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 15313201 : cstring_to_text(const char *s)
185 : {
186 15313201 : 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 16877840 : cstring_to_text_with_len(const char *s, int len)
197 : {
198 16877840 : text *result = (text *) palloc(len + VARHDRSZ);
199 :
200 16877840 : SET_VARSIZE(result, len + VARHDRSZ);
201 16877840 : memcpy(VARDATA(result), s, len);
202 :
203 16877840 : 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 10408781 : text_to_cstring(const text *t)
218 : {
219 : /* must cast away the const, unfortunately */
220 10408781 : text *tunpacked = pg_detoast_datum_packed(unconstify(text *, t));
221 10408781 : int len = VARSIZE_ANY_EXHDR(tunpacked);
222 : char *result;
223 :
224 10408781 : result = (char *) palloc(len + 1);
225 10408781 : memcpy(result, VARDATA_ANY(tunpacked), len);
226 10408781 : result[len] = '\0';
227 :
228 10408781 : if (tunpacked != t)
229 29728 : pfree(tunpacked);
230 :
231 10408781 : 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 13009391 : textin(PG_FUNCTION_ARGS)
279 : {
280 13009391 : char *inputText = PG_GETARG_CSTRING(0);
281 :
282 13009391 : PG_RETURN_TEXT_P(cstring_to_text(inputText));
283 : }
284 :
285 : /*
286 : * textout - converts internal representation to cstring
287 : */
288 : Datum
289 4814180 : textout(PG_FUNCTION_ARGS)
290 : {
291 4814180 : Datum txt = PG_GETARG_DATUM(0);
292 :
293 4814180 : 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 2394 : textsend(PG_FUNCTION_ARGS)
319 : {
320 2394 : text *t = PG_GETARG_TEXT_PP(0);
321 : StringInfoData buf;
322 :
323 2394 : pq_begintypsend(&buf);
324 2394 : pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
325 2394 : 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
345 655 : unknownout(PG_FUNCTION_ARGS)
346 : {
347 : /* representation is same as cstring */
348 655 : char *str = PG_GETARG_CSTRING(0);
349 :
350 655 : 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 286157 : textlen(PG_FUNCTION_ARGS)
393 : {
394 286157 : Datum str = PG_GETARG_DATUM(0);
395 :
396 : /* try to avoid decompressing argument */
397 286157 : 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 286167 : text_length(Datum str)
411 : {
412 : /* fastpath when max encoding length is one */
413 286167 : if (pg_database_encoding_max_length() == 1)
414 10 : return (toast_raw_datum_size(str) - VARHDRSZ);
415 : else
416 : {
417 286157 : text *t = DatumGetTextPP(str);
418 :
419 286157 : 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 1361982 : textcat(PG_FUNCTION_ARGS)
449 : {
450 1361982 : text *t1 = PG_GETARG_TEXT_PP(0);
451 1361982 : text *t2 = PG_GETARG_TEXT_PP(1);
452 :
453 1361982 : 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 1362038 : text_catenate(text *t1, text *t2)
464 : {
465 : text *result;
466 : int len1,
467 : len2,
468 : len;
469 : char *ptr;
470 :
471 1362038 : len1 = VARSIZE_ANY_EXHDR(t1);
472 1362038 : len2 = VARSIZE_ANY_EXHDR(t2);
473 :
474 : /* paranoia ... probably should throw error instead? */
475 1362038 : if (len1 < 0)
476 0 : len1 = 0;
477 1362038 : if (len2 < 0)
478 0 : len2 = 0;
479 :
480 1362038 : len = len1 + len2 + VARHDRSZ;
481 1362038 : result = (text *) palloc(len);
482 :
483 : /* Set size of result string... */
484 1362038 : SET_VARSIZE(result, len);
485 :
486 : /* Fill data field of result string... */
487 1362038 : ptr = VARDATA(result);
488 1362038 : if (len1 > 0)
489 1360315 : memcpy(ptr, VARDATA_ANY(t1), len1);
490 1362038 : if (len2 > 0)
491 1361899 : memcpy(ptr + len1, VARDATA_ANY(t2), len2);
492 :
493 1362038 : 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 11449 : charlen_to_bytelen(const char *p, int n)
508 : {
509 11449 : 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 3086678 : for (s = p; n > 0; n--)
519 3075319 : s += pg_mblen_unbounded(s); /* caller verified encoding */
520 :
521 11359 : 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 369589 : text_substr(PG_FUNCTION_ARGS)
554 : {
555 369589 : 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 396357 : text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
587 : {
588 396357 : int32 eml = pg_database_encoding_max_length();
589 396357 : 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 396357 : S1 = Max(S, 1);
600 :
601 : /* life is easy if the encoding max length is 1 */
602 396357 : 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 396346 : 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 396346 : slice_start = 0;
667 :
668 396346 : if (length_not_specified) /* special case - get length to end of
669 : * string */
670 52 : E = slice_size = L1 = -1;
671 396294 : 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 396286 : 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 396281 : 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 396273 : 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 396273 : 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 792572 : if (VARATT_IS_COMPRESSED(DatumGetPointer(str)) ||
717 396242 : VARATT_IS_EXTERNAL(DatumGetPointer(str)))
718 240 : slice = DatumGetTextPSlice(str, slice_start, slice_size);
719 : else
720 396090 : slice = (text *) DatumGetPointer(str);
721 :
722 : /* see if we got back an empty string */
723 396330 : slice_len = VARSIZE_ANY_EXHDR(slice);
724 396330 : 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 396322 : slice_strlen =
739 396330 : (slice_size == -1 ?
740 396330 : pg_mbstrlen_with_len(VARDATA_ANY(slice), slice_len) :
741 396268 : 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 396322 : 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 396297 : if (L1 > -1)
759 396253 : 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 396297 : p = VARDATA_ANY(slice);
767 4234116 : for (i = 0; i < S1 - 1; i++)
768 3837819 : p += pg_mblen_unbounded(p);
769 :
770 : /* hang onto a pointer to our start position */
771 396297 : s = p;
772 :
773 : /*
774 : * Count the actual bytes used by the substring of the requested
775 : * length.
776 : */
777 6234989 : for (i = S1; i < E1; i++)
778 5838692 : p += pg_mblen_unbounded(p);
779 :
780 396297 : ret = (text *) palloc(VARHDRSZ + (p - s));
781 396297 : SET_VARSIZE(ret, VARHDRSZ + (p - s));
782 396297 : memcpy(VARDATA(ret), s, (p - s));
783 :
784 396297 : if (slice != (text *) DatumGetPointer(str))
785 232 : pfree(slice);
786 :
787 396297 : 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 396268 : pg_mbcharcliplen_chars(const char *mbstr, int len, int limit)
805 : {
806 396268 : 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 8108672 : while (len > 0 && *mbstr)
814 : {
815 8108336 : l = pg_mblen_with_len(mbstr, len);
816 8108328 : nch++;
817 8108328 : if (nch == limit)
818 395924 : break;
819 7712404 : len -= l;
820 7712404 : mbstr += l;
821 : }
822 396260 : 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 1176 : text_position_setup(text *t1, text *t2, Oid collid, TextPositionState *state)
963 : {
964 1176 : int len1 = VARSIZE_ANY_EXHDR(t1);
965 1176 : int len2 = VARSIZE_ANY_EXHDR(t2);
966 :
967 1176 : check_collation_set(collid);
968 :
969 1176 : 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 1176 : 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 1176 : if (pg_database_encoding_max_length() == 1)
989 54 : state->is_multibyte_char_in_char = false;
990 1122 : else if (GetDatabaseEncoding() == PG_UTF8)
991 1122 : state->is_multibyte_char_in_char = false;
992 : else
993 0 : state->is_multibyte_char_in_char = true;
994 :
995 1176 : state->str1 = VARDATA_ANY(t1);
996 1176 : state->str2 = VARDATA_ANY(t2);
997 1176 : state->len1 = len1;
998 1176 : state->len2 = len2;
999 1176 : state->last_match = NULL;
1000 1176 : state->refpoint = state->str1;
1001 1176 : 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 1176 : if (len1 >= len2 && len2 > 1 && state->locale->deterministic)
1017 : {
1018 922 : int searchlength = len1 - len2;
1019 : int skiptablemask;
1020 : int last;
1021 : int i;
1022 922 : 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 922 : if (searchlength < 16)
1037 112 : skiptablemask = 3;
1038 810 : else if (searchlength < 64)
1039 24 : skiptablemask = 7;
1040 786 : else if (searchlength < 128)
1041 15 : skiptablemask = 15;
1042 771 : else if (searchlength < 512)
1043 186 : skiptablemask = 31;
1044 585 : else if (searchlength < 2048)
1045 433 : skiptablemask = 63;
1046 152 : else if (searchlength < 4096)
1047 110 : skiptablemask = 127;
1048 : else
1049 42 : skiptablemask = 255;
1050 922 : 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 60298 : for (i = 0; i <= skiptablemask; i++)
1058 59376 : 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 922 : last = len2 - 1;
1068 :
1069 11131 : for (i = 0; i < last; i++)
1070 10209 : state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1071 : }
1072 1176 : }
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 5530 : text_position_next(TextPositionState *state)
1084 : {
1085 5530 : int needle_len = state->len2;
1086 : char *start_ptr;
1087 : char *matchptr;
1088 :
1089 5530 : if (needle_len <= 0)
1090 0 : return false; /* result for empty pattern */
1091 :
1092 : /* Start from the point right after the previous match. */
1093 5530 : if (state->last_match)
1094 4340 : start_ptr = state->last_match + state->last_match_len;
1095 : else
1096 1190 : start_ptr = state->str1;
1097 :
1098 5530 : retry:
1099 5530 : matchptr = text_position_next_internal(start_ptr, state);
1100 :
1101 5530 : if (!matchptr)
1102 1112 : 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 4418 : 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 4418 : state->last_match = matchptr;
1140 4418 : state->last_match_len = state->last_match_len_tmp;
1141 4418 : 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 5530 : text_position_next_internal(char *start_ptr, TextPositionState *state)
1151 : {
1152 5530 : int haystack_len = state->len1;
1153 5530 : int needle_len = state->len2;
1154 5530 : int skiptablemask = state->skiptablemask;
1155 5530 : const char *haystack = state->str1;
1156 5530 : const char *needle = state->str2;
1157 5530 : 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 5530 : state->last_match_len_tmp = needle_len;
1164 :
1165 5530 : 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 5354 : else if (needle_len == 1)
1229 : {
1230 : /* No point in using B-M-H for a one-character needle */
1231 597 : char nchar = *needle;
1232 :
1233 597 : hptr = start_ptr;
1234 4246 : while (hptr < haystack_end)
1235 : {
1236 4108 : if (*hptr == nchar)
1237 459 : return (char *) hptr;
1238 3649 : hptr++;
1239 : }
1240 : }
1241 : else
1242 : {
1243 4757 : const char *needle_last = &needle[needle_len - 1];
1244 :
1245 : /* Start at startpos plus the length of the needle */
1246 4757 : hptr = start_ptr + needle_len - 1;
1247 115268 : while (hptr < haystack_end)
1248 : {
1249 : /* Match the needle scanning *backward* */
1250 : const char *nptr;
1251 : const char *p;
1252 :
1253 114373 : nptr = needle_last;
1254 114373 : p = hptr;
1255 171302 : while (*nptr == *p)
1256 : {
1257 : /* Matched it all? If so, return 1-based position */
1258 60791 : if (nptr == needle)
1259 3862 : return (char *) p;
1260 56929 : 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 110511 : hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1272 : }
1273 : }
1274 :
1275 1033 : 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 4331 : text_position_get_match_ptr(TextPositionState *state)
1285 : {
1286 4331 : 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 14 : text_position_reset(TextPositionState *state)
1312 : {
1313 14 : state->last_match = NULL;
1314 14 : state->refpoint = state->str1;
1315 14 : state->refpos = 0;
1316 14 : }
1317 :
1318 : static void
1319 1176 : text_position_cleanup(TextPositionState *state)
1320 : {
1321 : /* no cleanup needed */
1322 1176 : }
1323 :
1324 :
1325 : static void
1326 11723493 : check_collation_set(Oid collid)
1327 : {
1328 11723493 : 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 11723473 : }
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 5955297 : 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 5955297 : check_collation_set(collid);
1361 :
1362 5955285 : mylocale = pg_newlocale_from_collation(collid);
1363 :
1364 5955285 : if (mylocale->collate_is_c)
1365 : {
1366 2147769 : result = memcmp(arg1, arg2, Min(len1, len2));
1367 2147769 : if ((result == 0) && (len1 != len2))
1368 86434 : 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 3807516 : if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1382 1059135 : return 0;
1383 :
1384 2748381 : result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1385 :
1386 : /* Break tie if necessary. */
1387 2748381 : 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 4896150 : 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 4740979 : text_cmp(text *arg1, text *arg2, Oid collid)
1404 : {
1405 : char *a1p,
1406 : *a2p;
1407 : int len1,
1408 : len2;
1409 :
1410 4740979 : a1p = VARDATA_ANY(arg1);
1411 4740979 : a2p = VARDATA_ANY(arg2);
1412 :
1413 4740979 : len1 = VARSIZE_ANY_EXHDR(arg1);
1414 4740979 : len2 = VARSIZE_ANY_EXHDR(arg2);
1415 :
1416 4740979 : 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 5303056 : texteq(PG_FUNCTION_ARGS)
1429 : {
1430 5303056 : Oid collid = PG_GET_COLLATION();
1431 5303056 : pg_locale_t mylocale = 0;
1432 : bool result;
1433 :
1434 5303056 : check_collation_set(collid);
1435 :
1436 5303056 : mylocale = pg_newlocale_from_collation(collid);
1437 :
1438 5303056 : if (mylocale->deterministic)
1439 : {
1440 5297138 : Datum arg1 = PG_GETARG_DATUM(0);
1441 5297138 : 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 5297138 : len1 = toast_raw_datum_size(arg1);
1453 5297138 : len2 = toast_raw_datum_size(arg2);
1454 5297138 : if (len1 != len2)
1455 2682214 : result = false;
1456 : else
1457 : {
1458 2614924 : text *targ1 = DatumGetTextPP(arg1);
1459 2614924 : text *targ2 = DatumGetTextPP(arg2);
1460 :
1461 2614924 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1462 : len1 - VARHDRSZ) == 0);
1463 :
1464 2614924 : PG_FREE_IF_COPY(targ1, 0);
1465 2614924 : PG_FREE_IF_COPY(targ2, 1);
1466 : }
1467 : }
1468 : else
1469 : {
1470 5918 : text *arg1 = PG_GETARG_TEXT_PP(0);
1471 5918 : text *arg2 = PG_GETARG_TEXT_PP(1);
1472 :
1473 5918 : result = (text_cmp(arg1, arg2, collid) == 0);
1474 :
1475 5918 : PG_FREE_IF_COPY(arg1, 0);
1476 5918 : PG_FREE_IF_COPY(arg2, 1);
1477 : }
1478 :
1479 5303056 : PG_RETURN_BOOL(result);
1480 : }
1481 :
1482 : Datum
1483 207832 : textne(PG_FUNCTION_ARGS)
1484 : {
1485 207832 : Oid collid = PG_GET_COLLATION();
1486 : pg_locale_t mylocale;
1487 : bool result;
1488 :
1489 207832 : check_collation_set(collid);
1490 :
1491 207832 : mylocale = pg_newlocale_from_collation(collid);
1492 :
1493 207832 : if (mylocale->deterministic)
1494 : {
1495 207816 : Datum arg1 = PG_GETARG_DATUM(0);
1496 207816 : Datum arg2 = PG_GETARG_DATUM(1);
1497 : Size len1,
1498 : len2;
1499 :
1500 : /* See comment in texteq() */
1501 207816 : len1 = toast_raw_datum_size(arg1);
1502 207816 : len2 = toast_raw_datum_size(arg2);
1503 207816 : if (len1 != len2)
1504 11729 : result = true;
1505 : else
1506 : {
1507 196087 : text *targ1 = DatumGetTextPP(arg1);
1508 196087 : text *targ2 = DatumGetTextPP(arg2);
1509 :
1510 196087 : result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1511 : len1 - VARHDRSZ) != 0);
1512 :
1513 196087 : PG_FREE_IF_COPY(targ1, 0);
1514 196087 : 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 207832 : PG_RETURN_BOOL(result);
1529 : }
1530 :
1531 : Datum
1532 231689 : text_lt(PG_FUNCTION_ARGS)
1533 : {
1534 231689 : text *arg1 = PG_GETARG_TEXT_PP(0);
1535 231689 : text *arg2 = PG_GETARG_TEXT_PP(1);
1536 : bool result;
1537 :
1538 231689 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1539 :
1540 231677 : PG_FREE_IF_COPY(arg1, 0);
1541 231677 : PG_FREE_IF_COPY(arg2, 1);
1542 :
1543 231677 : PG_RETURN_BOOL(result);
1544 : }
1545 :
1546 : Datum
1547 213626 : text_le(PG_FUNCTION_ARGS)
1548 : {
1549 213626 : text *arg1 = PG_GETARG_TEXT_PP(0);
1550 213626 : text *arg2 = PG_GETARG_TEXT_PP(1);
1551 : bool result;
1552 :
1553 213626 : result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1554 :
1555 213626 : PG_FREE_IF_COPY(arg1, 0);
1556 213626 : PG_FREE_IF_COPY(arg2, 1);
1557 :
1558 213626 : 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 3760717 : bttextcmp(PG_FUNCTION_ARGS)
1632 : {
1633 3760717 : text *arg1 = PG_GETARG_TEXT_PP(0);
1634 3760717 : text *arg2 = PG_GETARG_TEXT_PP(1);
1635 : int32 result;
1636 :
1637 3760717 : result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1638 :
1639 3760717 : PG_FREE_IF_COPY(arg1, 0);
1640 3760717 : PG_FREE_IF_COPY(arg2, 1);
1641 :
1642 3760717 : PG_RETURN_INT32(result);
1643 : }
1644 :
1645 : Datum
1646 50934 : bttextsortsupport(PG_FUNCTION_ARGS)
1647 : {
1648 50934 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
1649 50934 : Oid collid = ssup->ssup_collation;
1650 : MemoryContext oldcontext;
1651 :
1652 50934 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1653 :
1654 : /* Use generic string SortSupport */
1655 50934 : varstr_sortsupport(ssup, TEXTOID, collid);
1656 :
1657 50926 : MemoryContextSwitchTo(oldcontext);
1658 :
1659 50926 : 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 87124 : varstr_sortsupport(SortSupport ssup, Oid typid, Oid collid)
1672 : {
1673 87124 : bool abbreviate = ssup->abbreviate;
1674 87124 : bool collate_c = false;
1675 : VarStringSortSupport *sss;
1676 : pg_locale_t locale;
1677 :
1678 87124 : check_collation_set(collid);
1679 :
1680 87116 : 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 87116 : if (locale->collate_is_c)
1696 : {
1697 59708 : if (typid == BPCHAROID)
1698 157 : ssup->comparator = bpcharfastcmp_c;
1699 59551 : else if (typid == NAMEOID)
1700 : {
1701 35512 : ssup->comparator = namefastcmp_c;
1702 : /* Not supporting abbreviation with type NAME, for now */
1703 35512 : abbreviate = false;
1704 : }
1705 : else
1706 24039 : ssup->comparator = varstrfastcmp_c;
1707 :
1708 59708 : collate_c = true;
1709 : }
1710 : else
1711 : {
1712 : /*
1713 : * We use varlenafastcmp_locale except for type NAME.
1714 : */
1715 27408 : 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 27408 : 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 27408 : if (!pg_strxfrm_enabled(locale))
1739 26870 : 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 87116 : if (abbreviate || !collate_c)
1750 : {
1751 40373 : sss = palloc_object(VarStringSortSupport);
1752 40373 : sss->buf1 = palloc(TEXTBUFLEN);
1753 40373 : sss->buflen1 = TEXTBUFLEN;
1754 40373 : sss->buf2 = palloc(TEXTBUFLEN);
1755 40373 : sss->buflen2 = TEXTBUFLEN;
1756 : /* Start with invalid values */
1757 40373 : sss->last_len1 = -1;
1758 40373 : sss->last_len2 = -1;
1759 : /* Initialize */
1760 40373 : sss->last_returned = 0;
1761 40373 : if (collate_c)
1762 12965 : sss->locale = NULL;
1763 : else
1764 27408 : 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 40373 : sss->cache_blob = true;
1781 40373 : sss->collate_c = collate_c;
1782 40373 : sss->typid = typid;
1783 40373 : 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 40373 : if (abbreviate)
1791 : {
1792 13363 : sss->prop_card = 0.20;
1793 13363 : initHyperLogLog(&sss->abbr_card, 10);
1794 13363 : initHyperLogLog(&sss->full_card, 10);
1795 13363 : ssup->abbrev_full_comparator = ssup->comparator;
1796 13363 : ssup->comparator = ssup_datum_unsigned_cmp;
1797 13363 : ssup->abbrev_converter = varstr_abbrev_convert;
1798 13363 : ssup->abbrev_abort = varstr_abbrev_abort;
1799 : }
1800 : }
1801 87116 : }
1802 :
1803 : /*
1804 : * sortsupport comparison func (for C locale case)
1805 : */
1806 : static int
1807 25250835 : varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
1808 : {
1809 25250835 : VarString *arg1 = DatumGetVarStringPP(x);
1810 25250835 : VarString *arg2 = DatumGetVarStringPP(y);
1811 : char *a1p,
1812 : *a2p;
1813 : int len1,
1814 : len2,
1815 : result;
1816 :
1817 25250835 : a1p = VARDATA_ANY(arg1);
1818 25250835 : a2p = VARDATA_ANY(arg2);
1819 :
1820 25250835 : len1 = VARSIZE_ANY_EXHDR(arg1);
1821 25250835 : len2 = VARSIZE_ANY_EXHDR(arg2);
1822 :
1823 25250835 : result = memcmp(a1p, a2p, Min(len1, len2));
1824 25250835 : if ((result == 0) && (len1 != len2))
1825 767225 : result = (len1 < len2) ? -1 : 1;
1826 :
1827 : /* We can't afford to leak memory here. */
1828 25250835 : if (PointerGetDatum(arg1) != x)
1829 0 : pfree(arg1);
1830 25250835 : if (PointerGetDatum(arg2) != y)
1831 0 : pfree(arg2);
1832 :
1833 25250835 : 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 31151 : bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
1845 : {
1846 31151 : BpChar *arg1 = DatumGetBpCharPP(x);
1847 31151 : BpChar *arg2 = DatumGetBpCharPP(y);
1848 : char *a1p,
1849 : *a2p;
1850 : int len1,
1851 : len2,
1852 : result;
1853 :
1854 31151 : a1p = VARDATA_ANY(arg1);
1855 31151 : a2p = VARDATA_ANY(arg2);
1856 :
1857 31151 : len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
1858 31151 : len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
1859 :
1860 31151 : result = memcmp(a1p, a2p, Min(len1, len2));
1861 31151 : if ((result == 0) && (len1 != len2))
1862 2 : result = (len1 < len2) ? -1 : 1;
1863 :
1864 : /* We can't afford to leak memory here. */
1865 31151 : if (PointerGetDatum(arg1) != x)
1866 0 : pfree(arg1);
1867 31151 : if (PointerGetDatum(arg2) != y)
1868 0 : pfree(arg2);
1869 :
1870 31151 : return result;
1871 : }
1872 :
1873 : /*
1874 : * sortsupport comparison func (for NAME C locale case)
1875 : */
1876 : static int
1877 24408188 : namefastcmp_c(Datum x, Datum y, SortSupport ssup)
1878 : {
1879 24408188 : Name arg1 = DatumGetName(x);
1880 24408188 : Name arg2 = DatumGetName(y);
1881 :
1882 24408188 : 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 21802811 : varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
1890 : {
1891 21802811 : VarString *arg1 = DatumGetVarStringPP(x);
1892 21802811 : VarString *arg2 = DatumGetVarStringPP(y);
1893 : char *a1p,
1894 : *a2p;
1895 : int len1,
1896 : len2,
1897 : result;
1898 :
1899 21802811 : a1p = VARDATA_ANY(arg1);
1900 21802811 : a2p = VARDATA_ANY(arg2);
1901 :
1902 21802811 : len1 = VARSIZE_ANY_EXHDR(arg1);
1903 21802811 : len2 = VARSIZE_ANY_EXHDR(arg2);
1904 :
1905 21802811 : result = varstrfastcmp_locale(a1p, len1, a2p, len2, ssup);
1906 :
1907 : /* We can't afford to leak memory here. */
1908 21802811 : if (PointerGetDatum(arg1) != x)
1909 3 : pfree(arg1);
1910 21802811 : if (PointerGetDatum(arg2) != y)
1911 3 : pfree(arg2);
1912 :
1913 21802811 : 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 21802811 : varstrfastcmp_locale(char *a1p, int len1, char *a2p, int len2, SortSupport ssup)
1935 : {
1936 21802811 : 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 21802811 : 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 6860309 : return 0;
1957 : }
1958 :
1959 14942502 : if (sss->typid == BPCHAROID)
1960 : {
1961 : /* Get true number of bytes, ignoring trailing spaces */
1962 18725 : len1 = bpchartruelen(a1p, len1);
1963 18725 : len2 = bpchartruelen(a2p, len2);
1964 : }
1965 :
1966 14942502 : 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 14942502 : 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 14942502 : arg1_match = true;
1988 14942502 : if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
1989 : {
1990 13583481 : arg1_match = false;
1991 13583481 : memcpy(sss->buf1, a1p, len1);
1992 13583481 : sss->buf1[len1] = '\0';
1993 13583481 : 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 14942502 : if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2003 : {
2004 2281778 : memcpy(sss->buf2, a2p, len2);
2005 2281778 : sss->buf2[len2] = '\0';
2006 2281778 : sss->last_len2 = len2;
2007 : }
2008 12660724 : else if (arg1_match && !sss->cache_blob)
2009 : {
2010 : /* Use result cached following last actual strcoll() call */
2011 1124065 : return sss->last_returned;
2012 : }
2013 :
2014 13818437 : result = pg_strcoll(sss->buf1, sss->buf2, sss->locale);
2015 :
2016 : /* Break tie if necessary. */
2017 13818437 : 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 13818437 : sss->cache_blob = false;
2022 13818437 : sss->last_returned = result;
2023 13818437 : 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 499844 : varstr_abbrev_convert(Datum original, SortSupport ssup)
2035 : {
2036 499844 : const size_t max_prefix_bytes = sizeof(Datum);
2037 499844 : VarStringSortSupport *sss = (VarStringSortSupport *) ssup->ssup_extra;
2038 499844 : VarString *authoritative = DatumGetVarStringPP(original);
2039 499844 : char *authoritative_data = VARDATA_ANY(authoritative);
2040 :
2041 : /* working state */
2042 : Datum res;
2043 : char *pres;
2044 : int len;
2045 : uint32 hash;
2046 :
2047 499844 : pres = (char *) &res;
2048 : /* memset(), so any non-overwritten bytes are NUL */
2049 499844 : memset(pres, 0, max_prefix_bytes);
2050 499844 : len = VARSIZE_ANY_EXHDR(authoritative);
2051 :
2052 : /* Get number of bytes, ignoring trailing spaces */
2053 499844 : 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 499844 : if (sss->collate_c)
2062 498622 : 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 499732 : hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2157 : Min(len, PG_CACHE_LINE_SIZE)));
2158 :
2159 499732 : if (len > PG_CACHE_LINE_SIZE)
2160 97 : hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2161 :
2162 499732 : addHyperLogLog(&sss->full_card, hash);
2163 :
2164 : /* Hash abbreviated key */
2165 : {
2166 : uint32 tmp;
2167 :
2168 499732 : tmp = DatumGetUInt32(res) ^ (uint32) (DatumGetUInt64(res) >> 32);
2169 499732 : hash = DatumGetUInt32(hash_uint32(tmp));
2170 : }
2171 :
2172 499732 : addHyperLogLog(&sss->abbr_card, hash);
2173 :
2174 : /* Cache result, perhaps saving an expensive strxfrm() call next time */
2175 499732 : sss->cache_blob = true;
2176 499844 : 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 499844 : res = DatumBigEndianToNative(res);
2187 :
2188 : /* Don't leak memory here */
2189 499844 : if (PointerGetDatum(authoritative) != original)
2190 1 : pfree(authoritative);
2191 :
2192 499844 : 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 1444 : varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2202 : {
2203 1444 : 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 1444 : if (memtupcount < 100)
2211 836 : 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 5608 : btvarstrequalimage(PG_FUNCTION_ARGS)
2316 : {
2317 : #ifdef NOT_USED
2318 : Oid opcintype = PG_GETARG_OID(0);
2319 : #endif
2320 5608 : Oid collid = PG_GET_COLLATION();
2321 : pg_locale_t locale;
2322 :
2323 5608 : check_collation_set(collid);
2324 :
2325 5608 : locale = pg_newlocale_from_collation(collid);
2326 :
2327 5608 : 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 132339 : nameeqtext(PG_FUNCTION_ARGS)
2361 : {
2362 132339 : Name arg1 = PG_GETARG_NAME(0);
2363 132339 : text *arg2 = PG_GETARG_TEXT_PP(1);
2364 132339 : size_t len1 = strlen(NameStr(*arg1));
2365 132339 : size_t len2 = VARSIZE_ANY_EXHDR(arg2);
2366 132339 : Oid collid = PG_GET_COLLATION();
2367 : bool result;
2368 :
2369 132339 : check_collation_set(collid);
2370 :
2371 132339 : if (collid == C_COLLATION_OID)
2372 172134 : result = (len1 == len2 &&
2373 83278 : memcmp(NameStr(*arg1), VARDATA_ANY(arg2), len1) == 0);
2374 : else
2375 43483 : result = (varstr_cmp(NameStr(*arg1), len1,
2376 43483 : VARDATA_ANY(arg2), len2,
2377 : collid) == 0);
2378 :
2379 132339 : PG_FREE_IF_COPY(arg2, 1);
2380 :
2381 132339 : PG_RETURN_BOOL(result);
2382 : }
2383 :
2384 : Datum
2385 5674 : texteqname(PG_FUNCTION_ARGS)
2386 : {
2387 5674 : text *arg1 = PG_GETARG_TEXT_PP(0);
2388 5674 : Name arg2 = PG_GETARG_NAME(1);
2389 5674 : size_t len1 = VARSIZE_ANY_EXHDR(arg1);
2390 5674 : size_t len2 = strlen(NameStr(*arg2));
2391 5674 : Oid collid = PG_GET_COLLATION();
2392 : bool result;
2393 :
2394 5674 : check_collation_set(collid);
2395 :
2396 5674 : if (collid == C_COLLATION_OID)
2397 378 : result = (len1 == len2 &&
2398 121 : memcmp(VARDATA_ANY(arg1), NameStr(*arg2), len1) == 0);
2399 : else
2400 5417 : result = (varstr_cmp(VARDATA_ANY(arg1), len1,
2401 5417 : NameStr(*arg2), len2,
2402 : collid) == 0);
2403 :
2404 5674 : PG_FREE_IF_COPY(arg1, 0);
2405 :
2406 5674 : 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 84752 : btnametextcmp(PG_FUNCTION_ARGS)
2461 : {
2462 84752 : Name arg1 = PG_GETARG_NAME(0);
2463 84752 : text *arg2 = PG_GETARG_TEXT_PP(1);
2464 : int32 result;
2465 :
2466 84752 : result = varstr_cmp(NameStr(*arg1), strlen(NameStr(*arg1)),
2467 84752 : VARDATA_ANY(arg2), VARSIZE_ANY_EXHDR(arg2),
2468 : PG_GET_COLLATION());
2469 :
2470 84752 : PG_FREE_IF_COPY(arg2, 1);
2471 :
2472 84752 : 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 40603 : namelttext(PG_FUNCTION_ARGS)
2499 : {
2500 40603 : 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 35805 : namegetext(PG_FUNCTION_ARGS)
2517 : {
2518 35805 : 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 20763 : text_name(PG_FUNCTION_ARGS)
2680 : {
2681 20763 : text *s = PG_GETARG_TEXT_PP(0);
2682 : Name result;
2683 : int len;
2684 :
2685 20763 : len = VARSIZE_ANY_EXHDR(s);
2686 :
2687 : /* Truncate oversize input */
2688 20763 : 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 20763 : result = (Name) palloc0(NAMEDATALEN);
2693 20763 : memcpy(NameStr(*result), VARDATA_ANY(s), len);
2694 :
2695 20763 : PG_RETURN_NAME(result);
2696 : }
2697 :
2698 : /* name_text()
2699 : * Converts a Name type to a text type.
2700 : */
2701 : Datum
2702 426050 : name_text(PG_FUNCTION_ARGS)
2703 : {
2704 426050 : Name s = PG_GETARG_NAME(0);
2705 :
2706 426050 : 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 3005 : textToQualifiedNameList(text *textval)
2720 : {
2721 : char *rawname;
2722 3005 : 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 3005 : rawname = text_to_cstring(textval);
2729 :
2730 3005 : if (!SplitIdentifierString(rawname, '.', &namelist))
2731 0 : ereport(ERROR,
2732 : (errcode(ERRCODE_INVALID_NAME),
2733 : errmsg("invalid name syntax")));
2734 :
2735 3005 : if (namelist == NIL)
2736 0 : ereport(ERROR,
2737 : (errcode(ERRCODE_INVALID_NAME),
2738 : errmsg("invalid name syntax")));
2739 :
2740 6103 : foreach(l, namelist)
2741 : {
2742 3098 : char *curname = (char *) lfirst(l);
2743 :
2744 3098 : result = lappend(result, makeString(pstrdup(curname)));
2745 : }
2746 :
2747 3005 : pfree(rawname);
2748 3005 : list_free(namelist);
2749 :
2750 3005 : return result;
2751 : }
2752 :
2753 : /*
2754 : * scan_quoted_identifier - In-place scanner for quoted identifiers.
2755 : *
2756 : * *nextp should point to the opening double-quote character, and will be
2757 : * updated to point just past the end. *endp is set to the position of
2758 : * the closing quote. The return value is the identifier, or NULL if the
2759 : * matching close-quote cannot be found.
2760 : *
2761 : * If we find two consecutive double quote characters, that doesn't end the
2762 : * identifier: instead, we collapse them into a double quote and include them
2763 : * in the resulting token. Note that this requires overwriting the rest of the
2764 : * string in place, including the portion beyond the final value of *nextp.
2765 : */
2766 : char *
2767 25119 : scan_quoted_identifier(char **endp, char **nextp)
2768 : {
2769 25119 : char *token = *nextp + 1;
2770 :
2771 : for (;;)
2772 : {
2773 25129 : *endp = strchr(*nextp + 1, '"');
2774 25124 : if (*endp == NULL)
2775 2 : return NULL; /* mismatched quotes */
2776 25122 : if ((*endp)[1] != '"')
2777 25117 : break; /* found end of quoted identifier */
2778 : /* Collapse adjacent quotes into one quote, and look again */
2779 5 : memmove(*endp, *endp + 1, strlen(*endp));
2780 5 : *nextp = *endp;
2781 : }
2782 : /* *endp now points at the terminating quote */
2783 25117 : *nextp = *endp + 1;
2784 :
2785 25117 : return token;
2786 : }
2787 :
2788 : /*
2789 : * scan_identifier - In-place scanner for quoted or unquoted identifiers.
2790 : *
2791 : * On success, *endp is set to the position where the caller should write '\0'
2792 : * to null-terminate the token, and *nextp is advanced past the token (and past
2793 : * the closing quote, if any). The return value is the token content, or NULL
2794 : * if there is a syntax error (mismatched quotes or empty unquoted token).
2795 : *
2796 : * Unquoted identifiers are terminated by whitespace or the first occurrence
2797 : * of the separator character. Additionally, if downcase_unquoted = true,
2798 : * unquoted identifiers are downcased in place. See scan_quoted_identifier for
2799 : * an additional way in which we modify the string in place.
2800 : */
2801 : char *
2802 351098 : scan_identifier(char **endp, char **nextp, char separator, bool downcase_unquoted)
2803 : {
2804 : char *token;
2805 :
2806 351098 : if (**nextp == '"')
2807 25119 : return scan_quoted_identifier(endp, nextp);
2808 :
2809 : /* Unquoted identifier --- extends to separator or whitespace */
2810 325979 : token = *nextp;
2811 :
2812 2981816 : while (**nextp && **nextp != separator && !scanner_isspace(**nextp))
2813 2655837 : (*nextp)++;
2814 :
2815 325979 : if (*nextp == token)
2816 2 : return NULL; /* empty token */
2817 :
2818 325977 : *endp = *nextp;
2819 :
2820 325977 : if (downcase_unquoted)
2821 : {
2822 : /*
2823 : * Downcase the identifier, using same code as main lexer does.
2824 : *
2825 : * XXX because we want to overwrite the input in-place, we cannot
2826 : * support a downcasing transformation that increases the string
2827 : * length. This is not a problem given the current implementation of
2828 : * downcase_truncate_identifier, but we'll probably have to do
2829 : * something about this someday.
2830 : */
2831 324107 : int len = *endp - token;
2832 324107 : char *downname = downcase_truncate_identifier(token, len, false);
2833 :
2834 : Assert(strlen(downname) <= len);
2835 324107 : strncpy(token, downname, len); /* strncpy is required here */
2836 324107 : pfree(downname);
2837 : }
2838 :
2839 325977 : return token;
2840 : }
2841 :
2842 :
2843 : /*
2844 : * SplitIdentifierString --- parse a string containing identifiers
2845 : *
2846 : * This is the guts of textToQualifiedNameList, and is exported for use in
2847 : * other situations such as parsing GUC variables. In the GUC case, it's
2848 : * important to avoid memory leaks, so the API is designed to minimize the
2849 : * amount of stuff that needs to be allocated and freed.
2850 : *
2851 : * Inputs:
2852 : * rawstring: the input string; must be overwritable! On return, it's
2853 : * been modified to contain the separated identifiers.
2854 : * separator: the separator punctuation expected between identifiers
2855 : * (typically '.' or ','). Whitespace may also appear around
2856 : * identifiers.
2857 : * Outputs:
2858 : * namelist: filled with a palloc'd list of pointers to identifiers within
2859 : * rawstring. Caller should list_free() this even on error return.
2860 : *
2861 : * Returns true if okay, false if there is a syntax error in the string.
2862 : *
2863 : * Note that an empty string is considered okay here, though not in
2864 : * textToQualifiedNameList.
2865 : */
2866 : bool
2867 205837 : SplitIdentifierString(char *rawstring, char separator,
2868 : List **namelist)
2869 : {
2870 205837 : char *nextp = rawstring;
2871 205837 : bool done = false;
2872 :
2873 205837 : *namelist = NIL;
2874 :
2875 205841 : while (scanner_isspace(*nextp))
2876 4 : nextp++; /* skip leading whitespace */
2877 :
2878 205837 : if (*nextp == '\0')
2879 17445 : return true; /* empty string represents empty list */
2880 :
2881 : /* At the top of the loop, we are at start of a new identifier. */
2882 : do
2883 : {
2884 : char *curname;
2885 : char *endp;
2886 :
2887 349162 : curname = scan_identifier(&endp, &nextp, separator, true);
2888 349162 : if (curname == NULL)
2889 1 : return false; /* mismatched quotes or empty name */
2890 :
2891 349163 : while (scanner_isspace(*nextp))
2892 1 : nextp++; /* skip trailing whitespace */
2893 :
2894 349162 : if (*nextp == separator)
2895 : {
2896 160770 : nextp++;
2897 306873 : while (scanner_isspace(*nextp))
2898 146103 : nextp++; /* skip leading whitespace for next */
2899 : /* we expect another name, so done remains false */
2900 : }
2901 188392 : else if (*nextp == '\0')
2902 188391 : done = true;
2903 : else
2904 1 : return false; /* invalid syntax */
2905 :
2906 : /* Now safe to overwrite separator with a null */
2907 349161 : *endp = '\0';
2908 :
2909 : /* Truncate name if it's overlength */
2910 349161 : truncate_identifier(curname, strlen(curname), false);
2911 :
2912 : /*
2913 : * Finished isolating current name --- add it to list
2914 : */
2915 349161 : *namelist = lappend(*namelist, curname);
2916 :
2917 : /* Loop back if we didn't reach end of string */
2918 349161 : } while (!done);
2919 :
2920 188391 : return true;
2921 : }
2922 :
2923 :
2924 : /*
2925 : * SplitDirectoriesString --- parse a string containing file/directory names
2926 : *
2927 : * This works fine on file names too; the function name is historical.
2928 : *
2929 : * This is similar to SplitIdentifierString, except that the parsing
2930 : * rules are meant to handle pathnames instead of identifiers: there is
2931 : * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
2932 : * and we apply canonicalize_path() to each extracted string. Because of the
2933 : * last, the returned strings are separately palloc'd rather than being
2934 : * pointers into rawstring --- but we still scribble on rawstring.
2935 : *
2936 : * Inputs:
2937 : * rawstring: the input string; must be modifiable!
2938 : * separator: the separator punctuation expected between directories
2939 : * (typically ',' or ';'). Whitespace may also appear around
2940 : * directories.
2941 : * Outputs:
2942 : * namelist: filled with a palloc'd list of directory names.
2943 : * Caller should list_free_deep() this even on error return.
2944 : *
2945 : * Returns true if okay, false if there is a syntax error in the string.
2946 : *
2947 : * Note that an empty string is considered okay here.
2948 : */
2949 : bool
2950 1033 : SplitDirectoriesString(char *rawstring, char separator,
2951 : List **namelist)
2952 : {
2953 1033 : char *nextp = rawstring;
2954 1033 : bool done = false;
2955 :
2956 1033 : *namelist = NIL;
2957 :
2958 1033 : while (scanner_isspace(*nextp))
2959 0 : nextp++; /* skip leading whitespace */
2960 :
2961 1033 : if (*nextp == '\0')
2962 1 : return true; /* empty string represents empty list */
2963 :
2964 : /* At the top of the loop, we are at start of a new directory. */
2965 : do
2966 : {
2967 : char *curname;
2968 : char *endp;
2969 :
2970 1051 : if (*nextp == '"')
2971 : {
2972 : /* Quoted name --- collapse quote-quote pairs */
2973 0 : curname = scan_quoted_identifier(&endp, &nextp);
2974 0 : if (curname == NULL)
2975 0 : return false; /* mismatched quotes */
2976 : }
2977 : else
2978 : {
2979 : /* Unquoted name --- extends to separator or end of string */
2980 1051 : curname = endp = nextp;
2981 17505 : while (*nextp && *nextp != separator)
2982 : {
2983 : /* trailing whitespace should not be included in name */
2984 16454 : if (!scanner_isspace(*nextp))
2985 16454 : endp = nextp + 1;
2986 16454 : nextp++;
2987 : }
2988 1051 : if (curname == endp)
2989 0 : return false; /* empty unquoted name not allowed */
2990 : }
2991 :
2992 1051 : while (scanner_isspace(*nextp))
2993 0 : nextp++; /* skip trailing whitespace */
2994 :
2995 1051 : if (*nextp == separator)
2996 : {
2997 19 : nextp++;
2998 22 : while (scanner_isspace(*nextp))
2999 3 : nextp++; /* skip leading whitespace for next */
3000 : /* we expect another name, so done remains false */
3001 : }
3002 1032 : else if (*nextp == '\0')
3003 1032 : done = true;
3004 : else
3005 0 : return false; /* invalid syntax */
3006 :
3007 : /* Now safe to overwrite separator with a null */
3008 1051 : *endp = '\0';
3009 :
3010 : /* Truncate path if it's overlength */
3011 1051 : if (strlen(curname) >= MAXPGPATH)
3012 0 : curname[MAXPGPATH - 1] = '\0';
3013 :
3014 : /*
3015 : * Finished isolating current name --- add it to list
3016 : */
3017 1051 : curname = pstrdup(curname);
3018 1051 : canonicalize_path(curname);
3019 1051 : *namelist = lappend(*namelist, curname);
3020 :
3021 : /* Loop back if we didn't reach end of string */
3022 1051 : } while (!done);
3023 :
3024 1032 : return true;
3025 : }
3026 :
3027 :
3028 : /*
3029 : * SplitGUCList --- parse a string containing identifiers or file names
3030 : *
3031 : * This is used to split the value of a GUC_LIST_QUOTE GUC variable, without
3032 : * presuming whether the elements will be taken as identifiers or file names.
3033 : * We assume the input has already been through flatten_set_variable_args(),
3034 : * so that we need never downcase (if appropriate, that was done already).
3035 : * Nor do we ever truncate, since we don't know the correct max length.
3036 : * We disallow embedded whitespace for simplicity (it shouldn't matter,
3037 : * because any embedded whitespace should have led to double-quoting).
3038 : * Otherwise the API is identical to SplitIdentifierString.
3039 : *
3040 : * XXX it's annoying to have so many copies of this string-splitting logic.
3041 : * However, it's not clear that having one function with a bunch of option
3042 : * flags would be much better.
3043 : *
3044 : * XXX there is a version of this function in src/bin/pg_dump/dumputils.c.
3045 : * Be sure to update that if you have to change this.
3046 : *
3047 : * Inputs:
3048 : * rawstring: the input string; must be overwritable! On return, it's
3049 : * been modified to contain the separated identifiers.
3050 : * separator: the separator punctuation expected between identifiers
3051 : * (typically '.' or ','). Whitespace may also appear around
3052 : * identifiers.
3053 : * Outputs:
3054 : * namelist: filled with a palloc'd list of pointers to identifiers within
3055 : * rawstring. Caller should list_free() this even on error return.
3056 : *
3057 : * Returns true if okay, false if there is a syntax error in the string.
3058 : */
3059 : bool
3060 4019 : SplitGUCList(char *rawstring, char separator,
3061 : List **namelist)
3062 : {
3063 4019 : char *nextp = rawstring;
3064 4019 : bool done = false;
3065 :
3066 4019 : *namelist = NIL;
3067 :
3068 4019 : while (scanner_isspace(*nextp))
3069 0 : nextp++; /* skip leading whitespace */
3070 :
3071 4019 : if (*nextp == '\0')
3072 2217 : return true; /* empty string represents empty list */
3073 :
3074 : /* At the top of the loop, we are at start of a new identifier. */
3075 : do
3076 : {
3077 : char *curname;
3078 : char *endp;
3079 :
3080 1886 : curname = scan_identifier(&endp, &nextp, separator, false);
3081 1886 : if (curname == NULL)
3082 0 : return false; /* mismatched quotes or empty name */
3083 :
3084 1886 : while (scanner_isspace(*nextp))
3085 0 : nextp++; /* skip trailing whitespace */
3086 :
3087 1886 : if (*nextp == separator)
3088 : {
3089 84 : nextp++;
3090 164 : while (scanner_isspace(*nextp))
3091 80 : nextp++; /* skip leading whitespace for next */
3092 : /* we expect another name, so done remains false */
3093 : }
3094 1802 : else if (*nextp == '\0')
3095 1802 : done = true;
3096 : else
3097 0 : return false; /* invalid syntax */
3098 :
3099 : /* Now safe to overwrite separator with a null */
3100 1886 : *endp = '\0';
3101 :
3102 : /*
3103 : * Finished isolating current name --- add it to list
3104 : */
3105 1886 : *namelist = lappend(*namelist, curname);
3106 :
3107 : /* Loop back if we didn't reach end of string */
3108 1886 : } while (!done);
3109 :
3110 1802 : return true;
3111 : }
3112 :
3113 : /*
3114 : * appendStringInfoText
3115 : *
3116 : * Append a text to str.
3117 : * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3118 : */
3119 : static void
3120 1146137 : appendStringInfoText(StringInfo str, const text *t)
3121 : {
3122 1146137 : appendBinaryStringInfo(str, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
3123 1146137 : }
3124 :
3125 : /*
3126 : * replace_text
3127 : * replace all occurrences of 'old_sub_str' in 'orig_str'
3128 : * with 'new_sub_str' to form 'new_str'
3129 : *
3130 : * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3131 : * otherwise returns 'new_str'
3132 : */
3133 : Datum
3134 909 : replace_text(PG_FUNCTION_ARGS)
3135 : {
3136 909 : text *src_text = PG_GETARG_TEXT_PP(0);
3137 909 : text *from_sub_text = PG_GETARG_TEXT_PP(1);
3138 909 : text *to_sub_text = PG_GETARG_TEXT_PP(2);
3139 : int src_text_len;
3140 : int from_sub_text_len;
3141 : TextPositionState state;
3142 : text *ret_text;
3143 : int chunk_len;
3144 : char *curr_ptr;
3145 : char *start_ptr;
3146 : StringInfoData str;
3147 : bool found;
3148 :
3149 909 : src_text_len = VARSIZE_ANY_EXHDR(src_text);
3150 909 : from_sub_text_len = VARSIZE_ANY_EXHDR(from_sub_text);
3151 :
3152 : /* Return unmodified source string if empty source or pattern */
3153 909 : if (src_text_len < 1 || from_sub_text_len < 1)
3154 : {
3155 0 : PG_RETURN_TEXT_P(src_text);
3156 : }
3157 :
3158 909 : text_position_setup(src_text, from_sub_text, PG_GET_COLLATION(), &state);
3159 :
3160 909 : found = text_position_next(&state);
3161 :
3162 : /* When the from_sub_text is not found, there is nothing to do. */
3163 909 : if (!found)
3164 : {
3165 188 : text_position_cleanup(&state);
3166 188 : PG_RETURN_TEXT_P(src_text);
3167 : }
3168 721 : curr_ptr = text_position_get_match_ptr(&state);
3169 721 : start_ptr = VARDATA_ANY(src_text);
3170 :
3171 721 : initStringInfo(&str);
3172 :
3173 : do
3174 : {
3175 3866 : CHECK_FOR_INTERRUPTS();
3176 :
3177 : /* copy the data skipped over by last text_position_next() */
3178 3866 : chunk_len = curr_ptr - start_ptr;
3179 3866 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3180 :
3181 3866 : appendStringInfoText(&str, to_sub_text);
3182 :
3183 3866 : start_ptr = curr_ptr + state.last_match_len;
3184 :
3185 3866 : found = text_position_next(&state);
3186 3866 : if (found)
3187 3145 : curr_ptr = text_position_get_match_ptr(&state);
3188 : }
3189 3866 : while (found);
3190 :
3191 : /* copy trailing data */
3192 721 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3193 721 : appendBinaryStringInfo(&str, start_ptr, chunk_len);
3194 :
3195 721 : text_position_cleanup(&state);
3196 :
3197 721 : ret_text = cstring_to_text_with_len(str.data, str.len);
3198 721 : pfree(str.data);
3199 :
3200 721 : PG_RETURN_TEXT_P(ret_text);
3201 : }
3202 :
3203 : /*
3204 : * check_replace_text_has_escape
3205 : *
3206 : * Returns 0 if text contains no backslashes that need processing.
3207 : * Returns 1 if text contains backslashes, but not regexp submatch specifiers.
3208 : * Returns 2 if text contains regexp submatch specifiers (\1 .. \9).
3209 : */
3210 : static int
3211 12528 : check_replace_text_has_escape(const text *replace_text)
3212 : {
3213 12528 : int result = 0;
3214 12528 : const char *p = VARDATA_ANY(replace_text);
3215 12528 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3216 :
3217 25084 : while (p < p_end)
3218 : {
3219 : /* Find next escape char, if any. */
3220 11690 : p = memchr(p, '\\', p_end - p);
3221 11690 : if (p == NULL)
3222 11142 : break;
3223 548 : p++;
3224 : /* Note: a backslash at the end doesn't require extra processing. */
3225 548 : if (p < p_end)
3226 : {
3227 548 : if (*p >= '1' && *p <= '9')
3228 520 : return 2; /* Found a submatch specifier, so done */
3229 28 : result = 1; /* Found some other sequence, keep looking */
3230 28 : p++;
3231 : }
3232 : }
3233 12008 : return result;
3234 : }
3235 :
3236 : /*
3237 : * appendStringInfoRegexpSubstr
3238 : *
3239 : * Append replace_text to str, substituting regexp back references for
3240 : * \n escapes. start_ptr is the start of the match in the source string,
3241 : * at logical character position data_pos.
3242 : */
3243 : static void
3244 170 : appendStringInfoRegexpSubstr(StringInfo str, text *replace_text,
3245 : regmatch_t *pmatch,
3246 : char *start_ptr, int data_pos)
3247 : {
3248 170 : const char *p = VARDATA_ANY(replace_text);
3249 170 : const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3250 :
3251 404 : while (p < p_end)
3252 : {
3253 361 : const char *chunk_start = p;
3254 : int so;
3255 : int eo;
3256 :
3257 : /* Find next escape char, if any. */
3258 361 : p = memchr(p, '\\', p_end - p);
3259 361 : if (p == NULL)
3260 122 : p = p_end;
3261 :
3262 : /* Copy the text we just scanned over, if any. */
3263 361 : if (p > chunk_start)
3264 225 : appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3265 :
3266 : /* Done if at end of string, else advance over escape char. */
3267 361 : if (p >= p_end)
3268 122 : break;
3269 239 : p++;
3270 :
3271 239 : if (p >= p_end)
3272 : {
3273 : /* Escape at very end of input. Treat same as unexpected char */
3274 5 : appendStringInfoChar(str, '\\');
3275 5 : break;
3276 : }
3277 :
3278 234 : if (*p >= '1' && *p <= '9')
3279 190 : {
3280 : /* Use the back reference of regexp. */
3281 190 : int idx = *p - '0';
3282 :
3283 190 : so = pmatch[idx].rm_so;
3284 190 : eo = pmatch[idx].rm_eo;
3285 190 : p++;
3286 : }
3287 44 : else if (*p == '&')
3288 : {
3289 : /* Use the entire matched string. */
3290 15 : so = pmatch[0].rm_so;
3291 15 : eo = pmatch[0].rm_eo;
3292 15 : p++;
3293 : }
3294 29 : else if (*p == '\\')
3295 : {
3296 : /* \\ means transfer one \ to output. */
3297 24 : appendStringInfoChar(str, '\\');
3298 24 : p++;
3299 24 : continue;
3300 : }
3301 : else
3302 : {
3303 : /*
3304 : * If escape char is not followed by any expected char, just treat
3305 : * it as ordinary data to copy. (XXX would it be better to throw
3306 : * an error?)
3307 : */
3308 5 : appendStringInfoChar(str, '\\');
3309 5 : continue;
3310 : }
3311 :
3312 205 : if (so >= 0 && eo >= 0)
3313 : {
3314 : /*
3315 : * Copy the text that is back reference of regexp. Note so and eo
3316 : * are counted in characters not bytes.
3317 : */
3318 : char *chunk_start;
3319 : int chunk_len;
3320 :
3321 : Assert(so >= data_pos);
3322 205 : chunk_start = start_ptr;
3323 205 : chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3324 205 : chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3325 205 : appendBinaryStringInfo(str, chunk_start, chunk_len);
3326 : }
3327 : }
3328 170 : }
3329 :
3330 : /*
3331 : * replace_text_regexp
3332 : *
3333 : * replace substring(s) in src_text that match pattern with replace_text.
3334 : * The replace_text can contain backslash markers to substitute
3335 : * (parts of) the matched text.
3336 : *
3337 : * cflags: regexp compile flags.
3338 : * collation: collation to use.
3339 : * search_start: the character (not byte) offset in src_text at which to
3340 : * begin searching.
3341 : * n: if 0, replace all matches; if > 0, replace only the N'th match.
3342 : */
3343 : text *
3344 12528 : replace_text_regexp(text *src_text, text *pattern_text,
3345 : text *replace_text,
3346 : int cflags, Oid collation,
3347 : int search_start, int n)
3348 : {
3349 : text *ret_text;
3350 : regex_t *re;
3351 12528 : int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3352 12528 : int nmatches = 0;
3353 : StringInfoData buf;
3354 : regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
3355 12528 : int nmatch = lengthof(pmatch);
3356 : pg_wchar *data;
3357 : size_t data_len;
3358 : int data_pos;
3359 : char *start_ptr;
3360 : int escape_status;
3361 :
3362 12528 : initStringInfo(&buf);
3363 :
3364 : /* Convert data string to wide characters. */
3365 12528 : data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3366 12528 : data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3367 :
3368 : /* Check whether replace_text has escapes, especially regexp submatches. */
3369 12528 : escape_status = check_replace_text_has_escape(replace_text);
3370 :
3371 : /* If no regexp submatches, we can use REG_NOSUB. */
3372 12528 : if (escape_status < 2)
3373 : {
3374 12008 : cflags |= REG_NOSUB;
3375 : /* Also tell pg_regexec we only want the whole-match location. */
3376 12008 : nmatch = 1;
3377 : }
3378 :
3379 : /* Prepare the regexp. */
3380 12528 : re = RE_compile_and_cache(pattern_text, cflags, collation);
3381 :
3382 : /* start_ptr points to the data_pos'th character of src_text */
3383 12528 : start_ptr = (char *) VARDATA_ANY(src_text);
3384 12528 : data_pos = 0;
3385 :
3386 16717 : while (search_start <= data_len)
3387 : {
3388 : int regexec_result;
3389 :
3390 16712 : CHECK_FOR_INTERRUPTS();
3391 :
3392 16712 : regexec_result = pg_regexec(re,
3393 : data,
3394 : data_len,
3395 : search_start,
3396 : NULL, /* no details */
3397 : nmatch,
3398 : pmatch,
3399 : 0);
3400 :
3401 16712 : if (regexec_result == REG_NOMATCH)
3402 11079 : break;
3403 :
3404 5633 : if (regexec_result != REG_OKAY)
3405 : {
3406 : char errMsg[100];
3407 :
3408 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3409 0 : ereport(ERROR,
3410 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3411 : errmsg("regular expression failed: %s", errMsg)));
3412 : }
3413 :
3414 : /*
3415 : * Count matches, and decide whether to replace this match.
3416 : */
3417 5633 : nmatches++;
3418 5633 : if (n > 0 && nmatches != n)
3419 : {
3420 : /*
3421 : * No, so advance search_start, but not start_ptr/data_pos. (Thus,
3422 : * we treat the matched text as if it weren't matched, and copy it
3423 : * to the output later.)
3424 : */
3425 50 : search_start = pmatch[0].rm_eo;
3426 50 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3427 0 : search_start++;
3428 50 : continue;
3429 : }
3430 :
3431 : /*
3432 : * Copy the text to the left of the match position. Note we are given
3433 : * character not byte indexes.
3434 : */
3435 5583 : if (pmatch[0].rm_so - data_pos > 0)
3436 : {
3437 : int chunk_len;
3438 :
3439 5456 : chunk_len = charlen_to_bytelen(start_ptr,
3440 5456 : pmatch[0].rm_so - data_pos);
3441 5456 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3442 :
3443 : /*
3444 : * Advance start_ptr over that text, to avoid multiple rescans of
3445 : * it if the replace_text contains multiple back-references.
3446 : */
3447 5456 : start_ptr += chunk_len;
3448 5456 : data_pos = pmatch[0].rm_so;
3449 : }
3450 :
3451 : /*
3452 : * Copy the replace_text, processing escapes if any are present.
3453 : */
3454 5583 : if (escape_status > 0)
3455 170 : appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3456 : start_ptr, data_pos);
3457 : else
3458 5413 : appendStringInfoText(&buf, replace_text);
3459 :
3460 : /* Advance start_ptr and data_pos over the matched text. */
3461 11166 : start_ptr += charlen_to_bytelen(start_ptr,
3462 5583 : pmatch[0].rm_eo - data_pos);
3463 5583 : data_pos = pmatch[0].rm_eo;
3464 :
3465 : /*
3466 : * If we only want to replace one occurrence, we're done.
3467 : */
3468 5583 : if (n > 0)
3469 1444 : break;
3470 :
3471 : /*
3472 : * Advance search position. Normally we start the next search at the
3473 : * end of the previous match; but if the match was of zero length, we
3474 : * have to advance by one character, or we'd just find the same match
3475 : * again.
3476 : */
3477 4139 : search_start = data_pos;
3478 4139 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
3479 10 : search_start++;
3480 : }
3481 :
3482 : /*
3483 : * Copy the text to the right of the last match.
3484 : */
3485 12528 : if (data_pos < data_len)
3486 : {
3487 : int chunk_len;
3488 :
3489 11901 : chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3490 11901 : appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3491 : }
3492 :
3493 12528 : ret_text = cstring_to_text_with_len(buf.data, buf.len);
3494 12528 : pfree(buf.data);
3495 12528 : pfree(data);
3496 :
3497 12528 : return ret_text;
3498 : }
3499 :
3500 : /*
3501 : * split_part
3502 : * parse input string based on provided field separator
3503 : * return N'th item (1 based, negative counts from end)
3504 : */
3505 : Datum
3506 124 : split_part(PG_FUNCTION_ARGS)
3507 : {
3508 124 : text *inputstring = PG_GETARG_TEXT_PP(0);
3509 124 : text *fldsep = PG_GETARG_TEXT_PP(1);
3510 124 : int fldnum = PG_GETARG_INT32(2);
3511 : int inputstring_len;
3512 : int fldsep_len;
3513 : TextPositionState state;
3514 : char *start_ptr;
3515 : char *end_ptr;
3516 : text *result_text;
3517 : bool found;
3518 :
3519 : /* field number is 1 based */
3520 124 : if (fldnum == 0)
3521 4 : ereport(ERROR,
3522 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3523 : errmsg("field position must not be zero")));
3524 :
3525 120 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3526 120 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3527 :
3528 : /* return empty string for empty input string */
3529 120 : if (inputstring_len < 1)
3530 10 : PG_RETURN_TEXT_P(cstring_to_text(""));
3531 :
3532 : /* handle empty field separator */
3533 110 : if (fldsep_len < 1)
3534 : {
3535 : /* if first or last field, return input string, else empty string */
3536 20 : if (fldnum == 1 || fldnum == -1)
3537 10 : PG_RETURN_TEXT_P(inputstring);
3538 : else
3539 10 : PG_RETURN_TEXT_P(cstring_to_text(""));
3540 : }
3541 :
3542 : /* find the first field separator */
3543 90 : text_position_setup(inputstring, fldsep, PG_GET_COLLATION(), &state);
3544 :
3545 90 : found = text_position_next(&state);
3546 :
3547 : /* special case if fldsep not found at all */
3548 90 : if (!found)
3549 : {
3550 18 : text_position_cleanup(&state);
3551 : /* if first or last field, return input string, else empty string */
3552 18 : if (fldnum == 1 || fldnum == -1)
3553 9 : PG_RETURN_TEXT_P(inputstring);
3554 : else
3555 9 : PG_RETURN_TEXT_P(cstring_to_text(""));
3556 : }
3557 :
3558 : /*
3559 : * take care of a negative field number (i.e. count from the right) by
3560 : * converting to a positive field number; we need total number of fields
3561 : */
3562 72 : if (fldnum < 0)
3563 : {
3564 : /* we found a fldsep, so there are at least two fields */
3565 36 : int numfields = 2;
3566 :
3567 54 : while (text_position_next(&state))
3568 18 : numfields++;
3569 :
3570 : /* special case of last field does not require an extra pass */
3571 36 : if (fldnum == -1)
3572 : {
3573 17 : start_ptr = text_position_get_match_ptr(&state) + state.last_match_len;
3574 17 : end_ptr = VARDATA_ANY(inputstring) + inputstring_len;
3575 17 : text_position_cleanup(&state);
3576 17 : PG_RETURN_TEXT_P(cstring_to_text_with_len(start_ptr,
3577 : end_ptr - start_ptr));
3578 : }
3579 :
3580 : /* else, convert fldnum to positive notation */
3581 19 : fldnum += numfields + 1;
3582 :
3583 : /* if nonexistent field, return empty string */
3584 19 : if (fldnum <= 0)
3585 : {
3586 5 : text_position_cleanup(&state);
3587 5 : PG_RETURN_TEXT_P(cstring_to_text(""));
3588 : }
3589 :
3590 : /* reset to pointing at first match, but now with positive fldnum */
3591 14 : text_position_reset(&state);
3592 14 : found = text_position_next(&state);
3593 : Assert(found);
3594 : }
3595 :
3596 : /* identify bounds of first field */
3597 50 : start_ptr = VARDATA_ANY(inputstring);
3598 50 : end_ptr = text_position_get_match_ptr(&state);
3599 :
3600 99 : while (found && --fldnum > 0)
3601 : {
3602 : /* identify bounds of next field */
3603 49 : start_ptr = end_ptr + state.last_match_len;
3604 49 : found = text_position_next(&state);
3605 49 : if (found)
3606 27 : end_ptr = text_position_get_match_ptr(&state);
3607 : }
3608 :
3609 50 : text_position_cleanup(&state);
3610 :
3611 50 : if (fldnum > 0)
3612 : {
3613 : /* N'th field separator not found */
3614 : /* if last field requested, return it, else empty string */
3615 22 : if (fldnum == 1)
3616 : {
3617 17 : int last_len = start_ptr - VARDATA_ANY(inputstring);
3618 :
3619 17 : result_text = cstring_to_text_with_len(start_ptr,
3620 : inputstring_len - last_len);
3621 : }
3622 : else
3623 5 : result_text = cstring_to_text("");
3624 : }
3625 : else
3626 : {
3627 : /* non-last field requested */
3628 28 : result_text = cstring_to_text_with_len(start_ptr, end_ptr - start_ptr);
3629 : }
3630 :
3631 50 : PG_RETURN_TEXT_P(result_text);
3632 : }
3633 :
3634 : /*
3635 : * Convenience function to return true when two text params are equal.
3636 : */
3637 : static bool
3638 306 : text_isequal(text *txt1, text *txt2, Oid collid)
3639 : {
3640 306 : return DatumGetBool(DirectFunctionCall2Coll(texteq,
3641 : collid,
3642 : PointerGetDatum(txt1),
3643 : PointerGetDatum(txt2)));
3644 : }
3645 :
3646 : /*
3647 : * text_to_array
3648 : * parse input string and return text array of elements,
3649 : * based on provided field separator
3650 : */
3651 : Datum
3652 136 : text_to_array(PG_FUNCTION_ARGS)
3653 : {
3654 : SplitTextOutputData tstate;
3655 :
3656 : /* For array output, tstate should start as all zeroes */
3657 136 : memset(&tstate, 0, sizeof(tstate));
3658 :
3659 136 : if (!split_text(fcinfo, &tstate))
3660 5 : PG_RETURN_NULL();
3661 :
3662 131 : if (tstate.astate == NULL)
3663 5 : PG_RETURN_ARRAYTYPE_P(construct_empty_array(TEXTOID));
3664 :
3665 126 : PG_RETURN_DATUM(makeArrayResult(tstate.astate,
3666 : CurrentMemoryContext));
3667 : }
3668 :
3669 : /*
3670 : * text_to_array_null
3671 : * parse input string and return text array of elements,
3672 : * based on provided field separator and null string
3673 : *
3674 : * This is a separate entry point only to prevent the regression tests from
3675 : * complaining about different argument sets for the same internal function.
3676 : */
3677 : Datum
3678 50 : text_to_array_null(PG_FUNCTION_ARGS)
3679 : {
3680 50 : return text_to_array(fcinfo);
3681 : }
3682 :
3683 : /*
3684 : * text_to_table
3685 : * parse input string and return table of elements,
3686 : * based on provided field separator
3687 : */
3688 : Datum
3689 56 : text_to_table(PG_FUNCTION_ARGS)
3690 : {
3691 56 : ReturnSetInfo *rsi = (ReturnSetInfo *) fcinfo->resultinfo;
3692 : SplitTextOutputData tstate;
3693 :
3694 56 : tstate.astate = NULL;
3695 56 : InitMaterializedSRF(fcinfo, MAT_SRF_USE_EXPECTED_DESC);
3696 56 : tstate.tupstore = rsi->setResult;
3697 56 : tstate.tupdesc = rsi->setDesc;
3698 :
3699 56 : (void) split_text(fcinfo, &tstate);
3700 :
3701 56 : return (Datum) 0;
3702 : }
3703 :
3704 : /*
3705 : * text_to_table_null
3706 : * parse input string and return table of elements,
3707 : * based on provided field separator and null string
3708 : *
3709 : * This is a separate entry point only to prevent the regression tests from
3710 : * complaining about different argument sets for the same internal function.
3711 : */
3712 : Datum
3713 16 : text_to_table_null(PG_FUNCTION_ARGS)
3714 : {
3715 16 : return text_to_table(fcinfo);
3716 : }
3717 :
3718 : /*
3719 : * Common code for text_to_array, text_to_array_null, text_to_table
3720 : * and text_to_table_null functions.
3721 : *
3722 : * These are not strict so we have to test for null inputs explicitly.
3723 : * Returns false if result is to be null, else returns true.
3724 : *
3725 : * Note that if the result is valid but empty (zero elements), we return
3726 : * without changing *tstate --- caller must handle that case, too.
3727 : */
3728 : static bool
3729 192 : split_text(FunctionCallInfo fcinfo, SplitTextOutputData *tstate)
3730 : {
3731 : text *inputstring;
3732 : text *fldsep;
3733 : text *null_string;
3734 192 : Oid collation = PG_GET_COLLATION();
3735 : int inputstring_len;
3736 : int fldsep_len;
3737 : char *start_ptr;
3738 : text *result_text;
3739 :
3740 : /* when input string is NULL, then result is NULL too */
3741 192 : if (PG_ARGISNULL(0))
3742 9 : return false;
3743 :
3744 183 : inputstring = PG_GETARG_TEXT_PP(0);
3745 :
3746 : /* fldsep can be NULL */
3747 183 : if (!PG_ARGISNULL(1))
3748 159 : fldsep = PG_GETARG_TEXT_PP(1);
3749 : else
3750 24 : fldsep = NULL;
3751 :
3752 : /* null_string can be NULL or omitted */
3753 183 : if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
3754 66 : null_string = PG_GETARG_TEXT_PP(2);
3755 : else
3756 117 : null_string = NULL;
3757 :
3758 183 : if (fldsep != NULL)
3759 : {
3760 : /*
3761 : * Normal case with non-null fldsep. Use the text_position machinery
3762 : * to search for occurrences of fldsep.
3763 : */
3764 : TextPositionState state;
3765 :
3766 159 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3767 159 : fldsep_len = VARSIZE_ANY_EXHDR(fldsep);
3768 :
3769 : /* return empty set for empty input string */
3770 159 : if (inputstring_len < 1)
3771 46 : return true;
3772 :
3773 : /* empty field separator: return input string as a one-element set */
3774 150 : if (fldsep_len < 1)
3775 : {
3776 37 : split_text_accum_result(tstate, inputstring,
3777 : null_string, collation);
3778 37 : return true;
3779 : }
3780 :
3781 113 : text_position_setup(inputstring, fldsep, collation, &state);
3782 :
3783 113 : start_ptr = VARDATA_ANY(inputstring);
3784 :
3785 : for (;;)
3786 371 : {
3787 : bool found;
3788 : char *end_ptr;
3789 : int chunk_len;
3790 :
3791 484 : CHECK_FOR_INTERRUPTS();
3792 :
3793 484 : found = text_position_next(&state);
3794 484 : if (!found)
3795 : {
3796 : /* fetch last field */
3797 113 : chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
3798 113 : end_ptr = NULL; /* not used, but some compilers complain */
3799 : }
3800 : else
3801 : {
3802 : /* fetch non-last field */
3803 371 : end_ptr = text_position_get_match_ptr(&state);
3804 371 : chunk_len = end_ptr - start_ptr;
3805 : }
3806 :
3807 : /* build a temp text datum to pass to split_text_accum_result */
3808 484 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3809 :
3810 : /* stash away this field */
3811 484 : split_text_accum_result(tstate, result_text,
3812 : null_string, collation);
3813 :
3814 484 : pfree(result_text);
3815 :
3816 484 : if (!found)
3817 113 : break;
3818 :
3819 371 : start_ptr = end_ptr + state.last_match_len;
3820 : }
3821 :
3822 113 : text_position_cleanup(&state);
3823 : }
3824 : else
3825 : {
3826 : const char *end_ptr;
3827 :
3828 : /*
3829 : * When fldsep is NULL, each character in the input string becomes a
3830 : * separate element in the result set. The separator is effectively
3831 : * the space between characters.
3832 : */
3833 24 : inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
3834 :
3835 24 : start_ptr = VARDATA_ANY(inputstring);
3836 24 : end_ptr = start_ptr + inputstring_len;
3837 :
3838 204 : while (inputstring_len > 0)
3839 : {
3840 180 : int chunk_len = pg_mblen_range(start_ptr, end_ptr);
3841 :
3842 180 : CHECK_FOR_INTERRUPTS();
3843 :
3844 : /* build a temp text datum to pass to split_text_accum_result */
3845 180 : result_text = cstring_to_text_with_len(start_ptr, chunk_len);
3846 :
3847 : /* stash away this field */
3848 180 : split_text_accum_result(tstate, result_text,
3849 : null_string, collation);
3850 :
3851 180 : pfree(result_text);
3852 :
3853 180 : start_ptr += chunk_len;
3854 180 : inputstring_len -= chunk_len;
3855 : }
3856 : }
3857 :
3858 137 : return true;
3859 : }
3860 :
3861 : /*
3862 : * Add text item to result set (table or array).
3863 : *
3864 : * This is also responsible for checking to see if the item matches
3865 : * the null_string, in which case we should emit NULL instead.
3866 : */
3867 : static void
3868 701 : split_text_accum_result(SplitTextOutputData *tstate,
3869 : text *field_value,
3870 : text *null_string,
3871 : Oid collation)
3872 : {
3873 701 : bool is_null = false;
3874 :
3875 701 : if (null_string && text_isequal(field_value, null_string, collation))
3876 56 : is_null = true;
3877 :
3878 701 : if (tstate->tupstore)
3879 : {
3880 : Datum values[1];
3881 : bool nulls[1];
3882 :
3883 152 : values[0] = PointerGetDatum(field_value);
3884 152 : nulls[0] = is_null;
3885 :
3886 152 : tuplestore_putvalues(tstate->tupstore,
3887 : tstate->tupdesc,
3888 : values,
3889 : nulls);
3890 : }
3891 : else
3892 : {
3893 549 : tstate->astate = accumArrayResult(tstate->astate,
3894 : PointerGetDatum(field_value),
3895 : is_null,
3896 : TEXTOID,
3897 : CurrentMemoryContext);
3898 : }
3899 701 : }
3900 :
3901 : /*
3902 : * array_to_text
3903 : * concatenate Cstring representation of input array elements
3904 : * using provided field separator
3905 : */
3906 : Datum
3907 44727 : array_to_text(PG_FUNCTION_ARGS)
3908 : {
3909 44727 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
3910 44727 : char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3911 :
3912 44727 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
3913 : }
3914 :
3915 : /*
3916 : * array_to_text_null
3917 : * concatenate Cstring representation of input array elements
3918 : * using provided field separator and null string
3919 : *
3920 : * This version is not strict so we have to test for null inputs explicitly.
3921 : */
3922 : Datum
3923 8 : array_to_text_null(PG_FUNCTION_ARGS)
3924 : {
3925 : ArrayType *v;
3926 : char *fldsep;
3927 : char *null_string;
3928 :
3929 : /* returns NULL when first or second parameter is NULL */
3930 8 : if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
3931 0 : PG_RETURN_NULL();
3932 :
3933 8 : v = PG_GETARG_ARRAYTYPE_P(0);
3934 8 : fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
3935 :
3936 : /* NULL null string is passed through as a null pointer */
3937 8 : if (!PG_ARGISNULL(2))
3938 4 : null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
3939 : else
3940 4 : null_string = NULL;
3941 :
3942 8 : PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
3943 : }
3944 :
3945 : /*
3946 : * common code for array_to_text and array_to_text_null functions
3947 : */
3948 : static text *
3949 44747 : array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v,
3950 : const char *fldsep, const char *null_string)
3951 : {
3952 : text *result;
3953 : int nitems,
3954 : *dims,
3955 : ndims;
3956 : Oid element_type;
3957 : int typlen;
3958 : bool typbyval;
3959 : char typalign;
3960 : uint8 typalignby;
3961 : StringInfoData buf;
3962 44747 : bool printed = false;
3963 : char *p;
3964 : uint8 *bitmap;
3965 : int bitmask;
3966 : int i;
3967 : ArrayMetaState *my_extra;
3968 :
3969 44747 : ndims = ARR_NDIM(v);
3970 44747 : dims = ARR_DIMS(v);
3971 44747 : nitems = ArrayGetNItems(ndims, dims);
3972 :
3973 : /* if there are no elements, return an empty string */
3974 44747 : if (nitems == 0)
3975 27939 : return cstring_to_text_with_len("", 0);
3976 :
3977 16808 : element_type = ARR_ELEMTYPE(v);
3978 16808 : initStringInfo(&buf);
3979 :
3980 : /*
3981 : * We arrange to look up info about element type, including its output
3982 : * conversion proc, only once per series of calls, assuming the element
3983 : * type doesn't change underneath us.
3984 : */
3985 16808 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3986 16808 : if (my_extra == NULL)
3987 : {
3988 879 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
3989 : sizeof(ArrayMetaState));
3990 879 : my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
3991 879 : my_extra->element_type = ~element_type;
3992 : }
3993 :
3994 16808 : if (my_extra->element_type != element_type)
3995 : {
3996 : /*
3997 : * Get info about element type, including its output conversion proc
3998 : */
3999 879 : get_type_io_data(element_type, IOFunc_output,
4000 : &my_extra->typlen, &my_extra->typbyval,
4001 : &my_extra->typalign, &my_extra->typdelim,
4002 : &my_extra->typioparam, &my_extra->typiofunc);
4003 879 : fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
4004 879 : fcinfo->flinfo->fn_mcxt);
4005 879 : my_extra->element_type = element_type;
4006 : }
4007 16808 : typlen = my_extra->typlen;
4008 16808 : typbyval = my_extra->typbyval;
4009 16808 : typalign = my_extra->typalign;
4010 16808 : typalignby = typalign_to_alignby(typalign);
4011 :
4012 16808 : p = ARR_DATA_PTR(v);
4013 16808 : bitmap = ARR_NULLBITMAP(v);
4014 16808 : bitmask = 1;
4015 :
4016 57325 : for (i = 0; i < nitems; i++)
4017 : {
4018 : Datum itemvalue;
4019 : char *value;
4020 :
4021 : /* Get source element, checking for NULL */
4022 40517 : if (bitmap && (*bitmap & bitmask) == 0)
4023 : {
4024 : /* if null_string is NULL, we just ignore null elements */
4025 12 : if (null_string != NULL)
4026 : {
4027 4 : if (printed)
4028 4 : appendStringInfo(&buf, "%s%s", fldsep, null_string);
4029 : else
4030 0 : appendStringInfoString(&buf, null_string);
4031 4 : printed = true;
4032 : }
4033 : }
4034 : else
4035 : {
4036 40505 : itemvalue = fetch_att(p, typbyval, typlen);
4037 :
4038 40505 : value = OutputFunctionCall(&my_extra->proc, itemvalue);
4039 :
4040 40505 : if (printed)
4041 23697 : appendStringInfo(&buf, "%s%s", fldsep, value);
4042 : else
4043 16808 : appendStringInfoString(&buf, value);
4044 40505 : printed = true;
4045 :
4046 40505 : p = att_addlength_pointer(p, typlen, p);
4047 40505 : p = (char *) att_nominal_alignby(p, typalignby);
4048 : }
4049 :
4050 : /* advance bitmap pointer if any */
4051 40517 : if (bitmap)
4052 : {
4053 72 : bitmask <<= 1;
4054 72 : if (bitmask == 0x100)
4055 : {
4056 0 : bitmap++;
4057 0 : bitmask = 1;
4058 : }
4059 : }
4060 : }
4061 :
4062 16808 : result = cstring_to_text_with_len(buf.data, buf.len);
4063 16808 : pfree(buf.data);
4064 :
4065 16808 : return result;
4066 : }
4067 :
4068 : /*
4069 : * Workhorse for to_bin, to_oct, and to_hex. Note that base must be > 1 and <=
4070 : * 16.
4071 : */
4072 : static inline text *
4073 20389 : convert_to_base(uint64 value, int base)
4074 : {
4075 20389 : const char *digits = "0123456789abcdef";
4076 :
4077 : /* We size the buffer for to_bin's longest possible return value. */
4078 : char buf[sizeof(uint64) * BITS_PER_BYTE];
4079 20389 : char *const end = buf + sizeof(buf);
4080 20389 : char *ptr = end;
4081 :
4082 : Assert(base > 1);
4083 : Assert(base <= 16);
4084 :
4085 : do
4086 : {
4087 40307 : *--ptr = digits[value % base];
4088 40307 : value /= base;
4089 40307 : } while (ptr > buf && value);
4090 :
4091 20389 : return cstring_to_text_with_len(ptr, end - ptr);
4092 : }
4093 :
4094 : /*
4095 : * Convert an integer to a string containing a base-2 (binary) representation
4096 : * of the number.
4097 : */
4098 : Datum
4099 10 : to_bin32(PG_FUNCTION_ARGS)
4100 : {
4101 10 : uint64 value = (uint32) PG_GETARG_INT32(0);
4102 :
4103 10 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4104 : }
4105 : Datum
4106 10 : to_bin64(PG_FUNCTION_ARGS)
4107 : {
4108 10 : uint64 value = (uint64) PG_GETARG_INT64(0);
4109 :
4110 10 : PG_RETURN_TEXT_P(convert_to_base(value, 2));
4111 : }
4112 :
4113 : /*
4114 : * Convert an integer to a string containing a base-8 (oct) representation of
4115 : * the number.
4116 : */
4117 : Datum
4118 10 : to_oct32(PG_FUNCTION_ARGS)
4119 : {
4120 10 : uint64 value = (uint32) PG_GETARG_INT32(0);
4121 :
4122 10 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4123 : }
4124 : Datum
4125 10 : to_oct64(PG_FUNCTION_ARGS)
4126 : {
4127 10 : uint64 value = (uint64) PG_GETARG_INT64(0);
4128 :
4129 10 : PG_RETURN_TEXT_P(convert_to_base(value, 8));
4130 : }
4131 :
4132 : /*
4133 : * Convert an integer to a string containing a base-16 (hex) representation of
4134 : * the number.
4135 : */
4136 : Datum
4137 20339 : to_hex32(PG_FUNCTION_ARGS)
4138 : {
4139 20339 : uint64 value = (uint32) PG_GETARG_INT32(0);
4140 :
4141 20339 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4142 : }
4143 : Datum
4144 10 : to_hex64(PG_FUNCTION_ARGS)
4145 : {
4146 10 : uint64 value = (uint64) PG_GETARG_INT64(0);
4147 :
4148 10 : PG_RETURN_TEXT_P(convert_to_base(value, 16));
4149 : }
4150 :
4151 : /*
4152 : * Return the size of a datum, possibly compressed
4153 : *
4154 : * Works on any data type
4155 : */
4156 : Datum
4157 71 : pg_column_size(PG_FUNCTION_ARGS)
4158 : {
4159 71 : Datum value = PG_GETARG_DATUM(0);
4160 : int32 result;
4161 : int typlen;
4162 :
4163 : /* On first call, get the input type's typlen, and save at *fn_extra */
4164 71 : if (fcinfo->flinfo->fn_extra == NULL)
4165 : {
4166 : /* Lookup the datatype of the supplied argument */
4167 71 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4168 :
4169 71 : typlen = get_typlen(argtypeid);
4170 71 : if (typlen == 0) /* should not happen */
4171 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4172 :
4173 71 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4174 : sizeof(int));
4175 71 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4176 : }
4177 : else
4178 0 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4179 :
4180 71 : if (typlen == -1)
4181 : {
4182 : /* varlena type, possibly toasted */
4183 71 : result = toast_datum_size(value);
4184 : }
4185 0 : else if (typlen == -2)
4186 : {
4187 : /* cstring */
4188 0 : result = strlen(DatumGetCString(value)) + 1;
4189 : }
4190 : else
4191 : {
4192 : /* ordinary fixed-width type */
4193 0 : result = typlen;
4194 : }
4195 :
4196 71 : PG_RETURN_INT32(result);
4197 : }
4198 :
4199 : /*
4200 : * Return the compression method stored in the compressed attribute. Return
4201 : * NULL for non varlena type or uncompressed data.
4202 : */
4203 : Datum
4204 128 : pg_column_compression(PG_FUNCTION_ARGS)
4205 : {
4206 : int typlen;
4207 : char *result;
4208 : ToastCompressionId cmid;
4209 :
4210 : /* On first call, get the input type's typlen, and save at *fn_extra */
4211 128 : if (fcinfo->flinfo->fn_extra == NULL)
4212 : {
4213 : /* Lookup the datatype of the supplied argument */
4214 104 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4215 :
4216 104 : typlen = get_typlen(argtypeid);
4217 104 : if (typlen == 0) /* should not happen */
4218 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4219 :
4220 104 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4221 : sizeof(int));
4222 104 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4223 : }
4224 : else
4225 24 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4226 :
4227 128 : if (typlen != -1)
4228 0 : PG_RETURN_NULL();
4229 :
4230 : /* get the compression method id stored in the compressed varlena */
4231 128 : cmid = toast_get_compression_id((varlena *)
4232 128 : DatumGetPointer(PG_GETARG_DATUM(0)));
4233 128 : if (cmid == TOAST_INVALID_COMPRESSION_ID)
4234 28 : PG_RETURN_NULL();
4235 :
4236 : /* convert compression method id to compression method name */
4237 100 : switch (cmid)
4238 : {
4239 56 : case TOAST_PGLZ_COMPRESSION_ID:
4240 56 : result = "pglz";
4241 56 : break;
4242 44 : case TOAST_LZ4_COMPRESSION_ID:
4243 44 : result = "lz4";
4244 44 : break;
4245 0 : default:
4246 0 : elog(ERROR, "invalid compression method id %d", cmid);
4247 : }
4248 :
4249 100 : PG_RETURN_TEXT_P(cstring_to_text(result));
4250 : }
4251 :
4252 : /*
4253 : * Return the chunk_id of the on-disk TOASTed value. Return NULL if the value
4254 : * is un-TOASTed or not on-disk.
4255 : */
4256 : Datum
4257 34 : pg_column_toast_chunk_id(PG_FUNCTION_ARGS)
4258 : {
4259 : int typlen;
4260 : varlena *attr;
4261 : varatt_external toast_pointer;
4262 :
4263 : /* On first call, get the input type's typlen, and save at *fn_extra */
4264 34 : if (fcinfo->flinfo->fn_extra == NULL)
4265 : {
4266 : /* Lookup the datatype of the supplied argument */
4267 26 : Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4268 :
4269 26 : typlen = get_typlen(argtypeid);
4270 26 : if (typlen == 0) /* should not happen */
4271 0 : elog(ERROR, "cache lookup failed for type %u", argtypeid);
4272 :
4273 26 : fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4274 : sizeof(int));
4275 26 : *((int *) fcinfo->flinfo->fn_extra) = typlen;
4276 : }
4277 : else
4278 8 : typlen = *((int *) fcinfo->flinfo->fn_extra);
4279 :
4280 34 : if (typlen != -1)
4281 0 : PG_RETURN_NULL();
4282 :
4283 34 : attr = (varlena *) DatumGetPointer(PG_GETARG_DATUM(0));
4284 :
4285 34 : if (!VARATT_IS_EXTERNAL_ONDISK(attr))
4286 8 : PG_RETURN_NULL();
4287 :
4288 26 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
4289 :
4290 26 : PG_RETURN_OID(toast_pointer.va_valueid);
4291 : }
4292 :
4293 : /*
4294 : * string_agg - Concatenates values and returns string.
4295 : *
4296 : * Syntax: string_agg(value text, delimiter text) RETURNS text
4297 : *
4298 : * Note: Any NULL values are ignored. The first-call delimiter isn't
4299 : * actually used at all, and on subsequent calls the delimiter precedes
4300 : * the associated value.
4301 : */
4302 :
4303 : /* subroutine to initialize state */
4304 : static StringInfo
4305 1607 : makeStringAggState(FunctionCallInfo fcinfo)
4306 : {
4307 : StringInfo state;
4308 : MemoryContext aggcontext;
4309 : MemoryContext oldcontext;
4310 :
4311 1607 : if (!AggCheckCallContext(fcinfo, &aggcontext))
4312 : {
4313 : /* cannot be called directly because of internal-type argument */
4314 0 : elog(ERROR, "string_agg_transfn called in non-aggregate context");
4315 : }
4316 :
4317 : /*
4318 : * Create state in aggregate context. It'll stay there across subsequent
4319 : * calls.
4320 : */
4321 1607 : oldcontext = MemoryContextSwitchTo(aggcontext);
4322 1607 : state = makeStringInfo();
4323 1607 : MemoryContextSwitchTo(oldcontext);
4324 :
4325 1607 : return state;
4326 : }
4327 :
4328 : Datum
4329 578461 : string_agg_transfn(PG_FUNCTION_ARGS)
4330 : {
4331 : StringInfo state;
4332 :
4333 578461 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4334 :
4335 : /* Append the value unless null, preceding it with the delimiter. */
4336 578461 : if (!PG_ARGISNULL(1))
4337 : {
4338 568429 : text *value = PG_GETARG_TEXT_PP(1);
4339 568429 : bool isfirst = false;
4340 :
4341 : /*
4342 : * You might think we can just throw away the first delimiter, however
4343 : * we must keep it as we may be a parallel worker doing partial
4344 : * aggregation building a state to send to the main process. We need
4345 : * to keep the delimiter of every aggregation so that the combine
4346 : * function can properly join up the strings of two separately
4347 : * partially aggregated results. The first delimiter is only stripped
4348 : * off in the final function. To know how much to strip off the front
4349 : * of the string, we store the length of the first delimiter in the
4350 : * StringInfo's cursor field, which we don't otherwise need here.
4351 : */
4352 568429 : if (state == NULL)
4353 : {
4354 1387 : state = makeStringAggState(fcinfo);
4355 1387 : isfirst = true;
4356 : }
4357 :
4358 568429 : if (!PG_ARGISNULL(2))
4359 : {
4360 568429 : text *delim = PG_GETARG_TEXT_PP(2);
4361 :
4362 568429 : appendStringInfoText(state, delim);
4363 568429 : if (isfirst)
4364 1387 : state->cursor = VARSIZE_ANY_EXHDR(delim);
4365 : }
4366 :
4367 568429 : appendStringInfoText(state, value);
4368 : }
4369 :
4370 : /*
4371 : * The transition type for string_agg() is declared to be "internal",
4372 : * which is a pass-by-value type the same size as a pointer.
4373 : */
4374 578461 : if (state)
4375 578401 : PG_RETURN_POINTER(state);
4376 60 : PG_RETURN_NULL();
4377 : }
4378 :
4379 : /*
4380 : * string_agg_combine
4381 : * Aggregate combine function for string_agg(text) and string_agg(bytea)
4382 : */
4383 : Datum
4384 140 : string_agg_combine(PG_FUNCTION_ARGS)
4385 : {
4386 : StringInfo state1;
4387 : StringInfo state2;
4388 : MemoryContext agg_context;
4389 :
4390 140 : if (!AggCheckCallContext(fcinfo, &agg_context))
4391 0 : elog(ERROR, "aggregate function called in non-aggregate context");
4392 :
4393 140 : state1 = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4394 140 : state2 = PG_ARGISNULL(1) ? NULL : (StringInfo) PG_GETARG_POINTER(1);
4395 :
4396 140 : if (state2 == NULL)
4397 : {
4398 : /*
4399 : * NULL state2 is easy, just return state1, which we know is already
4400 : * in the agg_context
4401 : */
4402 0 : if (state1 == NULL)
4403 0 : PG_RETURN_NULL();
4404 0 : PG_RETURN_POINTER(state1);
4405 : }
4406 :
4407 140 : if (state1 == NULL)
4408 : {
4409 : /* We must copy state2's data into the agg_context */
4410 : MemoryContext old_context;
4411 :
4412 80 : old_context = MemoryContextSwitchTo(agg_context);
4413 80 : state1 = makeStringAggState(fcinfo);
4414 80 : appendBinaryStringInfo(state1, state2->data, state2->len);
4415 80 : state1->cursor = state2->cursor;
4416 80 : MemoryContextSwitchTo(old_context);
4417 : }
4418 60 : else if (state2->len > 0)
4419 : {
4420 : /* Combine ... state1->cursor does not change in this case */
4421 60 : appendBinaryStringInfo(state1, state2->data, state2->len);
4422 : }
4423 :
4424 140 : PG_RETURN_POINTER(state1);
4425 : }
4426 :
4427 : /*
4428 : * string_agg_serialize
4429 : * Aggregate serialize function for string_agg(text) and string_agg(bytea)
4430 : *
4431 : * This is strict, so we need not handle NULL input
4432 : */
4433 : Datum
4434 140 : string_agg_serialize(PG_FUNCTION_ARGS)
4435 : {
4436 : StringInfo state;
4437 : StringInfoData buf;
4438 : bytea *result;
4439 :
4440 : /* cannot be called directly because of internal-type argument */
4441 : Assert(AggCheckCallContext(fcinfo, NULL));
4442 :
4443 140 : state = (StringInfo) PG_GETARG_POINTER(0);
4444 :
4445 140 : pq_begintypsend(&buf);
4446 :
4447 : /* cursor */
4448 140 : pq_sendint(&buf, state->cursor, 4);
4449 :
4450 : /* data */
4451 140 : pq_sendbytes(&buf, state->data, state->len);
4452 :
4453 140 : result = pq_endtypsend(&buf);
4454 :
4455 140 : PG_RETURN_BYTEA_P(result);
4456 : }
4457 :
4458 : /*
4459 : * string_agg_deserialize
4460 : * Aggregate deserial function for string_agg(text) and string_agg(bytea)
4461 : *
4462 : * This is strict, so we need not handle NULL input
4463 : */
4464 : Datum
4465 140 : string_agg_deserialize(PG_FUNCTION_ARGS)
4466 : {
4467 : bytea *sstate;
4468 : StringInfo result;
4469 : StringInfoData buf;
4470 : char *data;
4471 : int datalen;
4472 :
4473 : /* cannot be called directly because of internal-type argument */
4474 : Assert(AggCheckCallContext(fcinfo, NULL));
4475 :
4476 140 : sstate = PG_GETARG_BYTEA_PP(0);
4477 :
4478 : /*
4479 : * Initialize a StringInfo so that we can "receive" it using the standard
4480 : * recv-function infrastructure.
4481 : */
4482 140 : initReadOnlyStringInfo(&buf, VARDATA_ANY(sstate),
4483 140 : VARSIZE_ANY_EXHDR(sstate));
4484 :
4485 140 : result = makeStringAggState(fcinfo);
4486 :
4487 : /* cursor */
4488 140 : result->cursor = pq_getmsgint(&buf, 4);
4489 :
4490 : /* data */
4491 140 : datalen = VARSIZE_ANY_EXHDR(sstate) - 4;
4492 140 : data = (char *) pq_getmsgbytes(&buf, datalen);
4493 140 : appendBinaryStringInfo(result, data, datalen);
4494 :
4495 140 : pq_getmsgend(&buf);
4496 :
4497 140 : PG_RETURN_POINTER(result);
4498 : }
4499 :
4500 : Datum
4501 1413 : string_agg_finalfn(PG_FUNCTION_ARGS)
4502 : {
4503 : StringInfo state;
4504 :
4505 : /* cannot be called directly because of internal-type argument */
4506 : Assert(AggCheckCallContext(fcinfo, NULL));
4507 :
4508 1413 : state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4509 :
4510 1413 : if (state != NULL)
4511 : {
4512 : /* As per comment in transfn, strip data before the cursor position */
4513 1357 : PG_RETURN_TEXT_P(cstring_to_text_with_len(&state->data[state->cursor],
4514 : state->len - state->cursor));
4515 : }
4516 : else
4517 56 : PG_RETURN_NULL();
4518 : }
4519 :
4520 : /*
4521 : * Prepare cache with fmgr info for the output functions of the datatypes of
4522 : * the arguments of a concat-like function, beginning with argument "argidx".
4523 : * (Arguments before that will have corresponding slots in the resulting
4524 : * FmgrInfo array, but we don't fill those slots.)
4525 : */
4526 : static FmgrInfo *
4527 90 : build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
4528 : {
4529 : FmgrInfo *foutcache;
4530 : int i;
4531 :
4532 : /* We keep the info in fn_mcxt so it survives across calls */
4533 90 : foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4534 90 : PG_NARGS() * sizeof(FmgrInfo));
4535 :
4536 328 : for (i = argidx; i < PG_NARGS(); i++)
4537 : {
4538 : Oid valtype;
4539 : Oid typOutput;
4540 : bool typIsVarlena;
4541 :
4542 238 : valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4543 238 : if (!OidIsValid(valtype))
4544 0 : elog(ERROR, "could not determine data type of concat() input");
4545 :
4546 238 : getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4547 238 : fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
4548 : }
4549 :
4550 90 : fcinfo->flinfo->fn_extra = foutcache;
4551 :
4552 90 : return foutcache;
4553 : }
4554 :
4555 : /*
4556 : * Implementation of both concat() and concat_ws().
4557 : *
4558 : * sepstr is the separator string to place between values.
4559 : * argidx identifies the first argument to concatenate (counting from zero);
4560 : * note that this must be constant across any one series of calls.
4561 : *
4562 : * Returns NULL if result should be NULL, else text value.
4563 : */
4564 : static text *
4565 195 : concat_internal(const char *sepstr, int argidx,
4566 : FunctionCallInfo fcinfo)
4567 : {
4568 : text *result;
4569 : StringInfoData str;
4570 : FmgrInfo *foutcache;
4571 195 : bool first_arg = true;
4572 : int i;
4573 :
4574 : /*
4575 : * concat(VARIADIC some-array) is essentially equivalent to
4576 : * array_to_text(), ie concat the array elements with the given separator.
4577 : * So we just pass the case off to that code.
4578 : */
4579 195 : if (get_fn_expr_variadic(fcinfo->flinfo))
4580 : {
4581 : ArrayType *arr;
4582 :
4583 : /* Should have just the one argument */
4584 : Assert(argidx == PG_NARGS() - 1);
4585 :
4586 : /* concat(VARIADIC NULL) is defined as NULL */
4587 20 : if (PG_ARGISNULL(argidx))
4588 8 : return NULL;
4589 :
4590 : /*
4591 : * Non-null argument had better be an array. We assume that any call
4592 : * context that could let get_fn_expr_variadic return true will have
4593 : * checked that a VARIADIC-labeled parameter actually is an array. So
4594 : * it should be okay to just Assert that it's an array rather than
4595 : * doing a full-fledged error check.
4596 : */
4597 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, argidx))));
4598 :
4599 : /* OK, safe to fetch the array value */
4600 12 : arr = PG_GETARG_ARRAYTYPE_P(argidx);
4601 :
4602 : /*
4603 : * And serialize the array. We tell array_to_text to ignore null
4604 : * elements, which matches the behavior of the loop below.
4605 : */
4606 12 : return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4607 : }
4608 :
4609 : /* Normal case without explicit VARIADIC marker */
4610 175 : initStringInfo(&str);
4611 :
4612 : /* Get output function info, building it if first time through */
4613 175 : foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
4614 175 : if (foutcache == NULL)
4615 90 : foutcache = build_concat_foutcache(fcinfo, argidx);
4616 :
4617 608 : for (i = argidx; i < PG_NARGS(); i++)
4618 : {
4619 433 : if (!PG_ARGISNULL(i))
4620 : {
4621 381 : Datum value = PG_GETARG_DATUM(i);
4622 :
4623 : /* add separator if appropriate */
4624 381 : if (first_arg)
4625 171 : first_arg = false;
4626 : else
4627 210 : appendStringInfoString(&str, sepstr);
4628 :
4629 : /* call the appropriate type output function, append the result */
4630 381 : appendStringInfoString(&str,
4631 381 : OutputFunctionCall(&foutcache[i], value));
4632 : }
4633 : }
4634 :
4635 175 : result = cstring_to_text_with_len(str.data, str.len);
4636 175 : pfree(str.data);
4637 :
4638 175 : return result;
4639 : }
4640 :
4641 : /*
4642 : * Concatenate all arguments. NULL arguments are ignored.
4643 : */
4644 : Datum
4645 143 : text_concat(PG_FUNCTION_ARGS)
4646 : {
4647 : text *result;
4648 :
4649 143 : result = concat_internal("", 0, fcinfo);
4650 143 : if (result == NULL)
4651 4 : PG_RETURN_NULL();
4652 139 : PG_RETURN_TEXT_P(result);
4653 : }
4654 :
4655 : /*
4656 : * Concatenate all but first argument value with separators. The first
4657 : * parameter is used as the separator. NULL arguments are ignored.
4658 : */
4659 : Datum
4660 56 : text_concat_ws(PG_FUNCTION_ARGS)
4661 : {
4662 : char *sep;
4663 : text *result;
4664 :
4665 : /* return NULL when separator is NULL */
4666 56 : if (PG_ARGISNULL(0))
4667 4 : PG_RETURN_NULL();
4668 52 : sep = text_to_cstring(PG_GETARG_TEXT_PP(0));
4669 :
4670 52 : result = concat_internal(sep, 1, fcinfo);
4671 52 : if (result == NULL)
4672 4 : PG_RETURN_NULL();
4673 48 : PG_RETURN_TEXT_P(result);
4674 : }
4675 :
4676 : /*
4677 : * Return first n characters in the string. When n is negative,
4678 : * return all but last |n| characters.
4679 : */
4680 : Datum
4681 1432 : text_left(PG_FUNCTION_ARGS)
4682 : {
4683 1432 : int n = PG_GETARG_INT32(1);
4684 :
4685 1432 : if (n < 0)
4686 : {
4687 20 : text *str = PG_GETARG_TEXT_PP(0);
4688 20 : const char *p = VARDATA_ANY(str);
4689 20 : int len = VARSIZE_ANY_EXHDR(str);
4690 : int rlen;
4691 :
4692 20 : n = pg_mbstrlen_with_len(p, len) + n;
4693 20 : rlen = pg_mbcharcliplen(p, len, n);
4694 20 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p, rlen));
4695 : }
4696 : else
4697 1412 : PG_RETURN_TEXT_P(text_substring(PG_GETARG_DATUM(0), 1, n, false));
4698 : }
4699 :
4700 : /*
4701 : * Return last n characters in the string. When n is negative,
4702 : * return all but first |n| characters.
4703 : */
4704 : Datum
4705 44 : text_right(PG_FUNCTION_ARGS)
4706 : {
4707 44 : text *str = PG_GETARG_TEXT_PP(0);
4708 44 : const char *p = VARDATA_ANY(str);
4709 44 : int len = VARSIZE_ANY_EXHDR(str);
4710 44 : int n = PG_GETARG_INT32(1);
4711 : int off;
4712 :
4713 44 : if (n < 0)
4714 20 : n = -n;
4715 : else
4716 24 : n = pg_mbstrlen_with_len(p, len) - n;
4717 44 : off = pg_mbcharcliplen(p, len, n);
4718 :
4719 44 : PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4720 : }
4721 :
4722 : /*
4723 : * Return reversed string
4724 : */
4725 : Datum
4726 29 : text_reverse(PG_FUNCTION_ARGS)
4727 : {
4728 29 : text *str = PG_GETARG_TEXT_PP(0);
4729 29 : const char *p = VARDATA_ANY(str);
4730 29 : int len = VARSIZE_ANY_EXHDR(str);
4731 29 : const char *endp = p + len;
4732 : text *result;
4733 : char *dst;
4734 :
4735 29 : result = palloc(len + VARHDRSZ);
4736 29 : dst = (char *) VARDATA(result) + len;
4737 29 : SET_VARSIZE(result, len + VARHDRSZ);
4738 :
4739 29 : if (pg_database_encoding_max_length() > 1)
4740 : {
4741 : /* multibyte version */
4742 222 : while (p < endp)
4743 : {
4744 : int sz;
4745 :
4746 197 : sz = pg_mblen_range(p, endp);
4747 193 : dst -= sz;
4748 193 : memcpy(dst, p, sz);
4749 193 : p += sz;
4750 : }
4751 : }
4752 : else
4753 : {
4754 : /* single byte version */
4755 0 : while (p < endp)
4756 0 : *(--dst) = *p++;
4757 : }
4758 :
4759 25 : PG_RETURN_TEXT_P(result);
4760 : }
4761 :
4762 :
4763 : /*
4764 : * Support macros for text_format()
4765 : */
4766 : #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4767 :
4768 : #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4769 : do { \
4770 : if (++(ptr) >= (end_ptr)) \
4771 : ereport(ERROR, \
4772 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4773 : errmsg("unterminated format() type specifier"), \
4774 : errhint("For a single \"%%\" use \"%%%%\"."))); \
4775 : } while (0)
4776 :
4777 : /*
4778 : * Returns a formatted string
4779 : */
4780 : Datum
4781 22156 : text_format(PG_FUNCTION_ARGS)
4782 : {
4783 : text *fmt;
4784 : StringInfoData str;
4785 : const char *cp;
4786 : const char *start_ptr;
4787 : const char *end_ptr;
4788 : text *result;
4789 : int arg;
4790 : bool funcvariadic;
4791 : int nargs;
4792 22156 : Datum *elements = NULL;
4793 22156 : bool *nulls = NULL;
4794 22156 : Oid element_type = InvalidOid;
4795 22156 : Oid prev_type = InvalidOid;
4796 22156 : Oid prev_width_type = InvalidOid;
4797 : FmgrInfo typoutputfinfo;
4798 : FmgrInfo typoutputinfo_width;
4799 :
4800 : /* When format string is null, immediately return null */
4801 22156 : if (PG_ARGISNULL(0))
4802 4 : PG_RETURN_NULL();
4803 :
4804 : /* If argument is marked VARIADIC, expand array into elements */
4805 22152 : if (get_fn_expr_variadic(fcinfo->flinfo))
4806 : {
4807 : ArrayType *arr;
4808 : int16 elmlen;
4809 : bool elmbyval;
4810 : char elmalign;
4811 : int nitems;
4812 :
4813 : /* Should have just the one argument */
4814 : Assert(PG_NARGS() == 2);
4815 :
4816 : /* If argument is NULL, we treat it as zero-length array */
4817 32 : if (PG_ARGISNULL(1))
4818 4 : nitems = 0;
4819 : else
4820 : {
4821 : /*
4822 : * Non-null argument had better be an array. We assume that any
4823 : * call context that could let get_fn_expr_variadic return true
4824 : * will have checked that a VARIADIC-labeled parameter actually is
4825 : * an array. So it should be okay to just Assert that it's an
4826 : * array rather than doing a full-fledged error check.
4827 : */
4828 : Assert(OidIsValid(get_base_element_type(get_fn_expr_argtype(fcinfo->flinfo, 1))));
4829 :
4830 : /* OK, safe to fetch the array value */
4831 28 : arr = PG_GETARG_ARRAYTYPE_P(1);
4832 :
4833 : /* Get info about array element type */
4834 28 : element_type = ARR_ELEMTYPE(arr);
4835 28 : get_typlenbyvalalign(element_type,
4836 : &elmlen, &elmbyval, &elmalign);
4837 :
4838 : /* Extract all array elements */
4839 28 : deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
4840 : &elements, &nulls, &nitems);
4841 : }
4842 :
4843 32 : nargs = nitems + 1;
4844 32 : funcvariadic = true;
4845 : }
4846 : else
4847 : {
4848 : /* Non-variadic case, we'll process the arguments individually */
4849 22120 : nargs = PG_NARGS();
4850 22120 : funcvariadic = false;
4851 : }
4852 :
4853 : /* Setup for main loop. */
4854 22152 : fmt = PG_GETARG_TEXT_PP(0);
4855 22152 : start_ptr = VARDATA_ANY(fmt);
4856 22152 : end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
4857 22152 : initStringInfo(&str);
4858 22152 : arg = 1; /* next argument position to print */
4859 :
4860 : /* Scan format string, looking for conversion specifiers. */
4861 705316 : for (cp = start_ptr; cp < end_ptr; cp++)
4862 : {
4863 : int argpos;
4864 : int widthpos;
4865 : int flags;
4866 : int width;
4867 : Datum value;
4868 : bool isNull;
4869 : Oid typid;
4870 :
4871 : /*
4872 : * If it's not the start of a conversion specifier, just copy it to
4873 : * the output buffer.
4874 : */
4875 683204 : if (*cp != '%')
4876 : {
4877 637531 : appendStringInfoCharMacro(&str, *cp);
4878 637543 : continue;
4879 : }
4880 :
4881 45673 : ADVANCE_PARSE_POINTER(cp, end_ptr);
4882 :
4883 : /* Easy case: %% outputs a single % */
4884 45673 : if (*cp == '%')
4885 : {
4886 12 : appendStringInfoCharMacro(&str, *cp);
4887 12 : continue;
4888 : }
4889 :
4890 : /* Parse the optional portions of the format specifier */
4891 45661 : cp = text_format_parse_format(cp, end_ptr,
4892 : &argpos, &widthpos,
4893 : &flags, &width);
4894 :
4895 : /*
4896 : * Next we should see the main conversion specifier. Whether or not
4897 : * an argument position was present, it's known that at least one
4898 : * character remains in the string at this point. Experience suggests
4899 : * that it's worth checking that that character is one of the expected
4900 : * ones before we try to fetch arguments, so as to produce the least
4901 : * confusing response to a mis-formatted specifier.
4902 : */
4903 45645 : if (strchr("sIL", *cp) == NULL)
4904 4 : ereport(ERROR,
4905 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4906 : errmsg("unrecognized format() type specifier \"%.*s\"",
4907 : pg_mblen_range(cp, end_ptr), cp),
4908 : errhint("For a single \"%%\" use \"%%%%\".")));
4909 :
4910 : /* If indirect width was specified, get its value */
4911 45641 : if (widthpos >= 0)
4912 : {
4913 : /* Collect the specified or next argument position */
4914 28 : if (widthpos > 0)
4915 24 : arg = widthpos;
4916 28 : if (arg >= nargs)
4917 0 : ereport(ERROR,
4918 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4919 : errmsg("too few arguments for format()")));
4920 :
4921 : /* Get the value and type of the selected argument */
4922 28 : if (!funcvariadic)
4923 : {
4924 28 : value = PG_GETARG_DATUM(arg);
4925 28 : isNull = PG_ARGISNULL(arg);
4926 28 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4927 : }
4928 : else
4929 : {
4930 0 : value = elements[arg - 1];
4931 0 : isNull = nulls[arg - 1];
4932 0 : typid = element_type;
4933 : }
4934 28 : if (!OidIsValid(typid))
4935 0 : elog(ERROR, "could not determine data type of format() input");
4936 :
4937 28 : arg++;
4938 :
4939 : /* We can treat NULL width the same as zero */
4940 28 : if (isNull)
4941 4 : width = 0;
4942 24 : else if (typid == INT4OID)
4943 24 : width = DatumGetInt32(value);
4944 0 : else if (typid == INT2OID)
4945 0 : width = DatumGetInt16(value);
4946 : else
4947 : {
4948 : /* For less-usual datatypes, convert to text then to int */
4949 : char *str;
4950 :
4951 0 : if (typid != prev_width_type)
4952 : {
4953 : Oid typoutputfunc;
4954 : bool typIsVarlena;
4955 :
4956 0 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
4957 0 : fmgr_info(typoutputfunc, &typoutputinfo_width);
4958 0 : prev_width_type = typid;
4959 : }
4960 :
4961 0 : str = OutputFunctionCall(&typoutputinfo_width, value);
4962 :
4963 : /* pg_strtoint32 will complain about bad data or overflow */
4964 0 : width = pg_strtoint32(str);
4965 :
4966 0 : pfree(str);
4967 : }
4968 : }
4969 :
4970 : /* Collect the specified or next argument position */
4971 45641 : if (argpos > 0)
4972 88 : arg = argpos;
4973 45641 : if (arg >= nargs)
4974 16 : ereport(ERROR,
4975 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4976 : errmsg("too few arguments for format()")));
4977 :
4978 : /* Get the value and type of the selected argument */
4979 45625 : if (!funcvariadic)
4980 : {
4981 44777 : value = PG_GETARG_DATUM(arg);
4982 44777 : isNull = PG_ARGISNULL(arg);
4983 44777 : typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
4984 : }
4985 : else
4986 : {
4987 848 : value = elements[arg - 1];
4988 848 : isNull = nulls[arg - 1];
4989 848 : typid = element_type;
4990 : }
4991 45625 : if (!OidIsValid(typid))
4992 0 : elog(ERROR, "could not determine data type of format() input");
4993 :
4994 45625 : arg++;
4995 :
4996 : /*
4997 : * Get the appropriate typOutput function, reusing previous one if
4998 : * same type as previous argument. That's particularly useful in the
4999 : * variadic-array case, but often saves work even for ordinary calls.
5000 : */
5001 45625 : if (typid != prev_type)
5002 : {
5003 : Oid typoutputfunc;
5004 : bool typIsVarlena;
5005 :
5006 23502 : getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5007 23502 : fmgr_info(typoutputfunc, &typoutputfinfo);
5008 23502 : prev_type = typid;
5009 : }
5010 :
5011 : /*
5012 : * And now we can format the value.
5013 : */
5014 45625 : switch (*cp)
5015 : {
5016 45625 : case 's':
5017 : case 'I':
5018 : case 'L':
5019 45625 : text_format_string_conversion(&str, *cp, &typoutputfinfo,
5020 : value, isNull,
5021 : flags, width);
5022 45621 : break;
5023 0 : default:
5024 : /* should not get here, because of previous check */
5025 0 : ereport(ERROR,
5026 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5027 : errmsg("unrecognized format() type specifier \"%.*s\"",
5028 : pg_mblen_range(cp, end_ptr), cp),
5029 : errhint("For a single \"%%\" use \"%%%%\".")));
5030 : break;
5031 : }
5032 : }
5033 :
5034 : /* Don't need deconstruct_array results anymore. */
5035 22112 : if (elements != NULL)
5036 28 : pfree(elements);
5037 22112 : if (nulls != NULL)
5038 28 : pfree(nulls);
5039 :
5040 : /* Generate results. */
5041 22112 : result = cstring_to_text_with_len(str.data, str.len);
5042 22112 : pfree(str.data);
5043 :
5044 22112 : PG_RETURN_TEXT_P(result);
5045 : }
5046 :
5047 : /*
5048 : * Parse contiguous digits as a decimal number.
5049 : *
5050 : * Returns true if some digits could be parsed.
5051 : * The value is returned into *value, and *ptr is advanced to the next
5052 : * character to be parsed.
5053 : *
5054 : * Note parsing invariant: at least one character is known available before
5055 : * string end (end_ptr) at entry, and this is still true at exit.
5056 : */
5057 : static bool
5058 91298 : text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5059 : {
5060 91298 : bool found = false;
5061 91298 : const char *cp = *ptr;
5062 91298 : int val = 0;
5063 :
5064 91506 : while (*cp >= '0' && *cp <= '9')
5065 : {
5066 212 : int8 digit = (*cp - '0');
5067 :
5068 212 : if (unlikely(pg_mul_s32_overflow(val, 10, &val)) ||
5069 212 : unlikely(pg_add_s32_overflow(val, digit, &val)))
5070 0 : ereport(ERROR,
5071 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5072 : errmsg("number is out of range")));
5073 212 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5074 208 : found = true;
5075 : }
5076 :
5077 91294 : *ptr = cp;
5078 91294 : *value = val;
5079 :
5080 91294 : return found;
5081 : }
5082 :
5083 : /*
5084 : * Parse a format specifier (generally following the SUS printf spec).
5085 : *
5086 : * We have already advanced over the initial '%', and we are looking for
5087 : * [argpos][flags][width]type (but the type character is not consumed here).
5088 : *
5089 : * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5090 : * Output parameters:
5091 : * argpos: argument position for value to be printed. -1 means unspecified.
5092 : * widthpos: argument position for width. Zero means the argument position
5093 : * was unspecified (ie, take the next arg) and -1 means no width
5094 : * argument (width was omitted or specified as a constant).
5095 : * flags: bitmask of flags.
5096 : * width: directly-specified width value. Zero means the width was omitted
5097 : * (note it's not necessary to distinguish this case from an explicit
5098 : * zero width value).
5099 : *
5100 : * The function result is the next character position to be parsed, ie, the
5101 : * location where the type character is/should be.
5102 : *
5103 : * Note parsing invariant: at least one character is known available before
5104 : * string end (end_ptr) at entry, and this is still true at exit.
5105 : */
5106 : static const char *
5107 45661 : text_format_parse_format(const char *start_ptr, const char *end_ptr,
5108 : int *argpos, int *widthpos,
5109 : int *flags, int *width)
5110 : {
5111 45661 : const char *cp = start_ptr;
5112 : int n;
5113 :
5114 : /* set defaults for output parameters */
5115 45661 : *argpos = -1;
5116 45661 : *widthpos = -1;
5117 45661 : *flags = 0;
5118 45661 : *width = 0;
5119 :
5120 : /* try to identify first number */
5121 45661 : if (text_format_parse_digits(&cp, end_ptr, &n))
5122 : {
5123 116 : if (*cp != '$')
5124 : {
5125 : /* Must be just a width and a type, so we're done */
5126 16 : *width = n;
5127 16 : return cp;
5128 : }
5129 : /* The number was argument position */
5130 100 : *argpos = n;
5131 : /* Explicit 0 for argument index is immediately refused */
5132 100 : if (n == 0)
5133 4 : ereport(ERROR,
5134 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5135 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5136 96 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5137 : }
5138 :
5139 : /* Handle flags (only minus is supported now) */
5140 45657 : while (*cp == '-')
5141 : {
5142 20 : *flags |= TEXT_FORMAT_FLAG_MINUS;
5143 20 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5144 : }
5145 :
5146 45637 : if (*cp == '*')
5147 : {
5148 : /* Handle indirect width */
5149 32 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5150 32 : if (text_format_parse_digits(&cp, end_ptr, &n))
5151 : {
5152 : /* number in this position must be closed by $ */
5153 28 : if (*cp != '$')
5154 0 : ereport(ERROR,
5155 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5156 : errmsg("width argument position must be ended by \"$\"")));
5157 : /* The number was width argument position */
5158 28 : *widthpos = n;
5159 : /* Explicit 0 for argument index is immediately refused */
5160 28 : if (n == 0)
5161 4 : ereport(ERROR,
5162 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5163 : errmsg("format specifies argument 0, but arguments are numbered from 1")));
5164 24 : ADVANCE_PARSE_POINTER(cp, end_ptr);
5165 : }
5166 : else
5167 4 : *widthpos = 0; /* width's argument position is unspecified */
5168 : }
5169 : else
5170 : {
5171 : /* Check for direct width specification */
5172 45605 : if (text_format_parse_digits(&cp, end_ptr, &n))
5173 20 : *width = n;
5174 : }
5175 :
5176 : /* cp should now be pointing at type character */
5177 45629 : return cp;
5178 : }
5179 :
5180 : /*
5181 : * Format a %s, %I, or %L conversion
5182 : */
5183 : static void
5184 45625 : text_format_string_conversion(StringInfo buf, char conversion,
5185 : FmgrInfo *typOutputInfo,
5186 : Datum value, bool isNull,
5187 : int flags, int width)
5188 : {
5189 : char *str;
5190 :
5191 : /* Handle NULL arguments before trying to stringify the value. */
5192 45625 : if (isNull)
5193 : {
5194 228 : if (conversion == 's')
5195 180 : text_format_append_string(buf, "", flags, width);
5196 48 : else if (conversion == 'L')
5197 44 : text_format_append_string(buf, "NULL", flags, width);
5198 4 : else if (conversion == 'I')
5199 4 : ereport(ERROR,
5200 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5201 : errmsg("null values cannot be formatted as an SQL identifier")));
5202 224 : return;
5203 : }
5204 :
5205 : /* Stringify. */
5206 45397 : str = OutputFunctionCall(typOutputInfo, value);
5207 :
5208 : /* Escape. */
5209 45397 : if (conversion == 'I')
5210 : {
5211 : /* quote_identifier may or may not allocate a new string. */
5212 3264 : text_format_append_string(buf, quote_identifier(str), flags, width);
5213 : }
5214 42133 : else if (conversion == 'L')
5215 : {
5216 2166 : char *qstr = quote_literal_cstr(str);
5217 :
5218 2166 : text_format_append_string(buf, qstr, flags, width);
5219 : /* quote_literal_cstr() always allocates a new string */
5220 2166 : pfree(qstr);
5221 : }
5222 : else
5223 39967 : text_format_append_string(buf, str, flags, width);
5224 :
5225 : /* Cleanup. */
5226 45397 : pfree(str);
5227 : }
5228 :
5229 : /*
5230 : * Append str to buf, padding as directed by flags/width
5231 : */
5232 : static void
5233 45621 : text_format_append_string(StringInfo buf, const char *str,
5234 : int flags, int width)
5235 : {
5236 45621 : bool align_to_left = false;
5237 : int len;
5238 :
5239 : /* fast path for typical easy case */
5240 45621 : if (width == 0)
5241 : {
5242 45565 : appendStringInfoString(buf, str);
5243 45565 : return;
5244 : }
5245 :
5246 56 : if (width < 0)
5247 : {
5248 : /* Negative width: implicit '-' flag, then take absolute value */
5249 4 : align_to_left = true;
5250 : /* -INT_MIN is undefined */
5251 4 : if (width <= INT_MIN)
5252 0 : ereport(ERROR,
5253 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5254 : errmsg("number is out of range")));
5255 4 : width = -width;
5256 : }
5257 52 : else if (flags & TEXT_FORMAT_FLAG_MINUS)
5258 16 : align_to_left = true;
5259 :
5260 56 : len = pg_mbstrlen(str);
5261 56 : if (align_to_left)
5262 : {
5263 : /* left justify */
5264 20 : appendStringInfoString(buf, str);
5265 20 : if (len < width)
5266 20 : appendStringInfoSpaces(buf, width - len);
5267 : }
5268 : else
5269 : {
5270 : /* right justify */
5271 36 : if (len < width)
5272 36 : appendStringInfoSpaces(buf, width - len);
5273 36 : appendStringInfoString(buf, str);
5274 : }
5275 : }
5276 :
5277 : /*
5278 : * text_format_nv - nonvariadic wrapper for text_format function.
5279 : *
5280 : * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5281 : * which checks that all built-in functions that share the implementing C
5282 : * function take the same number of arguments.
5283 : */
5284 : Datum
5285 1910 : text_format_nv(PG_FUNCTION_ARGS)
5286 : {
5287 1910 : return text_format(fcinfo);
5288 : }
5289 :
5290 : /*
5291 : * Helper function for Levenshtein distance functions. Faster than memcmp(),
5292 : * for this use case.
5293 : */
5294 : static inline bool
5295 0 : rest_of_char_same(const char *s1, const char *s2, int len)
5296 : {
5297 0 : while (len > 0)
5298 : {
5299 0 : len--;
5300 0 : if (s1[len] != s2[len])
5301 0 : return false;
5302 : }
5303 0 : return true;
5304 : }
5305 :
5306 : /* Expand each Levenshtein distance variant */
5307 : #include "levenshtein.c"
5308 : #define LEVENSHTEIN_LESS_EQUAL
5309 : #include "levenshtein.c"
5310 :
5311 :
5312 : /*
5313 : * The following *ClosestMatch() functions can be used to determine whether a
5314 : * user-provided string resembles any known valid values, which is useful for
5315 : * providing hints in log messages, among other things. Use these functions
5316 : * like so:
5317 : *
5318 : * initClosestMatch(&state, source_string, max_distance);
5319 : *
5320 : * for (int i = 0; i < num_valid_strings; i++)
5321 : * updateClosestMatch(&state, valid_strings[i]);
5322 : *
5323 : * closestMatch = getClosestMatch(&state);
5324 : */
5325 :
5326 : /*
5327 : * Initialize the given state with the source string and maximum Levenshtein
5328 : * distance to consider.
5329 : */
5330 : void
5331 44 : initClosestMatch(ClosestMatchState *state, const char *source, int max_d)
5332 : {
5333 : Assert(state);
5334 : Assert(max_d >= 0);
5335 :
5336 44 : state->source = source;
5337 44 : state->min_d = -1;
5338 44 : state->max_d = max_d;
5339 44 : state->match = NULL;
5340 44 : }
5341 :
5342 : /*
5343 : * If the candidate string is a closer match than the current one saved (or
5344 : * there is no match saved), save it as the closest match.
5345 : *
5346 : * If the source or candidate string is NULL, empty, or too long, this function
5347 : * takes no action. Likewise, if the Levenshtein distance exceeds the maximum
5348 : * allowed or more than half the characters are different, no action is taken.
5349 : */
5350 : void
5351 432 : updateClosestMatch(ClosestMatchState *state, const char *candidate)
5352 : {
5353 : int dist;
5354 :
5355 : Assert(state);
5356 :
5357 432 : if (state->source == NULL || state->source[0] == '\0' ||
5358 432 : candidate == NULL || candidate[0] == '\0')
5359 0 : return;
5360 :
5361 : /*
5362 : * To avoid ERROR-ing, we check the lengths here instead of setting
5363 : * 'trusted' to false in the call to varstr_levenshtein_less_equal().
5364 : */
5365 432 : if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
5366 432 : strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
5367 0 : return;
5368 :
5369 432 : dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
5370 432 : candidate, strlen(candidate), 1, 1, 1,
5371 : state->max_d, true);
5372 432 : if (dist <= state->max_d &&
5373 39 : dist <= strlen(state->source) / 2 &&
5374 9 : (state->min_d == -1 || dist < state->min_d))
5375 : {
5376 9 : state->min_d = dist;
5377 9 : state->match = candidate;
5378 : }
5379 : }
5380 :
5381 : /*
5382 : * Return the closest match. If no suitable candidates were provided via
5383 : * updateClosestMatch(), return NULL.
5384 : */
5385 : const char *
5386 44 : getClosestMatch(ClosestMatchState *state)
5387 : {
5388 : Assert(state);
5389 :
5390 44 : return state->match;
5391 : }
5392 :
5393 :
5394 : /*
5395 : * Unicode support
5396 : */
5397 :
5398 : static UnicodeNormalizationForm
5399 149 : unicode_norm_form_from_string(const char *formstr)
5400 : {
5401 149 : UnicodeNormalizationForm form = -1;
5402 :
5403 : /*
5404 : * Might as well check this while we're here.
5405 : */
5406 149 : if (GetDatabaseEncoding() != PG_UTF8)
5407 0 : ereport(ERROR,
5408 : (errcode(ERRCODE_SYNTAX_ERROR),
5409 : errmsg("Unicode normalization can only be performed if server encoding is UTF8")));
5410 :
5411 149 : if (pg_strcasecmp(formstr, "NFC") == 0)
5412 50 : form = UNICODE_NFC;
5413 99 : else if (pg_strcasecmp(formstr, "NFD") == 0)
5414 41 : form = UNICODE_NFD;
5415 58 : else if (pg_strcasecmp(formstr, "NFKC") == 0)
5416 25 : form = UNICODE_NFKC;
5417 33 : else if (pg_strcasecmp(formstr, "NFKD") == 0)
5418 25 : form = UNICODE_NFKD;
5419 : else
5420 8 : ereport(ERROR,
5421 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5422 : errmsg("invalid normalization form: %s", formstr)));
5423 :
5424 141 : return form;
5425 : }
5426 :
5427 : /*
5428 : * Returns version of Unicode used by Postgres in "major.minor" format (the
5429 : * same format as the Unicode version reported by ICU). The third component
5430 : * ("update version") never involves additions to the character repertoire and
5431 : * is unimportant for most purposes.
5432 : *
5433 : * See: https://unicode.org/versions/
5434 : */
5435 : Datum
5436 20 : unicode_version(PG_FUNCTION_ARGS)
5437 : {
5438 20 : PG_RETURN_TEXT_P(cstring_to_text(PG_UNICODE_VERSION));
5439 : }
5440 :
5441 : /*
5442 : * Returns version of Unicode used by ICU, if enabled; otherwise NULL.
5443 : */
5444 : Datum
5445 1 : icu_unicode_version(PG_FUNCTION_ARGS)
5446 : {
5447 1 : const char *version = pg_icu_unicode_version();
5448 :
5449 1 : if (version)
5450 1 : PG_RETURN_TEXT_P(cstring_to_text(version));
5451 : else
5452 0 : PG_RETURN_NULL();
5453 : }
5454 :
5455 : /*
5456 : * Check whether the string contains only assigned Unicode code
5457 : * points. Requires that the database encoding is UTF-8.
5458 : */
5459 : Datum
5460 10 : unicode_assigned(PG_FUNCTION_ARGS)
5461 : {
5462 10 : text *input = PG_GETARG_TEXT_PP(0);
5463 : unsigned char *p;
5464 : int size;
5465 :
5466 10 : if (GetDatabaseEncoding() != PG_UTF8)
5467 0 : ereport(ERROR,
5468 : (errmsg("Unicode categorization can only be performed if server encoding is UTF8")));
5469 :
5470 : /* convert to char32_t */
5471 10 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5472 10 : p = (unsigned char *) VARDATA_ANY(input);
5473 40 : for (int i = 0; i < size; i++)
5474 : {
5475 35 : char32_t uchar = utf8_to_unicode(p);
5476 35 : int category = unicode_category(uchar);
5477 :
5478 35 : if (category == PG_U_UNASSIGNED)
5479 5 : PG_RETURN_BOOL(false);
5480 :
5481 30 : p += pg_utf_mblen(p);
5482 : }
5483 :
5484 5 : PG_RETURN_BOOL(true);
5485 : }
5486 :
5487 : Datum
5488 55 : unicode_normalize_func(PG_FUNCTION_ARGS)
5489 : {
5490 55 : text *input = PG_GETARG_TEXT_PP(0);
5491 55 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5492 : UnicodeNormalizationForm form;
5493 : int size;
5494 : char32_t *input_chars;
5495 : char32_t *output_chars;
5496 : unsigned char *p;
5497 : text *result;
5498 : int i;
5499 :
5500 55 : form = unicode_norm_form_from_string(formstr);
5501 :
5502 : /* convert to char32_t */
5503 51 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5504 51 : input_chars = palloc((size + 1) * sizeof(char32_t));
5505 51 : p = (unsigned char *) VARDATA_ANY(input);
5506 220 : for (i = 0; i < size; i++)
5507 : {
5508 169 : input_chars[i] = utf8_to_unicode(p);
5509 169 : p += pg_utf_mblen(p);
5510 : }
5511 51 : input_chars[i] = (char32_t) '\0';
5512 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5513 :
5514 : /* action */
5515 51 : output_chars = unicode_normalize(form, input_chars);
5516 :
5517 : /* convert back to UTF-8 string */
5518 51 : size = 0;
5519 231 : for (char32_t *wp = output_chars; *wp; wp++)
5520 : {
5521 : unsigned char buf[4];
5522 :
5523 180 : unicode_to_utf8(*wp, buf);
5524 180 : size += pg_utf_mblen(buf);
5525 : }
5526 :
5527 51 : result = palloc(size + VARHDRSZ);
5528 51 : SET_VARSIZE(result, size + VARHDRSZ);
5529 :
5530 51 : p = (unsigned char *) VARDATA_ANY(result);
5531 231 : for (char32_t *wp = output_chars; *wp; wp++)
5532 : {
5533 180 : unicode_to_utf8(*wp, p);
5534 180 : p += pg_utf_mblen(p);
5535 : }
5536 : Assert((char *) p == (char *) result + size + VARHDRSZ);
5537 :
5538 51 : PG_RETURN_TEXT_P(result);
5539 : }
5540 :
5541 : /*
5542 : * Check whether the string is in the specified Unicode normalization form.
5543 : *
5544 : * This is done by converting the string to the specified normal form and then
5545 : * comparing that to the original string. To speed that up, we also apply the
5546 : * "quick check" algorithm specified in UAX #15, which can give a yes or no
5547 : * answer for many strings by just scanning the string once.
5548 : *
5549 : * This function should generally be optimized for the case where the string
5550 : * is in fact normalized. In that case, we'll end up looking at the entire
5551 : * string, so it's probably not worth doing any incremental conversion etc.
5552 : */
5553 : Datum
5554 94 : unicode_is_normalized(PG_FUNCTION_ARGS)
5555 : {
5556 94 : text *input = PG_GETARG_TEXT_PP(0);
5557 94 : char *formstr = text_to_cstring(PG_GETARG_TEXT_PP(1));
5558 : UnicodeNormalizationForm form;
5559 : int size;
5560 : char32_t *input_chars;
5561 : char32_t *output_chars;
5562 : unsigned char *p;
5563 : int i;
5564 : UnicodeNormalizationQC quickcheck;
5565 : int output_size;
5566 : bool result;
5567 :
5568 94 : form = unicode_norm_form_from_string(formstr);
5569 :
5570 : /* convert to char32_t */
5571 90 : size = pg_mbstrlen_with_len(VARDATA_ANY(input), VARSIZE_ANY_EXHDR(input));
5572 90 : input_chars = palloc((size + 1) * sizeof(char32_t));
5573 90 : p = (unsigned char *) VARDATA_ANY(input);
5574 344 : for (i = 0; i < size; i++)
5575 : {
5576 254 : input_chars[i] = utf8_to_unicode(p);
5577 254 : p += pg_utf_mblen(p);
5578 : }
5579 90 : input_chars[i] = (char32_t) '\0';
5580 : Assert((char *) p == VARDATA_ANY(input) + VARSIZE_ANY_EXHDR(input));
5581 :
5582 : /* quick check (see UAX #15) */
5583 90 : quickcheck = unicode_is_normalized_quickcheck(form, input_chars);
5584 90 : if (quickcheck == UNICODE_NORM_QC_YES)
5585 30 : PG_RETURN_BOOL(true);
5586 60 : else if (quickcheck == UNICODE_NORM_QC_NO)
5587 8 : PG_RETURN_BOOL(false);
5588 :
5589 : /* normalize and compare with original */
5590 52 : output_chars = unicode_normalize(form, input_chars);
5591 :
5592 52 : output_size = 0;
5593 216 : for (char32_t *wp = output_chars; *wp; wp++)
5594 164 : output_size++;
5595 :
5596 76 : result = (size == output_size) &&
5597 24 : (memcmp(input_chars, output_chars, size * sizeof(char32_t)) == 0);
5598 :
5599 52 : PG_RETURN_BOOL(result);
5600 : }
5601 :
5602 : /*
5603 : * Check if first n chars are hexadecimal digits
5604 : */
5605 : static bool
5606 111 : isxdigits_n(const char *instr, size_t n)
5607 : {
5608 469 : for (size_t i = 0; i < n; i++)
5609 405 : if (!isxdigit((unsigned char) instr[i]))
5610 47 : return false;
5611 :
5612 64 : return true;
5613 : }
5614 :
5615 : static unsigned int
5616 358 : hexval(unsigned char c)
5617 : {
5618 358 : if (c >= '0' && c <= '9')
5619 278 : return c - '0';
5620 80 : if (c >= 'a' && c <= 'f')
5621 40 : return c - 'a' + 0xA;
5622 40 : if (c >= 'A' && c <= 'F')
5623 40 : return c - 'A' + 0xA;
5624 0 : elog(ERROR, "invalid hexadecimal digit");
5625 : return 0; /* not reached */
5626 : }
5627 :
5628 : /*
5629 : * Translate string with hexadecimal digits to number
5630 : */
5631 : static unsigned int
5632 64 : hexval_n(const char *instr, size_t n)
5633 : {
5634 64 : unsigned int result = 0;
5635 :
5636 422 : for (size_t i = 0; i < n; i++)
5637 358 : result += hexval(instr[i]) << (4 * (n - i - 1));
5638 :
5639 64 : return result;
5640 : }
5641 :
5642 : /*
5643 : * Replaces Unicode escape sequences by Unicode characters
5644 : */
5645 : Datum
5646 47 : unistr(PG_FUNCTION_ARGS)
5647 : {
5648 47 : text *input_text = PG_GETARG_TEXT_PP(0);
5649 : char *instr;
5650 : int len;
5651 : StringInfoData str;
5652 : text *result;
5653 47 : char16_t pair_first = 0;
5654 : char cbuf[MAX_UNICODE_EQUIVALENT_STRING + 1];
5655 :
5656 47 : instr = VARDATA_ANY(input_text);
5657 47 : len = VARSIZE_ANY_EXHDR(input_text);
5658 :
5659 47 : initStringInfo(&str);
5660 :
5661 356 : while (len > 0)
5662 : {
5663 337 : if (instr[0] == '\\')
5664 : {
5665 73 : if (len >= 2 &&
5666 73 : instr[1] == '\\')
5667 : {
5668 5 : if (pair_first)
5669 0 : goto invalid_pair;
5670 5 : appendStringInfoChar(&str, '\\');
5671 5 : instr += 2;
5672 5 : len -= 2;
5673 : }
5674 68 : else if ((len >= 5 && isxdigits_n(instr + 1, 4)) ||
5675 47 : (len >= 6 && instr[1] == 'u' && isxdigits_n(instr + 2, 4)))
5676 22 : {
5677 : char32_t unicode;
5678 30 : int offset = instr[1] == 'u' ? 2 : 1;
5679 :
5680 30 : unicode = hexval_n(instr + offset, 4);
5681 :
5682 30 : if (!is_valid_unicode_codepoint(unicode))
5683 0 : ereport(ERROR,
5684 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5685 : errmsg("invalid Unicode code point: %04X", unicode));
5686 :
5687 30 : if (pair_first)
5688 : {
5689 8 : if (is_utf16_surrogate_second(unicode))
5690 : {
5691 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5692 0 : pair_first = 0;
5693 : }
5694 : else
5695 8 : goto invalid_pair;
5696 : }
5697 22 : else if (is_utf16_surrogate_second(unicode))
5698 0 : goto invalid_pair;
5699 :
5700 22 : if (is_utf16_surrogate_first(unicode))
5701 12 : pair_first = unicode;
5702 : else
5703 : {
5704 10 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5705 10 : appendStringInfoString(&str, cbuf);
5706 : }
5707 :
5708 22 : instr += 4 + offset;
5709 22 : len -= 4 + offset;
5710 : }
5711 38 : else if (len >= 8 && instr[1] == '+' && isxdigits_n(instr + 2, 6))
5712 9 : {
5713 : char32_t unicode;
5714 :
5715 17 : unicode = hexval_n(instr + 2, 6);
5716 :
5717 17 : if (!is_valid_unicode_codepoint(unicode))
5718 4 : ereport(ERROR,
5719 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5720 : errmsg("invalid Unicode code point: %04X", unicode));
5721 :
5722 13 : if (pair_first)
5723 : {
5724 4 : if (is_utf16_surrogate_second(unicode))
5725 : {
5726 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5727 0 : pair_first = 0;
5728 : }
5729 : else
5730 4 : goto invalid_pair;
5731 : }
5732 9 : else if (is_utf16_surrogate_second(unicode))
5733 0 : goto invalid_pair;
5734 :
5735 9 : if (is_utf16_surrogate_first(unicode))
5736 4 : pair_first = unicode;
5737 : else
5738 : {
5739 5 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5740 5 : appendStringInfoString(&str, cbuf);
5741 : }
5742 :
5743 9 : instr += 8;
5744 9 : len -= 8;
5745 : }
5746 21 : else if (len >= 10 && instr[1] == 'U' && isxdigits_n(instr + 2, 8))
5747 9 : {
5748 : char32_t unicode;
5749 :
5750 17 : unicode = hexval_n(instr + 2, 8);
5751 :
5752 17 : if (!is_valid_unicode_codepoint(unicode))
5753 4 : ereport(ERROR,
5754 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5755 : errmsg("invalid Unicode code point: %04X", unicode));
5756 :
5757 13 : if (pair_first)
5758 : {
5759 4 : if (is_utf16_surrogate_second(unicode))
5760 : {
5761 0 : unicode = surrogate_pair_to_codepoint(pair_first, unicode);
5762 0 : pair_first = 0;
5763 : }
5764 : else
5765 4 : goto invalid_pair;
5766 : }
5767 9 : else if (is_utf16_surrogate_second(unicode))
5768 0 : goto invalid_pair;
5769 :
5770 9 : if (is_utf16_surrogate_first(unicode))
5771 4 : pair_first = unicode;
5772 : else
5773 : {
5774 5 : pg_unicode_to_server(unicode, (unsigned char *) cbuf);
5775 5 : appendStringInfoString(&str, cbuf);
5776 : }
5777 :
5778 9 : instr += 10;
5779 9 : len -= 10;
5780 : }
5781 : else
5782 4 : ereport(ERROR,
5783 : (errcode(ERRCODE_SYNTAX_ERROR),
5784 : errmsg("invalid Unicode escape"),
5785 : errhint("Unicode escapes must be \\XXXX, \\+XXXXXX, \\uXXXX, or \\UXXXXXXXX.")));
5786 : }
5787 : else
5788 : {
5789 264 : if (pair_first)
5790 0 : goto invalid_pair;
5791 :
5792 264 : appendStringInfoChar(&str, *instr++);
5793 264 : len--;
5794 : }
5795 : }
5796 :
5797 : /* unfinished surrogate pair? */
5798 19 : if (pair_first)
5799 4 : goto invalid_pair;
5800 :
5801 15 : result = cstring_to_text_with_len(str.data, str.len);
5802 15 : pfree(str.data);
5803 :
5804 15 : PG_RETURN_TEXT_P(result);
5805 :
5806 20 : invalid_pair:
5807 20 : ereport(ERROR,
5808 : (errcode(ERRCODE_SYNTAX_ERROR),
5809 : errmsg("invalid Unicode surrogate pair")));
5810 : PG_RETURN_NULL(); /* keep compiler quiet */
5811 : }
|
