length of the first delimiter in the
4341 : * StringInfo's cursor field, which we don't otherwise need here.
4342 : */
4343 558269 : if (state == NULL)
4344 : {
4345 1310 : state = makeStringAggState(fcinfo);
4346 1310 : isfirst = true;
4347 : }
4348 :
4349 558269 : if (!PG_ARGISNULL(2))
4350 : {
4351 558269 : text *delim = PG_GETARG_TEXT_PP(2);
4352 :
4353 558269 : appendStringInfoText(state, delim);
4354 558269 : if (isfirst)
4355 1310 : state->cursor = VARSIZE_ANY_EXHDR(delim);
4356 : }
4357 :
4358 558269 : appendStringInfoText(state, value);
4359 : }
4360 :
4361 : /*
4362 : * The transition type for string_agg() is declared to be "internal",
4363 : * which is a pass-by-value type the same size as a pointer.
4364 : */
4365 568301 : if (state)
4366 568240 : PG_RETURN_POINTER(state);
4367 61 : PG_RETURN_NULL();
4368 : }
4369 :
4370 : /*
4371 : * string_agg_combine
4372 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
4373 : */
4374 : Datum
4375 140 : string_agg_combine(PG_FUNCTION_ARGS)
4376 : {
4377 : StringInfo state1;
4378 : StringInfo state2;
4379 : MemoryContext agg_context;
4380 :
4381 140 : if (!AggCheckCallContext(fcinfo, &agg_context))
4382 0 : elog(ERROR, "aggregate function called in non-aggregate context");
4383 :
4384 140 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4385 140 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
4386 :
4387 140 : if (state2 == NULL)
4388 : {
4389 : /*
4390 : * NULL state2 is easy, just return state1, which we know is already
4391 : * in the agg_context
4392 : */
4393 0 : if (state1 == NULL)
4394 0 : PG_RETURN_NULL();
4395 0 : PG_RETURN_POINTER(state1);
4396 : }
4397 :
4398 140 : if (state1 == NULL)
4399 : {
4400 : /* We must copy state2's data into the agg_context */
4401 : MemoryContext old_context;
4402 :
4403 80 : old_context = MemoryContextSwitchTo(agg_context);
4404 80 : state1 = makeStringAggState(fcinfo);
4405 80 : appendBinaryStringInfo(state1, state2->data, state2->len);
4406 80 : state1->cursor = state2->cursor;
4407 80 : MemoryContextSwitchTo(old_context);
4408 : }
4409 60 : else if (state2->len > 0)
4410 : {
4411 : /* Combine ... state1->cursor does not change in this case */
4412 60 : appendBinaryStringInfo(state1, state2->data, state2->len);
4413 : }
4414 :
4415 140 : PG_RETURN_POINTER(state1);
4416 : }
4417 :
4418 : /*
4419 : * string_agg_serialize
4420 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
4421 : *
4422 : * This is strict, so we need not handle NULL input
4423 : */
4424 : Datum
4425 140 : string_agg_serialize(PG_FUNCTION_ARGS)
4426 : {
4427 : StringInfo state;
4428 : StringInfoData buf;
4429 : bytea *result;
4430 :
4431 : /* cannot be called directly because of internal-type argument */
4432 : Assert(AggCheckCallContext(fcinfo, NULL));
4433 :
4434 140 : state = (StringInfo) PG_GETARG_POINTER(0);
4435 :
4436 140 : pq_begintypsend(&buf);
4437 :
4438 : /* cursor */
4439 140 : pq_sendint(&buf, state->cursor, 4);
4440 :
4441 : /* data */
4442 140 : pq_sendbytes(&buf, state->data, state->len);
4443 :
4444 140 : result = pq_endtypsend(&buf);
4445 :
4446 140 : PG_RETURN_BYTEA_P(result);
4447 : }
4448 :
4449 : /*
4450 : * string_agg_deserialize
4451 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
4452 : *
4453 : * This is strict, so we need not handle NULL input
4454 : */
4455 : Datum
4456 140 : string_agg_deserialize(PG_FUNCTION_ARGS)
4457 : {
4458 : bytea *sstate;
4459 : StringInfo result;
4460 : StringInfoData buf;
4461 : char *data;
4462 : int datalen;
4463 :
4464 : /* cannot be called directly because of internal-type argument */
4465 : Assert(AggCheckCallContext(fcinfo, NULL));
4466 :
4467 140 : sstate = PG_GETARG_BYTEA_PP(0);
4468 :
4469 : /*
4470 : * Initialize a StringInfo so that we can "receive" it using the standard
4471 : * recv-function infrastructure.
4472 : */
4473 140 : initReadOnlyStringInfo(&buf, VARDATA_ANY(sstate),
4474 140 : VARSIZE_ANY_EXHDR(sstate));
4475 :
4476 140 : result = makeStringAggState(fcinfo);
4477 :
4478 : /* cursor */
4479 140 : result->cursor = pq_getmsgint(&buf, 4);
4480 :
4481 : /* data */
4482 140 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
4483 140 : data = (char *) pq_getmsgbytes(&buf, datalen);
4484 140 : appendBinaryStringInfo(result, data, datalen);
4485 :
4486 140 : pq_getmsgend(&buf);
4487 :
4488 140 : PG_RETURN_POINTER(result);
4489 : }
4490 :
4491 : Datum
4492 1336 : string_agg_finalfn(PG_FUNCTION_ARGS)
4493 : {
4494 : StringInfo state;
4495 :
4496 : /* cannot be called directly because of internal-type argument */
4497 : Assert(AggCheckCallContext(fcinfo, NULL));
4498 :
4499 1336 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4500 :
4501 1336 : if (state != NULL)
4502 : {
4503 : /* As per comment in transfn, strip data before the cursor position */
4504 1280 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
4505 : state->len - state->cursor));
4506 : }
4507 : else
4508 56 : PG_RETURN_NULL();
4509 : }
4510 :
4511 : /*
4512 : * Prepare cache with fmgr info for the output functions of the datatypes of
4513 : * the arguments of a concat-like function, beginning with argument "argidx".
4514 : * (Arguments before that will have corresponding slots in the resulting
4515 : * FmgrInfo array, but we don't fill those slots.)
4516 : */
4517 : static FmgrInfo *
4518 70 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
4519 : {
4520 : FmgrInfo *foutcache;
4521 : int i;
4522 :
4523 : /* We keep the info in fn_mcxt so it survives across calls */
4524 70 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4525 70 : PG_NARGS() * sizeof(FmgrInfo));
4526 :
4527 264 : for (i = argidx; i < PG_NARGS(); i++)
4528 : {
4529 : Oid valtype;
4530 : Oid typOutput;
4531 : bool typIsVarlena;
4532 :
4533 194 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4534 194 : if (!OidIsValid(valtype))
4535 0 : elog(ERROR, "could not determine data type of concat() input");
4536 :
4537 194 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4538 194 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
4539 : }
4540 :
4541 70 : fcinfo->flinfo->fn_extra = foutcache;
4542 :
4543 70 : return foutcache;
4544 : }
4545 :
4546 : /*
4547 : * Implementation of both concat() and concat_ws().
4548 : *
4549 : * sepstr is the separator string to place between values.
4550 : * argidx identifies the first argument to concatenate (counting from zero);
4551 : * note that this must be constant across any one series of calls.
4552 : *
4553 : * Returns NULL if result should be NULL, else text value.
4554 : */
4555 : static text *
4556 175 : concat_internal(const char *sepstr, int argidx,
4557 : FunctionCallInfo fcinfo)
4558 : {
4559 : text *result;
4560 : StringInfoData str;
4561 : FmgrInfo *foutcache;
4562 175 : bool first_arg = true;
4563 : int i;
4564 :
4565 : /*
4566 : * concat(VARIADIC some-array) is essentially equivalent to
4567 : * array_to_text(), ie concat the array elements with the given separator.
4568 : * So we just pass the case off to that code.
4569 : */
4570 175 : if (get_fn_expr_variadic(fcinfo->flinfo))
4571 : {
4572 : ArrayType *arr;
4573 :
4574 : /* Should have just the one argument */
4575 : Assert(argidx == PG_NARGS() - 1);
4576 :
4577 : /* concat(VARIADIC NULL) is defined as NULL */
4578 20 : if (PG_ARGISNULL(argidx))
4579 8 : return NULL;
4580 :
4581 : /*
4582 : * Non-null argument had better be an array. We assume that any call
4583 : * context that could let get_fn_expr_variadic return true will have
4584 : * checked that a VARIADIC-labeled parameter actually is an array. So
4585 : * it should be okay to just Assert that it's an array rather than
4586 : * doing a full-fledged error check.
4587 : */
4588 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
4589 :
4590 : /* OK, safe to fetch the array value */
4591 12 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
4592 :
4593 : /*
4594 : * And serialize the array. We tell array_to_text to ignore null
4595 : * elements, which matches the behavior of the loop below.
4596 : */
4597 12 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4598 : }
4599 :
4600 : /* Normal case without explicit VARIADIC marker */
4601 155 : initStringInfo(&str);
4602 :
4603 : /* Get output function info, building it if first time through */
4604 155 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
4605 155 : if (foutcache == NULL)
4606 70 : foutcache = build_concat_foutcache(fcinfo, argidx);
4607 :
4608 544 : for (i = argidx; i < PG_NARGS(); i++)
4609 : {
4610 389 : if (!PG_ARGISNULL(i))
4611 : {
4612 337 : Datum value = PG_GETARG_DATUM(i);
4613 :
4614 : /* add separator if appropriate */
4615 337 : if (first_arg)
4616 151 : first_arg = false;
4617 : else
4618 186 : appendStringInfoString(&str, sepstr);
4619 :
4620 : /* call the appropriate type output function, append the result */
4621 337 : appendStringInfoString(&str,
4622 337 : OutputFunctionCall(&foutcache[i], value));
4623 : }
4624 : }
4625 :
4626 155 : result = cstring_to_text_with_len(str.data, str.len);
4627 155 : pfree(str.data);
4628 :
4629 155 : return result;
4630 : }
4631 :
4632 : /*
4633 : * Concatenate all arguments. NULL arguments are ignored.
4634 : */
4635 : Datum
4636 123 : text_concat(PG_FUNCTION_ARGS)
4637 : {
4638 : text *result;
4639 :
4640 123 : result = concat_internal("", 0, fcinfo);
4641 123 : if (result == NULL)
4642 4 : PG_RETURN_NULL();
4643 119 : PG_RETURN_TEXT_P(result);
4644 : }
4645 :
4646 : /*
4647 : * Concatenate all but first argument value with separators. The first
4648 : * parameter is used as the separator. NULL arguments are ignored.
4649 : */
4650 : Datum
4651 56 : text_concat_ws(PG_FUNCTION_ARGS)
4652 : {
4653 : char *sep;
4654 : text *result;
4655 :
4656 : /* return NULL when separator is NULL */
4657 56 : if (PG_ARGISNULL(0))
4658 4 : PG_RETURN_NULL();
4659 52 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
4660 :
4661 52 : result = concat_internal(sep, 1, fcinfo);
4662 52 : if (result == NULL)
4663 4 : PG_RETURN_NULL();
4664 48 : PG_RETURN_TEXT_P(result);
4665 : }
4666 :
4667 : /*
4668 : * Return first n characters in the string. When n is negative,
4669 : * return all but last |n| characters.
4670 : */
4671 : Datum
4672 1432 : text_left(PG_FUNCTION_ARGS)
4673 : {
4674 1432 : int n = PG_GETARG_INT32(1);
4675 :
4676 1432 : if (n < 0)
4677 : {
4678 20 : text *str = PG_GETARG_TEXT_PP(0);
4679 20 : const char *p = VARDATA_ANY(str);
4680 20 : int len = VARSIZE_ANY_EXHDR(str);
4681 : int rlen;
4682 :
4683 20 : n = pg_mbstrlen_with_len(p, len) + n;
4684 20 : rlen = pg_mbcharcliplen(p, len, n);
4685 20 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
4686 : }
4687 : else
4688 1412 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
4689 : }
4690 :
4691 : /*
4692 : * Return last n characters in the string. When n is negative,
4693 : * return all but first |n| characters.
4694 : */
4695 : Datum
4696 44 : text_right(PG_FUNCTION_ARGS)
4697 : {
4698 44 : text *str = PG_GETARG_TEXT_PP(0);
4699 44 : const char *p = VARDATA_ANY(str);
4700 44 : int len = VARSIZE_ANY_EXHDR(str);
4701 44 : int n = PG_GETARG_INT32(1);
4702 : int off;
4703 :
4704 44 : if (n < 0)
4705 20 : n = -n;
4706 : else
4707 24 : n = pg_mbstrlen_with_len(p, len) - n;
4708 44 : off = pg_mbcharcliplen(p, len, n);
4709 :
4710 44 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4711 : }
4712 :
4713 : /*
4714 : * Return reversed string
4715 : */
4716 : Datum
4717 29 : text_reverse(PG_FUNCTION_ARGS)
4718 : {
4719 29 : text *str = PG_GETARG_TEXT_PP(0);
4720 29 : const char *p = VARDATA_ANY(str);
4721 29 : int len = VARSIZE_ANY_EXHDR(str);
4722 29 : const char *endp = p + len;
4723 : text *result;
4724 : char *dst;
4725 :
4726 29 : result = palloc(len + VARHDRSZ);
4727 29 : dst = (char *) VARDATA(result) + len;
4728 29 : SET_VARSIZE(result, len + VARHDRSZ);
4729 :
4730 29 : if (pg_database_encoding_max_length() > 1)
4731 : {
4732 : /* multibyte version */
4733 222 : while (p < endp)
4734 : {
4735 : int sz;
4736 :
4737 197 : sz = pg_mblen_range(p, endp);
4738 193 : dst -= sz;
4739 193 : memcpy(dst, p, sz);
4740 193 : p += sz;
4741 : }
4742 : }
4743 : else
4744 : {
4745 : /* single byte version */
4746 0 : while (p < endp)
4747 0 : *(--dst) = *p++;
4748 : }
4749 :
4750 25 : PG_RETURN_TEXT_P(result);
4751 : }
4752 :
4753 :
4754 : /*
4755 : * Support macros for text_format()
4756 : */
4757 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4758 :
4759 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4760 : do { \
4761 : if (++(ptr) >= (end_ptr)) \
4762 : ereport(ERROR, \
4763 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4764 : errmsg("unterminated format() type specifier"), \
4765 : errhint("For a single \"%%\" use \"%%%%\"."))); \
4766 : } while (0)
4767 :
4768 : /*
4769 : * Returns a formatted string
4770 : */
4771 : Datum
4772 21517 : text_format(PG_FUNCTION_ARGS)
4773 : {
4774 : text *fmt;
4775 : StringInfoData str;
4776 : const char *cp;
4777 : const char *start_ptr;
4778 : const char *end_ptr;
4779 : text *result;
4780 : int arg;
4781 : bool funcvariadic;
4782 : int nargs;
4783 21517 : Datum *elements = NULL;
4784 21517 : bool *nulls = NULL;
4785 21517 : Oid element_type = InvalidOid;
4786 21517 : Oid prev_type = InvalidOid;
4787 21517 : Oid prev_width_type = InvalidOid;
4788 : FmgrInfo typoutputfinfo;
4789 : FmgrInfo typoutputinfo_width;
4790 :
4791 : /* When format string is null, immediately return null */
4792 21517 : if (PG_ARGISNULL(0))
4793 4 : PG_RETURN_NULL();
4794 :
4795 : /* If argument is marked VARIADIC, expand array into elements */
4796 21513 : if (get_fn_expr_variadic(fcinfo->flinfo))
4797 : {
4798 : ArrayType *arr;
4799 : int16 elmlen;
4800 : bool elmbyval;
4801 : char elmalign;
4802 : int nitems;
4803 :
4804 : /* Should have just the one argument */
4805 : Assert(PG_NARGS() == 2);
4806 :
4807 : /* If argument is NULL, we treat it as zero-length array */
4808 32 : if (PG_ARGISNULL(1))
4809 4 : nitems = 0;
4810 : else
4811 : {
4812 : /*
4813 : * Non-null argument had better be an array. We assume that any
4814 : * call context that could let get_fn_expr_variadic return true
4815 : * will have checked that a VARIADIC-labeled parameter actually is
4816 : * an array. So it should be okay to just Assert that it's an
4817 : * array rather than doing a full-fledged error check.
4818 : */
4819 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
4820 :
4821 : /* OK, safe to fetch the array value */
4822 28 : arr = PG_GETARG_ARRAYTYPE_P(1);
4823 :
4824 : /* Get info about array element type */
4825 28 : element_type = ARR_ELEMTYPE(arr);
4826 28 : get_typlenbyvalalign(element_type,
4827 : &elmlen, &elmbyval, &elmalign);
4828 :
4829 : /* Extract all array elements */
4830 28 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
4831 : &elements, &nulls, &nitems);
4832 : }
4833 :
4834 32 : nargs = nitems + 1;
4835 32 : funcvariadic = true;
4836 : }
4837 : else
4838 : {
4839 : /* Non-variadic case, we'll process the arguments individually */
4840 21481 : nargs = PG_NARGS();
4841 21481 : funcvariadic = false;
4842 : }
4843 :
4844 : /* Setup for main loop. */
4845 21513 : fmt = PG_GETARG_TEXT_PP(0);
4846 21513 : start_ptr = VARDATA_ANY(fmt);
4847 21513 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
4848 21513 : initStringInfo(&str);
4849 21513 : arg = 1; /* next argument position to print */
4850 :
4851 : /* Scan format string, looking for conversion specifiers. */
4852 658508 : for (cp = start_ptr; cp < end_ptr; cp++)
4853 : {
4854 : int argpos;
4855 : int widthpos;
4856 : int flags;
4857 : int width;
4858 : Datum value;
4859 : bool isNull;
4860 : Oid typid;
4861 :
4862 : /*
4863 : * If it's not the start of a conversion specifier, just copy it to
4864 : * the output buffer.
4865 : */
4866 637035 : if (*cp != '%')
4867 : {
4868 593124 : appendStringInfoCharMacro(&str, *cp);
4869 593136 : continue;
4870 : }
4871 :
4872 43911 : ADVANCE_PARSE_POINTER(cp, end_ptr);
4873 :
4874 : /* Easy case: %% outputs a single % */
4875 43911 : if (*cp == '%')
4876 : {
4877 12 : appendStringInfoCharMacro(&str, *cp);
4878 12 : continue;
4879 : }
4880 :
4881 : /* Parse the optional portions of the format specifier */
4882 43899 : cp = text_format_parse_format(cp, end_ptr,
4883 : &argpos, &widthpos,
4884 : &flags, &width);
4885 :
4886 : /*
4887 : * Next we should see the main conversion specifier. Whether or not
4888 : * an argument position was present, it's known that at least one
4889 : * character remains in the string at this point. Experience suggests
4890 : * that it's worth checking that that character is one of the expected
4891 : * ones before we try to fetch arguments, so as to produce the least
4892 : * confusing response to a mis-formatted specifier.
4893 : */
4894 43883 : if (strchr("sIL", *cp) == NULL)
4895 4 : ereport(ERROR,
4896 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4897 : errmsg("unrecognized format() type specifier \"%.*s\"",
4898 : pg_mblen_range(cp, end_ptr), cp),
4899 : errhint("For a single \"%%\" use \"%%%%\".")));
4900 :
4901 : /* If indirect width was specified, get its value */
4902 43879 : if (widthpos >= 0)
4903 : {
4904 : /* Collect the specified or next argument position */
4905 28 : if (widthpos > 0)
4906 24 : arg = widthpos;
4907 28 : if (arg >= nargs)
4908 0 : ereport(ERROR,
4909 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4910 : errmsg("too few arguments for format()")));
4911 :
4912 : /* Get the value and type of the selected argument */
4913 28 : if (!funcvariadic)
4914 : {
4915 28 : value = PG_GETARG_DATUM(arg);
4916 28 : isNull = PG_ARGISNULL(arg);
4917 28 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4918 : }
4919 : else
4920 : {
4921 0 : value = elements[arg - 1];
4922 0 : isNull = nulls[arg - 1];
4923 0 : typid = element_type;
4924 : }
4925 28 : if (!OidIsValid(typid))
4926 0 : elog(ERROR, "could not determine data type of format() input");
4927 :
4928 28 : arg++;
4929 :
4930 : /* We can treat NULL width the same as zero */
4931 28 : if (isNull)
4932 4 : width = 0;
4933 24 : else if (typid == INT4OID)
4934 24 : width = DatumGetInt32(value);
4935 0 : else if (typid == INT2OID)
4936 0 : width = DatumGetInt16(value);
4937 : else
4938 : {
4939 : /* For less-usual datatypes, convert to text then to int */
4940 : char *str;
4941 :
4942 0 : if (typid != prev_width_type)
4943 : {
4944 : Oid typoutputfunc;
4945 : bool typIsVarlena;
4946 :
4947 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4948 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
4949 0 : prev_width_type = typid;
4950 : }
4951 :
4952 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
4953 :
4954 : /* pg_strtoint32 will complain about bad data or overflow */
4955 0 : width = pg_strtoint32(str);
4956 :
4957 0 : pfree(str);
4958 : }
4959 : }
4960 :
4961 : /* Collect the specified or next argument position */
4962 43879 : if (argpos > 0)
4963 88 : arg = argpos;
4964 43879 : if (arg >= nargs)
4965 16 : ereport(ERROR,
4966 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4967 : errmsg("too few arguments for format()")));
4968 :
4969 : /* Get the value and type of the selected argument */
4970 43863 : if (!funcvariadic)
4971 : {
4972 43015 : value = PG_GETARG_DATUM(arg);
4973 43015 : isNull = PG_ARGISNULL(arg);
4974 43015 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4975 : }
4976 : else
4977 : {
4978 848 : value = elements[arg - 1];
4979 848 : isNull = nulls[arg - 1];
4980 848 : typid = element_type;
4981 : }
4982 43863 : if (!OidIsValid(typid))
4983 0 : elog(ERROR, "could not determine data type of format() input");
4984 :
4985 43863 : arg++;
4986 :
4987 : /*
4988 : * Get the appropriate typOutput function, reusing previous one if
4989 : * same type as previous argument. That's particularly useful in the
4990 : * variadic-array case, but often saves work even for ordinary calls.
4991 : */
4992 43863 : if (typid != prev_type)
4993 : {
4994 : Oid typoutputfunc;
4995 : bool typIsVarlena;
4996 :
4997 22855 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4998 22855 : fmgr_info(typoutputfunc, &typoutputfinfo);
4999 22855 : prev_type = typid;
5000 : }
5001 :
5002 : /*
5003 : * And now we can format the value.
5004 : */
5005 43863 : switch (*cp)
5006 : {
5007 43863 : case 's':
5008 : case 'I':
5009 : case 'L':
5010 43863 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
5011 : value, isNull,
5012 : flags, width);
5013 43859 : break;
5014 0 : default:
5015 : /* should not get here, because of previous check */
5016 0 : ereport(ERROR,
5017 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5018 : errmsg("unrecognized format() type specifier \"%.*s\"",
5019 : pg_mblen_range(cp, end_ptr), cp),
5020 : errhint("For a single \"%%\" use \"%%%%\".")));
5021 : break;
5022 : }
5023 : }
5024 :
5025 : /* Don't need deconstruct_array results anymore. */
5026 21473 : if (elements != NULL)
5027 28 : pfree(elements);
5028 21473 : if (nulls != NULL)
5029 28 : pfree(nulls);
5030 :
5031 : /* Generate results. */
5032 21473 : result = cstring_to_text_with_len(str.data, str.len);
5033 21473 : pfree(str.data);
5034 :
5035 21473 : PG_RETURN_TEXT_P(result);
5036 : }
5037 :
5038 : /*
5039 : * Parse contiguous digits as a decimal number.
5040 : *
5041 : * Returns true if some digits could be parsed.
5042 : * The value is returned into *value, and *ptr is advanced to the next
5043 : * character to be parsed.
5044 : *
5045 : * Note parsing invariant: at least one character is known available before
5046 : * string end (end_ptr) at entry, and this is still true at exit.
5047 : */
5048 : static bool
5049 87774 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5050 : {
5051 87774 : bool found = false;
5052 87774 : const char *cp = *ptr;
5053 87774 : int val = 0;
5054 :
5055 87982 : while (*cp >= '0' && *cp <= '9')
5056 : {
5057 212 : int8 digit = (*cp - '0');
5058 :
5059 212 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5060 212 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5061 0 : ereport(ERROR,
5062 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5063 : errmsg("number is out of range")));
5064 212 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5065 208 : found = true;
5066 : }
5067 :
5068 87770 : *ptr = cp;
5069 87770 : *value = val;
5070 :
5071 87770 : return found;
5072 : }
5073 :
5074 : /*
5075 : * Parse a format specifier (generally following the SUS printf spec).
5076 : *
5077 : * We have already advanced over the initial '%', and we are looking for
5078 : * [argpos][flags][width]type (but the type character is not consumed here).
5079 : *
5080 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5081 : * Output parameters:
5082 : * argpos: argument position for value to be printed. -1 means unspecified.
5083 : * widthpos: argument position for width. Zero means the argument position
5084 : * was unspecified (ie, take the next arg) and -1 means no width
5085 : * argument (width was omitted or specified as a constant).
5086 : * flags: bitmask of flags.
5087 : * width: directly-specified width value. Zero means the width was omitted
5088 : * (note it's not necessary to distinguish this case from an explicit
5089 : * zero width value).
5090 : *
5091 : * The function result is the next character position to be parsed, ie, the
5092 : * location where the type character is/should be.
5093 : *
5094 : * Note parsing invariant: at least one character is known available before
5095 : * string end (end_ptr) at entry, and this is still true at exit.
5096 : */
5097 : static const char *
5098 43899 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5099 : int *argpos, int *widthpos,
5100 : int *flags, int *width)
5101 : {
5102 43899 : const char *cp = start_ptr;
5103 : int n;
5104 :
5105 : /* set defaults for output parameters */
5106 43899 : *argpos = -1;
5107 43899 : *widthpos = -1;
5108 43899 : *flags = 0;
5109 43899 : *width = 0;
5110 :
5111 : /* try to identify first number */
5112 43899 : if (text_format_parse_digits(&cp, end_ptr, &n))
5113 : {
5114 116 : if (*cp != '$')
5115 : {
5116 : /* Must be just a width and a type, so we're done */
5117 16 : *width = n;
5118 16 : return cp;
5119 : }
5120 : /* The number was argument position */
5121 100 : *argpos = n;
5122 : /* Explicit 0 for argument index is immediately refused */
5123 100 : if (n == 0)
5124 4 : ereport(ERROR,
5125 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5126 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5127 96 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5128 : }
5129 :
5130 : /* Handle flags (only minus is supported now) */
5131 43895 : while (*cp == '-')
5132 : {
5133 20 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5134 20 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5135 : }
5136 :
5137 43875 : if (*cp == '*')
5138 : {
5139 : /* Handle indirect width */
5140 32 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5141 32 : if (text_format_parse_digits(&cp, end_ptr, &n))
5142 : {
5143 : /* number in this position must be closed by $ */
5144 28 : if (*cp != '$')
5145 0 : ereport(ERROR,
5146 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5147 : errmsg("width argument position must be ended by \"$\"")));
5148 : /* The number was width argument position */
5149 28 : *widthpos = n;
5150 : /* Explicit 0 for argument index is immediately refused */
5151 28 : if (n == 0)
5152 4 : ereport(ERROR,
5153 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5154 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5155 24 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5156 : }
5157 : else
5158 4 : *widthpos = 0; /* width's argument position is unspecified */
5159 : }
5160 : else
5161 : {
5162 : /* Check for direct width specification */
5163 43843 : if (text_format_parse_digits(&cp, end_ptr, &n))
5164 20 : *width = n;
5165 : }
5166 :
5167 : /* cp should now be pointing at type character */
5168 43867 : return cp;
5169 : }
5170 :
5171 : /*
5172 : * Format a %s, %I, or %L conversion
5173 : */
5174 : static void
5175 43863 : text_format_string_conversion(StringInfo buf, char conversion,
5176 : FmgrInfo *typOutputInfo,
5177 : Datum value, bool isNull,
5178 : int flags, int width)
5179 : {
5180 : char *str;
5181 :
5182 : /* Handle NULL arguments before trying to stringify the value. */
5183 43863 : if (isNull)
5184 : {
5185 228 : if (conversion == 's')
5186 180 : text_format_append_string(buf, "", flags, width);
5187 48 : else if (conversion == 'L')
5188 44 : text_format_append_string(buf, "NULL", flags, width);
5189 4 : else if (conversion == 'I')
5190 4 : ereport(ERROR,
5191 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5192 : errmsg("null values cannot be formatted as an SQL identifier")));
5193 224 : return;
5194 : }
5195 :
5196 : /* Stringify. */
5197 43635 : str = OutputFunctionCall(typOutputInfo, value);
5198 :
5199 : /* Escape. */
5200 43635 : if (conversion == 'I')
5201 : {
5202 : /* quote_identifier may or may not allocate a new string. */
5203 3268 : text_format_append_string(buf, quote_identifier(str), flags, width);
5204 : }
5205 40367 : else if (conversion == 'L')
5206 : {
5207 2166 : char *qstr = quote_literal_cstr(str);
5208 :
5209 2166 : text_format_append_string(buf, qstr, flags, width);
5210 : /* quote_literal_cstr() always allocates a new string */
5211 2166 : pfree(qstr);
5212 : }
5213 : else
5214 38201 : text_format_append_string(buf, str, flags, width);
5215 :
5216 : /* Cleanup. */
5217 43635 : pfree(str);
5218 : }
5219 :
5220 : /*
5221 : * Append str to buf, padding as directed by flags/width
5222 : */
5223 : static void
5224 43859 : text_format_append_string(StringInfo buf, const char *str,
5225 : int flags, int width)
5226 : {
5227 43859 : bool align_to_left = false;
5228 : int len;
5229 :
5230 : /* fast path for typical easy case */
5231 43859 : if (width == 0)
5232 : {
5233 43803 : appendStringInfoString(buf, str);
5234 43803 : return;
5235 : }
5236 :
5237 56 : if (width < 0)
5238 : {
5239 : /* Negative width: implicit '-' flag, then take absolute value */
5240 4 : align_to_left = true;
5241 : /* -INT_MIN is undefined */
5242 4 : if (width <= INT_MIN)
5243 0 : ereport(ERROR,
5244 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5245 : errmsg("number is out of range")));
5246 4 : width = -width;
5247 : }
5248 52 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
5249 16 : align_to_left = true;
5250 :
5251 56 : len = pg_mbstrlen(str);
5252 56 : if (align_to_left)
5253 : {
5254 : /* left justify */
5255 20 : appendStringInfoString(buf, str);
5256 20 : if (len < width)
5257 20 : appendStringInfoSpaces(buf, width - len);
5258 : }
5259 : else
5260 : {
5261 : /* right justify */
5262 36 : if (len < width)
5263 36 : appendStringInfoSpaces(buf, width - len);
5264 36 : appendStringInfoString(buf, str);
5265 : }
5266 : }
5267 :
5268 : /*
5269 : * text_format_nv - nonvariadic wrapper for text_format function.
5270 : *
5271 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5272 : * which checks that all built-in functions that share the implementing C
5273 : * function take the same number of arguments.
5274 : */
5275 : Datum
5276 1910 : text_format_nv(PG_FUNCTION_ARGS)
5277 : {
5278 1910 : return text_format(fcinfo);
5279 : }
5280 :
5281 : /*
5282 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
5283 : * for this use case.
5284 : */
5285 : static inline bool
5286 0 : rest_of_char_same(const char *s1, const char *s2, int len)
5287 : {
5288 0 : while (len > 0)
5289 : {
5290 0 : len--;
5291 0 : if (s1[len] != s2[len])
5292 0 : return false;
5293 : }
5294 0 : return true;
5295 : }
5296 :
5297 : /* Expand each Levenshtein distance variant */
5298 : #include "levenshtein.c"
5299 : #define LEVENSHTEIN_LESS_EQUAL
5300 : #include "levenshtein.c"
5301 :
5302 :
5303 : /*
5304 : * The following *ClosestMatch() functions can be used to determine whether a
5305 : * user-provided string resembles any known valid values, which is useful for
5306 : * providing hints in log messages, among other things. Use these functions
5307 : * like so:
5308 : *
5309 : * initClosestMatch(&state, source_string, max_distance);
5310 : *
5311 : * for (int i = 0; i < num_valid_strings; i++)
5312 : * updateClosestMatch(&state, valid_strings[i]);
5313 : *
5314 : * closestMatch = getClosestMatch(&state);
5315 : */
5316 :
5317 : /*
5318 : * Initialize the given state with the source string and maximum Levenshtein
5319 : * distance to consider.
5320 : */
5321 : void
5322 44 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
5323 : {
5324 : Assert(state);
5325 : Assert(max_d >= 0);
5326 :
5327 44 : state->source = source;
5328 44 : state->min_d = -1;
5329 44 : state->max_d = max_d;
5330 44 : state->match = NULL;
5331 44 : }
5332 :
5333 : /*
5334 : * If the candidate string is a closer match than the current one saved (or
5335 : * there is no match saved), save it as the closest match.
5336 : *
5337 : * If the source or candidate string is NULL, empty, or too long, this function
5338 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
5339 : * allowed or more than half the characters are different, no action is taken.
5340 : */
5341 : void
5342 432 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
5343 : {
5344 : int dist;
5345 :
5346 : Assert(state);
5347 :
5348 432 : if (state->source == NULL || state->source[0] == '\0' ||
5349 432 : candidate == NULL || candidate[0] == '\0')
5350 0 : return;
5351 :
5352 : /*
5353 : * To avoid ERROR-ing, we check the lengths here instead of setting
5354 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
5355 : */
5356 432 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
5357 432 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
5358 0 : return;
5359 :
5360 432 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
5361 432 : candidate, strlen(candidate), 1, 1, 1,
5362 : state->max_d, true);
5363 432 : if (dist <= state->max_d &&
5364 39 : dist <= strlen(state->source) / 2 &&
5365 9 : (state->min_d == -1 || dist < state->min_d))
5366 : {
5367 9 : state->min_d = dist;
5368 9 : state->match = candidate;
5369 : }
5370 : }
5371 :
5372 : /*
5373 : * Return the closest match. If no suitable candidates were provided via
5374 : * updateClosestMatch(), return NULL.
5375 : */
5376 : const char *
5377 44 : getClosestMatch(ClosestMatchState *state)
5378 : {
5379 : Assert(state);
5380 :
5381 44 : return state->match;
5382 : }
5383 :
5384 :
5385 : /*
5386 : * Unicode support
5387 : */
5388 :
5389 : static UnicodeNormalizationForm
5390 149 : unicode_norm_form_from_string(const char *formstr)
5391 : {
5392 149 : UnicodeNormalizationForm form = -1;
5393 :
5394 : /*
5395 : * Might as well check this while we're here.
5396 : */
5397 149 : if (GetDatabaseEncoding() != PG_UTF8)
5398 0 : ereport(ERROR,
5399 : (errcode(ERRCODE_SYNTAX_ERROR),
5400 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
5401 :
5402 149 : if (pg_strcasecmp(formstr, "NFC") == 0)
5403 50 : form = UNICODE_NFC;
5404 99 : else if (pg_strcasecmp(formstr, "NFD") == 0)
5405 41 : form = UNICODE_NFD;
5406 58 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
5407 25 : form = UNICODE_NFKC;
5408 33 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
5409 25 : form = UNICODE_NFKD;
5410 : else
5411 8 : ereport(ERROR,
5412 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5413 : errmsg("invalid normalization form: %s", formstr)));
5414 :
5415 141 : return form;
5416 : }
5417 :
5418 : /*
5419 : * Returns version of Unicode used by Postgres in "major.minor" format (the
5420 : * same format as the Unicode version reported by ICU). The third component
5421 : * ("update version") never involves additions to the character repertoire and
5422 : * is unimportant for most purposes.
5423 : *
5424 : * See: https://unicode.org/versions/
5425 : */
5426 : Datum
5427 20 : unicode_version(PG_FUNCTION_ARGS)
5428 : {
5429 20 : PG_RETURN_TEXT_P(cstring_to_text(PG_UNICODE_VERSION));
5430 : }
5431 :
5432 : /*
5433 : * Returns version of Unicode used by ICU, if enabled; otherwise NULL.
5434 : */
5435 : Datum
5436 1 : icu_unicode_version(PG_FUNCTION_ARGS)
5437 : {
5438 1 : const char *version = pg_icu_unicode_version();
5439 :
5440 1 : if (version)
5441 1 : PG_RETURN_TEXT_P(cstring_to_text(version));
5442 : else
5443 0 : PG_RETURN_NULL();
5444 : }
5445 :
5446 : /*
5447 : * Check whether the string contains only assigned Unicode code
5448 : * points. Requires that the database encoding is UTF-8.
5449 : */
5450 : Datum
5451 10 : unicode_assigned(PG_FUNCTION_ARGS)
5452 : {
5453 10 : text *input = PG_GETARG_TEXT_PP(0);
5454 : unsigned char *p;
5455 : int size;
5456 :
5457 10 : if (GetDatabaseEncoding() != PG_UTF8)
5458 0 : ereport(ERROR,
5459 : (errmsg("Unicode categorization can only be performed if server encoding is UTF8")));
5460 :
5461 : /* convert to char32_t */
5462 10 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5463 10 : p = (unsigned char *) VARDATA_ANY(input);
5464 40 : for (int i = 0; i < size; i++)
5465 : {
5466 35 : char32_t uchar = utf8_to_unicode(p);
5467 35 : int category = unicode_category(uchar);
5468 :
5469 35 : if (category == PG_U_UNASSIGNED)
5470 5 : PG_RETURN_BOOL(false);
5471 :
5472 30 : p += pg_utf_mblen(p);
5473 : }
5474 :
5475 5 : PG_RETURN_BOOL(true);
5476 : }
5477 :
5478 : Datum
5479 55 : unicode_normalize_func(PG_FUNCTION_ARGS)
5480 : {
5481 55 : text *input = PG_GETARG_TEXT_PP(0);
5482 55 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5483 : UnicodeNormalizationForm form;
5484 : int size;
5485 : char32_t *input_chars;
5486 : char32_t *output_chars;
5487 : unsigned char *p;
5488 : text *result;
5489 : int i;
5490 :
5491 55 : form = unicode_norm_form_from_string(formstr);
5492 :
5493 : /* convert to char32_t */
5494 51 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5495 51 : input_chars = palloc((size + 1) * sizeof(char32_t));
5496 51 : p = (unsigned char *) VARDATA_ANY(input);
5497 220 : for (i = 0; i < size; i++)
5498 : {
5499 169 : input_chars[i] = utf8_to_unicode(p);
5500 169 : p += pg_utf_mblen(p);
5501 : }
5502 51 : input_chars[i] = (char32_t) '\0';
5503 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5504 :
5505 : /* action */
5506 51 : output_chars = unicode_normalize(form, input_chars);
5507 :
5508 : /* convert back to UTF-8 string */
5509 51 : size = 0;
5510 231 : for (char32_t *wp = output_chars; *wp; wp++)
5511 : {
5512 : unsigned char buf[4];
5513 :
5514 180 : unicode_to_utf8(*wp, buf);
5515 180 : size += pg_utf_mblen(buf);
5516 : }
5517 :
5518 51 : result = palloc(size + VARHDRSZ);
5519 51 : SET_VARSIZE(result, size + VARHDRSZ);
5520 :
5521 51 : p = (unsigned char *) VARDATA_ANY(result);
5522 231 : for (char32_t *wp = output_chars; *wp; wp++)
5523 : {
5524 180 : unicode_to_utf8(*wp, p);
5525 180 : p += pg_utf_mblen(p);
5526 : }
5527 : Assert((char *) p == (char *) result + size + VARHDRSZ);
5528 :
5529 51 : PG_RETURN_TEXT_P(result);
5530 : }
5531 :
5532 : /*
5533 : * Check whether the string is in the specified Unicode normalization form.
5534 : *
5535 : * This is done by converting the string to the specified normal form and then
5536 : * comparing that to the original string. To speed that up, we also apply the
5537 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
5538 : * answer for many strings by just scanning the string once.
5539 : *
5540 : * This function should generally be optimized for the case where the string
5541 : * is in fact normalized. In that case, we'll end up looking at the entire
5542 : * string, so it's probably not worth doing any incremental conversion etc.
5543 : */
5544 : Datum
5545 94 : unicode_is_normalized(PG_FUNCTION_ARGS)
5546 : {
5547 94 : text *input = PG_GETARG_TEXT_PP(0);
5548 94 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5549 : UnicodeNormalizationForm form;
5550 : int size;
5551 : char32_t *input_chars;
5552 : char32_t *output_chars;
5553 : unsigned char *p;
5554 : int i;
5555 : UnicodeNormalizationQC quickcheck;
5556 : int output_size;
5557 : bool result;
5558 :
5559 94 : form = unicode_norm_form_from_string(formstr);
5560 :
5561 : /* convert to char32_t */
5562 90 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5563 90 : input_chars = palloc((size + 1) * sizeof(char32_t));
5564 90 : p = (unsigned char *) VARDATA_ANY(input);
5565 344 : for (i = 0; i < size; i++)
5566 : {
5567 254 : input_chars[i] = utf8_to_unicode(p);
5568 254 : p += pg_utf_mblen(p);
5569 : }
5570 90 : input_chars[i] = (char32_t) '\0';
5571 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5572 :
5573 : /* quick check (see UAX #15) */
5574 90 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
5575 90 : if (quickcheck == UNICODE_NORM_QC_YES)
5576 30 : PG_RETURN_BOOL(true);
5577 60 : else if (quickcheck == UNICODE_NORM_QC_NO)
5578 8 : PG_RETURN_BOOL(false);
5579 :
5580 : /* normalize and compare with original */
5581 52 : output_chars = unicode_normalize(form, input_chars);
5582 :
5583 52 : output_size = 0;
5584 216 : for (char32_t *wp = output_chars; *wp; wp++)
5585 164 : output_size++;
5586 :
5587 76 : result = (size == output_size) &&
5588 24 : (memcmp(input_chars, output_chars, size * sizeof(char32_t)) == 0);
5589 :
5590 52 : PG_RETURN_BOOL(result);
5591 : }
5592 :
5593 : /*
5594 : * Check if first n chars are hexadecimal digits
5595 : */
5596 : static bool
5597 111 : isxdigits_n(const char *instr, size_t n)
5598 : {
5599 469 : for (size_t i = 0; i < n; i++)
5600 405 : if (!isxdigit((unsigned char) instr[i]))
5601 47 : return false;
5602 :
5603 64 : return true;
5604 : }
5605 :
5606 : static unsigned int
5607 358 : hexval(unsigned char c)
5608 : {
5609 358 : if (c >= '0' && c <= '9')
5610 278 : return c - '0';
5611 80 : if (c >= 'a' && c <= 'f')
5612 40 : return c - 'a' + 0xA;
5613 40 : if (c >= 'A' && c <= 'F')
5614 40 : return c - 'A' + 0xA;
5615 0 : elog(ERROR, "invalid hexadecimal digit");
5616 : return 0; /* not reached */
5617 : }
5618 :
5619 : /*
5620 : * Translate string with hexadecimal digits to number
5621 : */
5622 : static unsigned int
5623 64 : hexval_n(const char *instr, size_t n)
5624 : {
5625 64 : unsigned int result = 0;
5626 :
5627 422 : for (size_t i = 0; i < n; i++)
5628 358 : result += hexval(instr[i]) << (4 * (n - i - 1));
5629 :
5630 64 : return result;
5631 : }
5632 :
5633 : /*
5634 : * Replaces Unicode escape sequences by Unicode characters
5635 : */
5636 : Datum
5637 47 : unistr(PG_FUNCTION_ARGS)
5638 : {
5639 47 : text *input_text = PG_GETARG_TEXT_PP(0);
5640 : char *instr;
5641 : int len;
5642 : StringInfoData str;
5643 : text *result;
5644 47 : char16_t pair_first = 0;
5645 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
5646 :
5647 47 : instr = VARDATA_ANY(input_text);
5648 47 : len = VARSIZE_ANY_EXHDR(input_text);
5649 :
5650 47 : initStringInfo(&str);
5651 :
5652 356 : while (len > 0)
5653 : {
5654 337 : if (instr[0] == '\\')
5655 : {
5656 73 : if (len >= 2 &&
5657 73 : instr[1] == '\\')
5658 : {
5659 5 : if (pair_first)
5660 0 : goto invalid_pair;
5661 5 : appendStringInfoChar(&str, '\\');
5662 5 : instr += 2;
5663 5 : len -= 2;
5664 : }
5665 68 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
5666 47 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
5667 22 : {
5668 : char32_t unicode;
5669 30 : int offset = instr[1] == 'u' ? 2 : 1;
5670 :
5671 30 : unicode = hexval_n(instr + offset, 4);
5672 :
5673 30 : if (!is_valid_unicode_codepoint(unicode))
5674 0 : ereport(ERROR,
5675 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5676 : errmsg("invalid Unicode code point: %04X", unicode));
5677 :
5678 30 : if (pair_first)
5679 : {
5680 8 : if (is_utf16_surrogate_second(unicode))
5681 : {
5682 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5683 0 : pair_first = 0;
5684 : }
5685 : else
5686 8 : goto invalid_pair;
5687 : }
5688 22 : else if (is_utf16_surrogate_second(unicode))
5689 0 : goto invalid_pair;
5690 :
5691 22 : if (is_utf16_surrogate_first(unicode))
5692 12 : pair_first = unicode;
5693 : else
5694 : {
5695 10 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5696 10 : appendStringInfoString(&str, cbuf);
5697 : }
5698 :
5699 22 : instr += 4 + offset;
5700 22 : len -= 4 + offset;
5701 : }
5702 38 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
5703 9 : {
5704 : char32_t unicode;
5705 :
5706 17 : unicode = hexval_n(instr + 2, 6);
5707 :
5708 17 : if (!is_valid_unicode_codepoint(unicode))
5709 4 : ereport(ERROR,
5710 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5711 : errmsg("invalid Unicode code point: %04X", unicode));
5712 :
5713 13 : if (pair_first)
5714 : {
5715 4 : if (is_utf16_surrogate_second(unicode))
5716 : {
5717 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5718 0 : pair_first = 0;
5719 : }
5720 : else
5721 4 : goto invalid_pair;
5722 : }
5723 9 : else if (is_utf16_surrogate_second(unicode))
5724 0 : goto invalid_pair;
5725 :
5726 9 : if (is_utf16_surrogate_first(unicode))
5727 4 : pair_first = unicode;
5728 : else
5729 : {
5730 5 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5731 5 : appendStringInfoString(&str, cbuf);
5732 : }
5733 :
5734 9 : instr += 8;
5735 9 : len -= 8;
5736 : }
5737 21 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
5738 9 : {
5739 : char32_t unicode;
5740 :
5741 17 : unicode = hexval_n(instr + 2, 8);
5742 :
5743 17 : if (!is_valid_unicode_codepoint(unicode))
5744 4 : ereport(ERROR,
5745 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5746 : errmsg("invalid Unicode code point: %04X", unicode));
5747 :
5748 13 : if (pair_first)
5749 : {
5750 4 : if (is_utf16_surrogate_second(unicode))
5751 : {
5752 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5753 0 : pair_first = 0;
5754 : }
5755 : else
5756 4 : goto invalid_pair;
5757 : }
5758 9 : else if (is_utf16_surrogate_second(unicode))
5759 0 : goto invalid_pair;
5760 :
5761 9 : if (is_utf16_surrogate_first(unicode))
5762 4 : pair_first = unicode;
5763 : else
5764 : {
5765 5 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5766 5 : appendStringInfoString(&str, cbuf);
5767 : }
5768 :
5769 9 : instr += 10;
5770 9 : len -= 10;
5771 : }
5772 : else
5773 4 : ereport(ERROR,
5774 : (errcode(ERRCODE_SYNTAX_ERROR),
5775 : errmsg("invalid Unicode escape"),
5776 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
5777 : }
5778 : else
5779 : {
5780 264 : if (pair_first)
5781 0 : goto invalid_pair;
5782 :
5783 264 : appendStringInfoChar(&str, *instr++);
5784 264 : len--;
5785 : }
5786 : }
5787 :
5788 : /* unfinished surrogate pair? */
5789 19 : if (pair_first)
5790 4 : goto invalid_pair;
5791 :
5792 15 : result = cstring_to_text_with_len(str.data, str.len);
5793 15 : pfree(str.data);
5794 :
5795 15 : PG_RETURN_TEXT_P(result);
5796 :
5797 20 : invalid_pair:
5798 20 : ereport(ERROR,
5799 : (errcode(ERRCODE_SYNTAX_ERROR),
5800 : errmsg("invalid Unicode surrogate pair")));
5801 : PG_RETURN_NULL(); /* keep compiler quiet */
5802 : }
|
