PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
varlena.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * varlena.c
4  * Functions for the variable-length built-in types.
5  *
6  * Portions Copyright (c) 1996-2017, 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/hash.h"
21 #include "access/tuptoaster.h"
22 #include "catalog/pg_collation.h"
23 #include "catalog/pg_type.h"
24 #include "common/md5.h"
25 #include "lib/hyperloglog.h"
26 #include "libpq/pqformat.h"
27 #include "miscadmin.h"
28 #include "parser/scansup.h"
29 #include "port/pg_bswap.h"
30 #include "regex/regex.h"
31 #include "utils/builtins.h"
32 #include "utils/bytea.h"
33 #include "utils/lsyscache.h"
34 #include "utils/memutils.h"
35 #include "utils/pg_locale.h"
36 #include "utils/sortsupport.h"
37 #include "utils/varlena.h"
38 
39 
40 /* GUC variable */
42 
43 typedef struct varlena unknown;
44 typedef struct varlena VarString;
45 
46 typedef struct
47 {
48  bool use_wchar; /* T if multibyte encoding */
49  char *str1; /* use these if not use_wchar */
50  char *str2; /* note: these point to original texts */
51  pg_wchar *wstr1; /* use these if use_wchar */
52  pg_wchar *wstr2; /* note: these are palloc'd */
53  int len1; /* string lengths in logical characters */
54  int len2;
55  /* Skip table for Boyer-Moore-Horspool search algorithm: */
56  int skiptablemask; /* mask for ANDing with skiptable subscripts */
57  int skiptable[256]; /* skip distance for given mismatched char */
59 
60 typedef struct
61 {
62  char *buf1; /* 1st string, or abbreviation original string
63  * buf */
64  char *buf2; /* 2nd string, or abbreviation strxfrm() buf */
65  int buflen1;
66  int buflen2;
67  int last_len1; /* Length of last buf1 string/strxfrm() input */
68  int last_len2; /* Length of last buf2 string/strxfrm() blob */
69  int last_returned; /* Last comparison result (cache) */
70  bool cache_blob; /* Does buf2 contain strxfrm() blob, etc? */
71  bool collate_c;
72  bool bpchar; /* Sorting bpchar, not varchar/text/bytea? */
73  hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
74  hyperLogLogState full_card; /* Full key cardinality state */
75  double prop_card; /* Required cardinality proportion */
78 
79 /*
80  * This should be large enough that most strings will fit, but small enough
81  * that we feel comfortable putting it on the stack
82  */
83 #define TEXTBUFLEN 1024
84 
85 #define DatumGetUnknownP(X) ((unknown *) PG_DETOAST_DATUM(X))
86 #define DatumGetUnknownPCopy(X) ((unknown *) PG_DETOAST_DATUM_COPY(X))
87 #define PG_GETARG_UNKNOWN_P(n) DatumGetUnknownP(PG_GETARG_DATUM(n))
88 #define PG_GETARG_UNKNOWN_P_COPY(n) DatumGetUnknownPCopy(PG_GETARG_DATUM(n))
89 #define PG_RETURN_UNKNOWN_P(x) PG_RETURN_POINTER(x)
90 
91 #define DatumGetVarStringP(X) ((VarString *) PG_DETOAST_DATUM(X))
92 #define DatumGetVarStringPP(X) ((VarString *) PG_DETOAST_DATUM_PACKED(X))
93 
94 static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup);
95 static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup);
96 static int varstrfastcmp_locale(Datum x, Datum y, SortSupport ssup);
97 static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup);
98 static Datum varstr_abbrev_convert(Datum original, SortSupport ssup);
99 static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup);
100 static int32 text_length(Datum str);
101 static text *text_catenate(text *t1, text *t2);
102 static text *text_substring(Datum str,
103  int32 start,
104  int32 length,
105  bool length_not_specified);
106 static text *text_overlay(text *t1, text *t2, int sp, int sl);
107 static int text_position(text *t1, text *t2);
108 static void text_position_setup(text *t1, text *t2, TextPositionState *state);
109 static int text_position_next(int start_pos, TextPositionState *state);
111 static int text_cmp(text *arg1, text *arg2, Oid collid);
112 static bytea *bytea_catenate(bytea *t1, bytea *t2);
113 static bytea *bytea_substring(Datum str,
114  int S,
115  int L,
116  bool length_not_specified);
117 static bytea *bytea_overlay(bytea *t1, bytea *t2, int sp, int sl);
118 static void appendStringInfoText(StringInfo str, const text *t);
121  const char *fldsep, const char *null_string);
123 static bool text_format_parse_digits(const char **ptr, const char *end_ptr,
124  int *value);
125 static const char *text_format_parse_format(const char *start_ptr,
126  const char *end_ptr,
127  int *argpos, int *widthpos,
128  int *flags, int *width);
129 static void text_format_string_conversion(StringInfo buf, char conversion,
130  FmgrInfo *typOutputInfo,
131  Datum value, bool isNull,
132  int flags, int width);
133 static void text_format_append_string(StringInfo buf, const char *str,
134  int flags, int width);
135 
136 
137 /*****************************************************************************
138  * CONVERSION ROUTINES EXPORTED FOR USE BY C CODE *
139  *****************************************************************************/
140 
141 /*
142  * cstring_to_text
143  *
144  * Create a text value from a null-terminated C string.
145  *
146  * The new text value is freshly palloc'd with a full-size VARHDR.
147  */
148 text *
149 cstring_to_text(const char *s)
150 {
151  return cstring_to_text_with_len(s, strlen(s));
152 }
153 
154 /*
155  * cstring_to_text_with_len
156  *
157  * Same as cstring_to_text except the caller specifies the string length;
158  * the string need not be null_terminated.
159  */
160 text *
161 cstring_to_text_with_len(const char *s, int len)
162 {
163  text *result = (text *) palloc(len + VARHDRSZ);
164 
165  SET_VARSIZE(result, len + VARHDRSZ);
166  memcpy(VARDATA(result), s, len);
167 
168  return result;
169 }
170 
171 /*
172  * text_to_cstring
173  *
174  * Create a palloc'd, null-terminated C string from a text value.
175  *
176  * We support being passed a compressed or toasted text value.
177  * This is a bit bogus since such values shouldn't really be referred to as
178  * "text *", but it seems useful for robustness. If we didn't handle that
179  * case here, we'd need another routine that did, anyway.
180  */
181 char *
183 {
184  /* must cast away the const, unfortunately */
185  text *tunpacked = pg_detoast_datum_packed((struct varlena *) t);
186  int len = VARSIZE_ANY_EXHDR(tunpacked);
187  char *result;
188 
189  result = (char *) palloc(len + 1);
190  memcpy(result, VARDATA_ANY(tunpacked), len);
191  result[len] = '\0';
192 
193  if (tunpacked != t)
194  pfree(tunpacked);
195 
196  return result;
197 }
198 
199 /*
200  * text_to_cstring_buffer
201  *
202  * Copy a text value into a caller-supplied buffer of size dst_len.
203  *
204  * The text string is truncated if necessary to fit. The result is
205  * guaranteed null-terminated (unless dst_len == 0).
206  *
207  * We support being passed a compressed or toasted text value.
208  * This is a bit bogus since such values shouldn't really be referred to as
209  * "text *", but it seems useful for robustness. If we didn't handle that
210  * case here, we'd need another routine that did, anyway.
211  */
212 void
213 text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
214 {
215  /* must cast away the const, unfortunately */
216  text *srcunpacked = pg_detoast_datum_packed((struct varlena *) src);
217  size_t src_len = VARSIZE_ANY_EXHDR(srcunpacked);
218 
219  if (dst_len > 0)
220  {
221  dst_len--;
222  if (dst_len >= src_len)
223  dst_len = src_len;
224  else /* ensure truncation is encoding-safe */
225  dst_len = pg_mbcliplen(VARDATA_ANY(srcunpacked), src_len, dst_len);
226  memcpy(dst, VARDATA_ANY(srcunpacked), dst_len);
227  dst[dst_len] = '\0';
228  }
229 
230  if (srcunpacked != src)
231  pfree(srcunpacked);
232 }
233 
234 
235 /*****************************************************************************
236  * USER I/O ROUTINES *
237  *****************************************************************************/
238 
239 
240 #define VAL(CH) ((CH) - '0')
241 #define DIG(VAL) ((VAL) + '0')
242 
243 /*
244  * byteain - converts from printable representation of byte array
245  *
246  * Non-printable characters must be passed as '\nnn' (octal) and are
247  * converted to internal form. '\' must be passed as '\\'.
248  * ereport(ERROR, ...) if bad form.
249  *
250  * BUGS:
251  * The input is scanned twice.
252  * The error checking of input is minimal.
253  */
254 Datum
256 {
257  char *inputText = PG_GETARG_CSTRING(0);
258  char *tp;
259  char *rp;
260  int bc;
261  bytea *result;
262 
263  /* Recognize hex input */
264  if (inputText[0] == '\\' && inputText[1] == 'x')
265  {
266  size_t len = strlen(inputText);
267 
268  bc = (len - 2) / 2 + VARHDRSZ; /* maximum possible length */
269  result = palloc(bc);
270  bc = hex_decode(inputText + 2, len - 2, VARDATA(result));
271  SET_VARSIZE(result, bc + VARHDRSZ); /* actual length */
272 
273  PG_RETURN_BYTEA_P(result);
274  }
275 
276  /* Else, it's the traditional escaped style */
277  for (bc = 0, tp = inputText; *tp != '\0'; bc++)
278  {
279  if (tp[0] != '\\')
280  tp++;
281  else if ((tp[0] == '\\') &&
282  (tp[1] >= '0' && tp[1] <= '3') &&
283  (tp[2] >= '0' && tp[2] <= '7') &&
284  (tp[3] >= '0' && tp[3] <= '7'))
285  tp += 4;
286  else if ((tp[0] == '\\') &&
287  (tp[1] == '\\'))
288  tp += 2;
289  else
290  {
291  /*
292  * one backslash, not followed by another or ### valid octal
293  */
294  ereport(ERROR,
295  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
296  errmsg("invalid input syntax for type %s", "bytea")));
297  }
298  }
299 
300  bc += VARHDRSZ;
301 
302  result = (bytea *) palloc(bc);
303  SET_VARSIZE(result, bc);
304 
305  tp = inputText;
306  rp = VARDATA(result);
307  while (*tp != '\0')
308  {
309  if (tp[0] != '\\')
310  *rp++ = *tp++;
311  else if ((tp[0] == '\\') &&
312  (tp[1] >= '0' && tp[1] <= '3') &&
313  (tp[2] >= '0' && tp[2] <= '7') &&
314  (tp[3] >= '0' && tp[3] <= '7'))
315  {
316  bc = VAL(tp[1]);
317  bc <<= 3;
318  bc += VAL(tp[2]);
319  bc <<= 3;
320  *rp++ = bc + VAL(tp[3]);
321 
322  tp += 4;
323  }
324  else if ((tp[0] == '\\') &&
325  (tp[1] == '\\'))
326  {
327  *rp++ = '\\';
328  tp += 2;
329  }
330  else
331  {
332  /*
333  * We should never get here. The first pass should not allow it.
334  */
335  ereport(ERROR,
336  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
337  errmsg("invalid input syntax for type %s", "bytea")));
338  }
339  }
340 
341  PG_RETURN_BYTEA_P(result);
342 }
343 
344 /*
345  * byteaout - converts to printable representation of byte array
346  *
347  * In the traditional escaped format, non-printable characters are
348  * printed as '\nnn' (octal) and '\' as '\\'.
349  */
350 Datum
352 {
353  bytea *vlena = PG_GETARG_BYTEA_PP(0);
354  char *result;
355  char *rp;
356 
358  {
359  /* Print hex format */
360  rp = result = palloc(VARSIZE_ANY_EXHDR(vlena) * 2 + 2 + 1);
361  *rp++ = '\\';
362  *rp++ = 'x';
363  rp += hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp);
364  }
365  else if (bytea_output == BYTEA_OUTPUT_ESCAPE)
366  {
367  /* Print traditional escaped format */
368  char *vp;
369  int len;
370  int i;
371 
372  len = 1; /* empty string has 1 char */
373  vp = VARDATA_ANY(vlena);
374  for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
375  {
376  if (*vp == '\\')
377  len += 2;
378  else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
379  len += 4;
380  else
381  len++;
382  }
383  rp = result = (char *) palloc(len);
384  vp = VARDATA_ANY(vlena);
385  for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
386  {
387  if (*vp == '\\')
388  {
389  *rp++ = '\\';
390  *rp++ = '\\';
391  }
392  else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
393  {
394  int val; /* holds unprintable chars */
395 
396  val = *vp;
397  rp[0] = '\\';
398  rp[3] = DIG(val & 07);
399  val >>= 3;
400  rp[2] = DIG(val & 07);
401  val >>= 3;
402  rp[1] = DIG(val & 03);
403  rp += 4;
404  }
405  else
406  *rp++ = *vp;
407  }
408  }
409  else
410  {
411  elog(ERROR, "unrecognized bytea_output setting: %d",
412  bytea_output);
413  rp = result = NULL; /* keep compiler quiet */
414  }
415  *rp = '\0';
416  PG_RETURN_CSTRING(result);
417 }
418 
419 /*
420  * bytearecv - converts external binary format to bytea
421  */
422 Datum
424 {
426  bytea *result;
427  int nbytes;
428 
429  nbytes = buf->len - buf->cursor;
430  result = (bytea *) palloc(nbytes + VARHDRSZ);
431  SET_VARSIZE(result, nbytes + VARHDRSZ);
432  pq_copymsgbytes(buf, VARDATA(result), nbytes);
433  PG_RETURN_BYTEA_P(result);
434 }
435 
436 /*
437  * byteasend - converts bytea to binary format
438  *
439  * This is a special case: just copy the input...
440  */
441 Datum
443 {
444  bytea *vlena = PG_GETARG_BYTEA_P_COPY(0);
445 
446  PG_RETURN_BYTEA_P(vlena);
447 }
448 
449 Datum
451 {
453 
454  state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
455 
456  /* Append the value unless null. */
457  if (!PG_ARGISNULL(1))
458  {
460 
461  /* On the first time through, we ignore the delimiter. */
462  if (state == NULL)
463  state = makeStringAggState(fcinfo);
464  else if (!PG_ARGISNULL(2))
465  {
466  bytea *delim = PG_GETARG_BYTEA_PP(2);
467 
469  }
470 
472  }
473 
474  /*
475  * The transition type for string_agg() is declared to be "internal",
476  * which is a pass-by-value type the same size as a pointer.
477  */
478  PG_RETURN_POINTER(state);
479 }
480 
481 Datum
483 {
485 
486  /* cannot be called directly because of internal-type argument */
487  Assert(AggCheckCallContext(fcinfo, NULL));
488 
489  state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
490 
491  if (state != NULL)
492  {
493  bytea *result;
494 
495  result = (bytea *) palloc(state->len + VARHDRSZ);
496  SET_VARSIZE(result, state->len + VARHDRSZ);
497  memcpy(VARDATA(result), state->data, state->len);
498  PG_RETURN_BYTEA_P(result);
499  }
500  else
501  PG_RETURN_NULL();
502 }
503 
504 /*
505  * textin - converts "..." to internal representation
506  */
507 Datum
509 {
510  char *inputText = PG_GETARG_CSTRING(0);
511 
512  PG_RETURN_TEXT_P(cstring_to_text(inputText));
513 }
514 
515 /*
516  * textout - converts internal representation to "..."
517  */
518 Datum
520 {
521  Datum txt = PG_GETARG_DATUM(0);
522 
524 }
525 
526 /*
527  * textrecv - converts external binary format to text
528  */
529 Datum
531 {
533  text *result;
534  char *str;
535  int nbytes;
536 
537  str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
538 
539  result = cstring_to_text_with_len(str, nbytes);
540  pfree(str);
541  PG_RETURN_TEXT_P(result);
542 }
543 
544 /*
545  * textsend - converts text to binary format
546  */
547 Datum
549 {
550  text *t = PG_GETARG_TEXT_PP(0);
552 
553  pq_begintypsend(&buf);
556 }
557 
558 
559 /*
560  * unknownin - converts "..." to internal representation
561  */
562 Datum
564 {
565  char *str = PG_GETARG_CSTRING(0);
566 
567  /* representation is same as cstring */
569 }
570 
571 /*
572  * unknownout - converts internal representation to "..."
573  */
574 Datum
576 {
577  /* representation is same as cstring */
578  char *str = PG_GETARG_CSTRING(0);
579 
581 }
582 
583 /*
584  * unknownrecv - converts external binary format to unknown
585  */
586 Datum
588 {
590  char *str;
591  int nbytes;
592 
593  str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
594  /* representation is same as cstring */
595  PG_RETURN_CSTRING(str);
596 }
597 
598 /*
599  * unknownsend - converts unknown to binary format
600  */
601 Datum
603 {
604  /* representation is same as cstring */
605  char *str = PG_GETARG_CSTRING(0);
607 
608  pq_begintypsend(&buf);
609  pq_sendtext(&buf, str, strlen(str));
611 }
612 
613 
614 /* ========== PUBLIC ROUTINES ========== */
615 
616 /*
617  * textlen -
618  * returns the logical length of a text*
619  * (which is less than the VARSIZE of the text*)
620  */
621 Datum
623 {
624  Datum str = PG_GETARG_DATUM(0);
625 
626  /* try to avoid decompressing argument */
628 }
629 
630 /*
631  * text_length -
632  * Does the real work for textlen()
633  *
634  * This is broken out so it can be called directly by other string processing
635  * functions. Note that the argument is passed as a Datum, to indicate that
636  * it may still be in compressed form. We can avoid decompressing it at all
637  * in some cases.
638  */
639 static int32
641 {
642  /* fastpath when max encoding length is one */
645  else
646  {
647  text *t = DatumGetTextPP(str);
648 
650  VARSIZE_ANY_EXHDR(t)));
651  }
652 }
653 
654 /*
655  * textoctetlen -
656  * returns the physical length of a text*
657  * (which is less than the VARSIZE of the text*)
658  */
659 Datum
661 {
662  Datum str = PG_GETARG_DATUM(0);
663 
664  /* We need not detoast the input at all */
666 }
667 
668 /*
669  * textcat -
670  * takes two text* and returns a text* that is the concatenation of
671  * the two.
672  *
673  * Rewritten by Sapa, sapa@hq.icb.chel.su. 8-Jul-96.
674  * Updated by Thomas, Thomas.Lockhart@jpl.nasa.gov 1997-07-10.
675  * Allocate space for output in all cases.
676  * XXX - thomas 1997-07-10
677  */
678 Datum
680 {
681  text *t1 = PG_GETARG_TEXT_PP(0);
682  text *t2 = PG_GETARG_TEXT_PP(1);
683 
685 }
686 
687 /*
688  * text_catenate
689  * Guts of textcat(), broken out so it can be used by other functions
690  *
691  * Arguments can be in short-header form, but not compressed or out-of-line
692  */
693 static text *
695 {
696  text *result;
697  int len1,
698  len2,
699  len;
700  char *ptr;
701 
702  len1 = VARSIZE_ANY_EXHDR(t1);
703  len2 = VARSIZE_ANY_EXHDR(t2);
704 
705  /* paranoia ... probably should throw error instead? */
706  if (len1 < 0)
707  len1 = 0;
708  if (len2 < 0)
709  len2 = 0;
710 
711  len = len1 + len2 + VARHDRSZ;
712  result = (text *) palloc(len);
713 
714  /* Set size of result string... */
715  SET_VARSIZE(result, len);
716 
717  /* Fill data field of result string... */
718  ptr = VARDATA(result);
719  if (len1 > 0)
720  memcpy(ptr, VARDATA_ANY(t1), len1);
721  if (len2 > 0)
722  memcpy(ptr + len1, VARDATA_ANY(t2), len2);
723 
724  return result;
725 }
726 
727 /*
728  * charlen_to_bytelen()
729  * Compute the number of bytes occupied by n characters starting at *p
730  *
731  * It is caller's responsibility that there actually are n characters;
732  * the string need not be null-terminated.
733  */
734 static int
735 charlen_to_bytelen(const char *p, int n)
736 {
738  {
739  /* Optimization for single-byte encodings */
740  return n;
741  }
742  else
743  {
744  const char *s;
745 
746  for (s = p; n > 0; n--)
747  s += pg_mblen(s);
748 
749  return s - p;
750  }
751 }
752 
753 /*
754  * text_substr()
755  * Return a substring starting at the specified position.
756  * - thomas 1997-12-31
757  *
758  * Input:
759  * - string
760  * - starting position (is one-based)
761  * - string length
762  *
763  * If the starting position is zero or less, then return from the start of the string
764  * adjusting the length to be consistent with the "negative start" per SQL.
765  * If the length is less than zero, return the remaining string.
766  *
767  * Added multibyte support.
768  * - Tatsuo Ishii 1998-4-21
769  * Changed behavior if starting position is less than one to conform to SQL behavior.
770  * Formerly returned the entire string; now returns a portion.
771  * - Thomas Lockhart 1998-12-10
772  * Now uses faster TOAST-slicing interface
773  * - John Gray 2002-02-22
774  * Remove "#ifdef MULTIBYTE" and test for encoding_max_length instead. Change
775  * behaviors conflicting with SQL to meet SQL (if E = S + L < S throw
776  * error; if E < 1, return '', not entire string). Fixed MB related bug when
777  * S > LC and < LC + 4 sometimes garbage characters are returned.
778  * - Joe Conway 2002-08-10
779  */
780 Datum
782 {
784  PG_GETARG_INT32(1),
785  PG_GETARG_INT32(2),
786  false));
787 }
788 
789 /*
790  * text_substr_no_len -
791  * Wrapper to avoid opr_sanity failure due to
792  * one function accepting a different number of args.
793  */
794 Datum
796 {
798  PG_GETARG_INT32(1),
799  -1, true));
800 }
801 
802 /*
803  * text_substring -
804  * Does the real work for text_substr() and text_substr_no_len()
805  *
806  * This is broken out so it can be called directly by other string processing
807  * functions. Note that the argument is passed as a Datum, to indicate that
808  * it may still be in compressed/toasted form. We can avoid detoasting all
809  * of it in some cases.
810  *
811  * The result is always a freshly palloc'd datum.
812  */
813 static text *
814 text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
815 {
817  int32 S = start; /* start position */
818  int32 S1; /* adjusted start position */
819  int32 L1; /* adjusted substring length */
820 
821  /* life is easy if the encoding max length is 1 */
822  if (eml == 1)
823  {
824  S1 = Max(S, 1);
825 
826  if (length_not_specified) /* special case - get length to end of
827  * string */
828  L1 = -1;
829  else
830  {
831  /* end position */
832  int E = S + length;
833 
834  /*
835  * A negative value for L is the only way for the end position to
836  * be before the start. SQL99 says to throw an error.
837  */
838  if (E < S)
839  ereport(ERROR,
840  (errcode(ERRCODE_SUBSTRING_ERROR),
841  errmsg("negative substring length not allowed")));
842 
843  /*
844  * A zero or negative value for the end position can happen if the
845  * start was negative or one. SQL99 says to return a zero-length
846  * string.
847  */
848  if (E < 1)
849  return cstring_to_text("");
850 
851  L1 = E - S1;
852  }
853 
854  /*
855  * If the start position is past the end of the string, SQL99 says to
856  * return a zero-length string -- PG_GETARG_TEXT_P_SLICE() will do
857  * that for us. Convert to zero-based starting position
858  */
859  return DatumGetTextPSlice(str, S1 - 1, L1);
860  }
861  else if (eml > 1)
862  {
863  /*
864  * When encoding max length is > 1, we can't get LC without
865  * detoasting, so we'll grab a conservatively large slice now and go
866  * back later to do the right thing
867  */
868  int32 slice_start;
869  int32 slice_size;
870  int32 slice_strlen;
871  text *slice;
872  int32 E1;
873  int32 i;
874  char *p;
875  char *s;
876  text *ret;
877 
878  /*
879  * if S is past the end of the string, the tuple toaster will return a
880  * zero-length string to us
881  */
882  S1 = Max(S, 1);
883 
884  /*
885  * We need to start at position zero because there is no way to know
886  * in advance which byte offset corresponds to the supplied start
887  * position.
888  */
889  slice_start = 0;
890 
891  if (length_not_specified) /* special case - get length to end of
892  * string */
893  slice_size = L1 = -1;
894  else
895  {
896  int E = S + length;
897 
898  /*
899  * A negative value for L is the only way for the end position to
900  * be before the start. SQL99 says to throw an error.
901  */
902  if (E < S)
903  ereport(ERROR,
904  (errcode(ERRCODE_SUBSTRING_ERROR),
905  errmsg("negative substring length not allowed")));
906 
907  /*
908  * A zero or negative value for the end position can happen if the
909  * start was negative or one. SQL99 says to return a zero-length
910  * string.
911  */
912  if (E < 1)
913  return cstring_to_text("");
914 
915  /*
916  * if E is past the end of the string, the tuple toaster will
917  * truncate the length for us
918  */
919  L1 = E - S1;
920 
921  /*
922  * Total slice size in bytes can't be any longer than the start
923  * position plus substring length times the encoding max length.
924  */
925  slice_size = (S1 + L1) * eml;
926  }
927 
928  /*
929  * If we're working with an untoasted source, no need to do an extra
930  * copying step.
931  */
934  slice = DatumGetTextPSlice(str, slice_start, slice_size);
935  else
936  slice = (text *) DatumGetPointer(str);
937 
938  /* see if we got back an empty string */
939  if (VARSIZE_ANY_EXHDR(slice) == 0)
940  {
941  if (slice != (text *) DatumGetPointer(str))
942  pfree(slice);
943  return cstring_to_text("");
944  }
945 
946  /* Now we can get the actual length of the slice in MB characters */
947  slice_strlen = pg_mbstrlen_with_len(VARDATA_ANY(slice),
948  VARSIZE_ANY_EXHDR(slice));
949 
950  /*
951  * Check that the start position wasn't > slice_strlen. If so, SQL99
952  * says to return a zero-length string.
953  */
954  if (S1 > slice_strlen)
955  {
956  if (slice != (text *) DatumGetPointer(str))
957  pfree(slice);
958  return cstring_to_text("");
959  }
960 
961  /*
962  * Adjust L1 and E1 now that we know the slice string length. Again
963  * remember that S1 is one based, and slice_start is zero based.
964  */
965  if (L1 > -1)
966  E1 = Min(S1 + L1, slice_start + 1 + slice_strlen);
967  else
968  E1 = slice_start + 1 + slice_strlen;
969 
970  /*
971  * Find the start position in the slice; remember S1 is not zero based
972  */
973  p = VARDATA_ANY(slice);
974  for (i = 0; i < S1 - 1; i++)
975  p += pg_mblen(p);
976 
977  /* hang onto a pointer to our start position */
978  s = p;
979 
980  /*
981  * Count the actual bytes used by the substring of the requested
982  * length.
983  */
984  for (i = S1; i < E1; i++)
985  p += pg_mblen(p);
986 
987  ret = (text *) palloc(VARHDRSZ + (p - s));
988  SET_VARSIZE(ret, VARHDRSZ + (p - s));
989  memcpy(VARDATA(ret), s, (p - s));
990 
991  if (slice != (text *) DatumGetPointer(str))
992  pfree(slice);
993 
994  return ret;
995  }
996  else
997  elog(ERROR, "invalid backend encoding: encoding max length < 1");
998 
999  /* not reached: suppress compiler warning */
1000  return NULL;
1001 }
1002 
1003 /*
1004  * textoverlay
1005  * Replace specified substring of first string with second
1006  *
1007  * The SQL standard defines OVERLAY() in terms of substring and concatenation.
1008  * This code is a direct implementation of what the standard says.
1009  */
1010 Datum
1012 {
1013  text *t1 = PG_GETARG_TEXT_PP(0);
1014  text *t2 = PG_GETARG_TEXT_PP(1);
1015  int sp = PG_GETARG_INT32(2); /* substring start position */
1016  int sl = PG_GETARG_INT32(3); /* substring length */
1017 
1018  PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1019 }
1020 
1021 Datum
1023 {
1024  text *t1 = PG_GETARG_TEXT_PP(0);
1025  text *t2 = PG_GETARG_TEXT_PP(1);
1026  int sp = PG_GETARG_INT32(2); /* substring start position */
1027  int sl;
1028 
1029  sl = text_length(PointerGetDatum(t2)); /* defaults to length(t2) */
1030  PG_RETURN_TEXT_P(text_overlay(t1, t2, sp, sl));
1031 }
1032 
1033 static text *
1034 text_overlay(text *t1, text *t2, int sp, int sl)
1035 {
1036  text *result;
1037  text *s1;
1038  text *s2;
1039  int sp_pl_sl;
1040 
1041  /*
1042  * Check for possible integer-overflow cases. For negative sp, throw a
1043  * "substring length" error because that's what should be expected
1044  * according to the spec's definition of OVERLAY().
1045  */
1046  if (sp <= 0)
1047  ereport(ERROR,
1048  (errcode(ERRCODE_SUBSTRING_ERROR),
1049  errmsg("negative substring length not allowed")));
1050  sp_pl_sl = sp + sl;
1051  if (sp_pl_sl <= sl)
1052  ereport(ERROR,
1053  (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
1054  errmsg("integer out of range")));
1055 
1056  s1 = text_substring(PointerGetDatum(t1), 1, sp - 1, false);
1057  s2 = text_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
1058  result = text_catenate(s1, t2);
1059  result = text_catenate(result, s2);
1060 
1061  return result;
1062 }
1063 
1064 /*
1065  * textpos -
1066  * Return the position of the specified substring.
1067  * Implements the SQL POSITION() function.
1068  * Ref: A Guide To The SQL Standard, Date & Darwen, 1997
1069  * - thomas 1997-07-27
1070  */
1071 Datum
1073 {
1074  text *str = PG_GETARG_TEXT_PP(0);
1075  text *search_str = PG_GETARG_TEXT_PP(1);
1076 
1077  PG_RETURN_INT32((int32) text_position(str, search_str));
1078 }
1079 
1080 /*
1081  * text_position -
1082  * Does the real work for textpos()
1083  *
1084  * Inputs:
1085  * t1 - string to be searched
1086  * t2 - pattern to match within t1
1087  * Result:
1088  * Character index of the first matched char, starting from 1,
1089  * or 0 if no match.
1090  *
1091  * This is broken out so it can be called directly by other string processing
1092  * functions.
1093  */
1094 static int
1096 {
1098  int result;
1099 
1100  text_position_setup(t1, t2, &state);
1101  result = text_position_next(1, &state);
1102  text_position_cleanup(&state);
1103  return result;
1104 }
1105 
1106 
1107 /*
1108  * text_position_setup, text_position_next, text_position_cleanup -
1109  * Component steps of text_position()
1110  *
1111  * These are broken out so that a string can be efficiently searched for
1112  * multiple occurrences of the same pattern. text_position_next may be
1113  * called multiple times with increasing values of start_pos, which is
1114  * the 1-based character position to start the search from. The "state"
1115  * variable is normally just a local variable in the caller.
1116  */
1117 
1118 static void
1120 {
1121  int len1 = VARSIZE_ANY_EXHDR(t1);
1122  int len2 = VARSIZE_ANY_EXHDR(t2);
1123 
1125  {
1126  /* simple case - single byte encoding */
1127  state->use_wchar = false;
1128  state->str1 = VARDATA_ANY(t1);
1129  state->str2 = VARDATA_ANY(t2);
1130  state->len1 = len1;
1131  state->len2 = len2;
1132  }
1133  else
1134  {
1135  /* not as simple - multibyte encoding */
1136  pg_wchar *p1,
1137  *p2;
1138 
1139  p1 = (pg_wchar *) palloc((len1 + 1) * sizeof(pg_wchar));
1140  len1 = pg_mb2wchar_with_len(VARDATA_ANY(t1), p1, len1);
1141  p2 = (pg_wchar *) palloc((len2 + 1) * sizeof(pg_wchar));
1142  len2 = pg_mb2wchar_with_len(VARDATA_ANY(t2), p2, len2);
1143 
1144  state->use_wchar = true;
1145  state->wstr1 = p1;
1146  state->wstr2 = p2;
1147  state->len1 = len1;
1148  state->len2 = len2;
1149  }
1150 
1151  /*
1152  * Prepare the skip table for Boyer-Moore-Horspool searching. In these
1153  * notes we use the terminology that the "haystack" is the string to be
1154  * searched (t1) and the "needle" is the pattern being sought (t2).
1155  *
1156  * If the needle is empty or bigger than the haystack then there is no
1157  * point in wasting cycles initializing the table. We also choose not to
1158  * use B-M-H for needles of length 1, since the skip table can't possibly
1159  * save anything in that case.
1160  */
1161  if (len1 >= len2 && len2 > 1)
1162  {
1163  int searchlength = len1 - len2;
1164  int skiptablemask;
1165  int last;
1166  int i;
1167 
1168  /*
1169  * First we must determine how much of the skip table to use. The
1170  * declaration of TextPositionState allows up to 256 elements, but for
1171  * short search problems we don't really want to have to initialize so
1172  * many elements --- it would take too long in comparison to the
1173  * actual search time. So we choose a useful skip table size based on
1174  * the haystack length minus the needle length. The closer the needle
1175  * length is to the haystack length the less useful skipping becomes.
1176  *
1177  * Note: since we use bit-masking to select table elements, the skip
1178  * table size MUST be a power of 2, and so the mask must be 2^N-1.
1179  */
1180  if (searchlength < 16)
1181  skiptablemask = 3;
1182  else if (searchlength < 64)
1183  skiptablemask = 7;
1184  else if (searchlength < 128)
1185  skiptablemask = 15;
1186  else if (searchlength < 512)
1187  skiptablemask = 31;
1188  else if (searchlength < 2048)
1189  skiptablemask = 63;
1190  else if (searchlength < 4096)
1191  skiptablemask = 127;
1192  else
1193  skiptablemask = 255;
1194  state->skiptablemask = skiptablemask;
1195 
1196  /*
1197  * Initialize the skip table. We set all elements to the needle
1198  * length, since this is the correct skip distance for any character
1199  * not found in the needle.
1200  */
1201  for (i = 0; i <= skiptablemask; i++)
1202  state->skiptable[i] = len2;
1203 
1204  /*
1205  * Now examine the needle. For each character except the last one,
1206  * set the corresponding table element to the appropriate skip
1207  * distance. Note that when two characters share the same skip table
1208  * entry, the one later in the needle must determine the skip
1209  * distance.
1210  */
1211  last = len2 - 1;
1212 
1213  if (!state->use_wchar)
1214  {
1215  const char *str2 = state->str2;
1216 
1217  for (i = 0; i < last; i++)
1218  state->skiptable[(unsigned char) str2[i] & skiptablemask] = last - i;
1219  }
1220  else
1221  {
1222  const pg_wchar *wstr2 = state->wstr2;
1223 
1224  for (i = 0; i < last; i++)
1225  state->skiptable[wstr2[i] & skiptablemask] = last - i;
1226  }
1227  }
1228 }
1229 
1230 static int
1232 {
1233  int haystack_len = state->len1;
1234  int needle_len = state->len2;
1235  int skiptablemask = state->skiptablemask;
1236 
1237  Assert(start_pos > 0); /* else caller error */
1238 
1239  if (needle_len <= 0)
1240  return start_pos; /* result for empty pattern */
1241 
1242  start_pos--; /* adjust for zero based arrays */
1243 
1244  /* Done if the needle can't possibly fit */
1245  if (haystack_len < start_pos + needle_len)
1246  return 0;
1247 
1248  if (!state->use_wchar)
1249  {
1250  /* simple case - single byte encoding */
1251  const char *haystack = state->str1;
1252  const char *needle = state->str2;
1253  const char *haystack_end = &haystack[haystack_len];
1254  const char *hptr;
1255 
1256  if (needle_len == 1)
1257  {
1258  /* No point in using B-M-H for a one-character needle */
1259  char nchar = *needle;
1260 
1261  hptr = &haystack[start_pos];
1262  while (hptr < haystack_end)
1263  {
1264  if (*hptr == nchar)
1265  return hptr - haystack + 1;
1266  hptr++;
1267  }
1268  }
1269  else
1270  {
1271  const char *needle_last = &needle[needle_len - 1];
1272 
1273  /* Start at startpos plus the length of the needle */
1274  hptr = &haystack[start_pos + needle_len - 1];
1275  while (hptr < haystack_end)
1276  {
1277  /* Match the needle scanning *backward* */
1278  const char *nptr;
1279  const char *p;
1280 
1281  nptr = needle_last;
1282  p = hptr;
1283  while (*nptr == *p)
1284  {
1285  /* Matched it all? If so, return 1-based position */
1286  if (nptr == needle)
1287  return p - haystack + 1;
1288  nptr--, p--;
1289  }
1290 
1291  /*
1292  * No match, so use the haystack char at hptr to decide how
1293  * far to advance. If the needle had any occurrence of that
1294  * character (or more precisely, one sharing the same
1295  * skiptable entry) before its last character, then we advance
1296  * far enough to align the last such needle character with
1297  * that haystack position. Otherwise we can advance by the
1298  * whole needle length.
1299  */
1300  hptr += state->skiptable[(unsigned char) *hptr & skiptablemask];
1301  }
1302  }
1303  }
1304  else
1305  {
1306  /* The multibyte char version. This works exactly the same way. */
1307  const pg_wchar *haystack = state->wstr1;
1308  const pg_wchar *needle = state->wstr2;
1309  const pg_wchar *haystack_end = &haystack[haystack_len];
1310  const pg_wchar *hptr;
1311 
1312  if (needle_len == 1)
1313  {
1314  /* No point in using B-M-H for a one-character needle */
1315  pg_wchar nchar = *needle;
1316 
1317  hptr = &haystack[start_pos];
1318  while (hptr < haystack_end)
1319  {
1320  if (*hptr == nchar)
1321  return hptr - haystack + 1;
1322  hptr++;
1323  }
1324  }
1325  else
1326  {
1327  const pg_wchar *needle_last = &needle[needle_len - 1];
1328 
1329  /* Start at startpos plus the length of the needle */
1330  hptr = &haystack[start_pos + needle_len - 1];
1331  while (hptr < haystack_end)
1332  {
1333  /* Match the needle scanning *backward* */
1334  const pg_wchar *nptr;
1335  const pg_wchar *p;
1336 
1337  nptr = needle_last;
1338  p = hptr;
1339  while (*nptr == *p)
1340  {
1341  /* Matched it all? If so, return 1-based position */
1342  if (nptr == needle)
1343  return p - haystack + 1;
1344  nptr--, p--;
1345  }
1346 
1347  /*
1348  * No match, so use the haystack char at hptr to decide how
1349  * far to advance. If the needle had any occurrence of that
1350  * character (or more precisely, one sharing the same
1351  * skiptable entry) before its last character, then we advance
1352  * far enough to align the last such needle character with
1353  * that haystack position. Otherwise we can advance by the
1354  * whole needle length.
1355  */
1356  hptr += state->skiptable[*hptr & skiptablemask];
1357  }
1358  }
1359  }
1360 
1361  return 0; /* not found */
1362 }
1363 
1364 static void
1366 {
1367  if (state->use_wchar)
1368  {
1369  pfree(state->wstr1);
1370  pfree(state->wstr2);
1371  }
1372 }
1373 
1374 /* varstr_cmp()
1375  * Comparison function for text strings with given lengths.
1376  * Includes locale support, but must copy strings to temporary memory
1377  * to allow null-termination for inputs to strcoll().
1378  * Returns an integer less than, equal to, or greater than zero, indicating
1379  * whether arg1 is less than, equal to, or greater than arg2.
1380  */
1381 int
1382 varstr_cmp(char *arg1, int len1, char *arg2, int len2, Oid collid)
1383 {
1384  int result;
1385 
1386  /*
1387  * Unfortunately, there is no strncoll(), so in the non-C locale case we
1388  * have to do some memory copying. This turns out to be significantly
1389  * slower, so we optimize the case where LC_COLLATE is C. We also try to
1390  * optimize relatively-short strings by avoiding palloc/pfree overhead.
1391  */
1392  if (lc_collate_is_c(collid))
1393  {
1394  result = memcmp(arg1, arg2, Min(len1, len2));
1395  if ((result == 0) && (len1 != len2))
1396  result = (len1 < len2) ? -1 : 1;
1397  }
1398  else
1399  {
1400  char a1buf[TEXTBUFLEN];
1401  char a2buf[TEXTBUFLEN];
1402  char *a1p,
1403  *a2p;
1404  pg_locale_t mylocale = 0;
1405 
1406  if (collid != DEFAULT_COLLATION_OID)
1407  {
1408  if (!OidIsValid(collid))
1409  {
1410  /*
1411  * This typically means that the parser could not resolve a
1412  * conflict of implicit collations, so report it that way.
1413  */
1414  ereport(ERROR,
1415  (errcode(ERRCODE_INDETERMINATE_COLLATION),
1416  errmsg("could not determine which collation to use for string comparison"),
1417  errhint("Use the COLLATE clause to set the collation explicitly.")));
1418  }
1419  mylocale = pg_newlocale_from_collation(collid);
1420  }
1421 
1422  /*
1423  * memcmp() can't tell us which of two unequal strings sorts first,
1424  * but it's a cheap way to tell if they're equal. Testing shows that
1425  * memcmp() followed by strcoll() is only trivially slower than
1426  * strcoll() by itself, so we don't lose much if this doesn't work out
1427  * very often, and if it does - for example, because there are many
1428  * equal strings in the input - then we win big by avoiding expensive
1429  * collation-aware comparisons.
1430  */
1431  if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1432  return 0;
1433 
1434 #ifdef WIN32
1435  /* Win32 does not have UTF-8, so we need to map to UTF-16 */
1436  if (GetDatabaseEncoding() == PG_UTF8)
1437  {
1438  int a1len;
1439  int a2len;
1440  int r;
1441 
1442  if (len1 >= TEXTBUFLEN / 2)
1443  {
1444  a1len = len1 * 2 + 2;
1445  a1p = palloc(a1len);
1446  }
1447  else
1448  {
1449  a1len = TEXTBUFLEN;
1450  a1p = a1buf;
1451  }
1452  if (len2 >= TEXTBUFLEN / 2)
1453  {
1454  a2len = len2 * 2 + 2;
1455  a2p = palloc(a2len);
1456  }
1457  else
1458  {
1459  a2len = TEXTBUFLEN;
1460  a2p = a2buf;
1461  }
1462 
1463  /* stupid Microsloth API does not work for zero-length input */
1464  if (len1 == 0)
1465  r = 0;
1466  else
1467  {
1468  r = MultiByteToWideChar(CP_UTF8, 0, arg1, len1,
1469  (LPWSTR) a1p, a1len / 2);
1470  if (!r)
1471  ereport(ERROR,
1472  (errmsg("could not convert string to UTF-16: error code %lu",
1473  GetLastError())));
1474  }
1475  ((LPWSTR) a1p)[r] = 0;
1476 
1477  if (len2 == 0)
1478  r = 0;
1479  else
1480  {
1481  r = MultiByteToWideChar(CP_UTF8, 0, arg2, len2,
1482  (LPWSTR) a2p, a2len / 2);
1483  if (!r)
1484  ereport(ERROR,
1485  (errmsg("could not convert string to UTF-16: error code %lu",
1486  GetLastError())));
1487  }
1488  ((LPWSTR) a2p)[r] = 0;
1489 
1490  errno = 0;
1491 #ifdef HAVE_LOCALE_T
1492  if (mylocale)
1493  result = wcscoll_l((LPWSTR) a1p, (LPWSTR) a2p, mylocale->info.lt);
1494  else
1495 #endif
1496  result = wcscoll((LPWSTR) a1p, (LPWSTR) a2p);
1497  if (result == 2147483647) /* _NLSCMPERROR; missing from mingw
1498  * headers */
1499  ereport(ERROR,
1500  (errmsg("could not compare Unicode strings: %m")));
1501 
1502  /*
1503  * In some locales wcscoll() can claim that nonidentical strings
1504  * are equal. Believing that would be bad news for a number of
1505  * reasons, so we follow Perl's lead and sort "equal" strings
1506  * according to strcmp (on the UTF-8 representation).
1507  */
1508  if (result == 0)
1509  {
1510  result = memcmp(arg1, arg2, Min(len1, len2));
1511  if ((result == 0) && (len1 != len2))
1512  result = (len1 < len2) ? -1 : 1;
1513  }
1514 
1515  if (a1p != a1buf)
1516  pfree(a1p);
1517  if (a2p != a2buf)
1518  pfree(a2p);
1519 
1520  return result;
1521  }
1522 #endif /* WIN32 */
1523 
1524  if (len1 >= TEXTBUFLEN)
1525  a1p = (char *) palloc(len1 + 1);
1526  else
1527  a1p = a1buf;
1528  if (len2 >= TEXTBUFLEN)
1529  a2p = (char *) palloc(len2 + 1);
1530  else
1531  a2p = a2buf;
1532 
1533  memcpy(a1p, arg1, len1);
1534  a1p[len1] = '\0';
1535  memcpy(a2p, arg2, len2);
1536  a2p[len2] = '\0';
1537 
1538  if (mylocale)
1539  {
1540  if (mylocale->provider == COLLPROVIDER_ICU)
1541  {
1542 #ifdef USE_ICU
1543 #ifdef HAVE_UCOL_STRCOLLUTF8
1544  if (GetDatabaseEncoding() == PG_UTF8)
1545  {
1546  UErrorCode status;
1547 
1548  status = U_ZERO_ERROR;
1549  result = ucol_strcollUTF8(mylocale->info.icu.ucol,
1550  arg1, len1,
1551  arg2, len2,
1552  &status);
1553  if (U_FAILURE(status))
1554  ereport(ERROR,
1555  (errmsg("collation failed: %s", u_errorName(status))));
1556  }
1557  else
1558 #endif
1559  {
1560  int32_t ulen1, ulen2;
1561  UChar *uchar1, *uchar2;
1562 
1563  ulen1 = icu_to_uchar(&uchar1, arg1, len1);
1564  ulen2 = icu_to_uchar(&uchar2, arg2, len2);
1565 
1566  result = ucol_strcoll(mylocale->info.icu.ucol,
1567  uchar1, ulen1,
1568  uchar2, ulen2);
1569  }
1570 #else /* not USE_ICU */
1571  /* shouldn't happen */
1572  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1573 #endif /* not USE_ICU */
1574  }
1575  else
1576  {
1577 #ifdef HAVE_LOCALE_T
1578  result = strcoll_l(a1p, a2p, mylocale->info.lt);
1579 #else
1580  /* shouldn't happen */
1581  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1582 #endif
1583  }
1584  }
1585  else
1586  result = strcoll(a1p, a2p);
1587 
1588  /*
1589  * In some locales strcoll() can claim that nonidentical strings are
1590  * equal. Believing that would be bad news for a number of reasons,
1591  * so we follow Perl's lead and sort "equal" strings according to
1592  * strcmp().
1593  */
1594  if (result == 0)
1595  result = strcmp(a1p, a2p);
1596 
1597  if (a1p != a1buf)
1598  pfree(a1p);
1599  if (a2p != a2buf)
1600  pfree(a2p);
1601  }
1602 
1603  return result;
1604 }
1605 
1606 /* text_cmp()
1607  * Internal comparison function for text strings.
1608  * Returns -1, 0 or 1
1609  */
1610 static int
1611 text_cmp(text *arg1, text *arg2, Oid collid)
1612 {
1613  char *a1p,
1614  *a2p;
1615  int len1,
1616  len2;
1617 
1618  a1p = VARDATA_ANY(arg1);
1619  a2p = VARDATA_ANY(arg2);
1620 
1621  len1 = VARSIZE_ANY_EXHDR(arg1);
1622  len2 = VARSIZE_ANY_EXHDR(arg2);
1623 
1624  return varstr_cmp(a1p, len1, a2p, len2, collid);
1625 }
1626 
1627 /*
1628  * Comparison functions for text strings.
1629  *
1630  * Note: btree indexes need these routines not to leak memory; therefore,
1631  * be careful to free working copies of toasted datums. Most places don't
1632  * need to be so careful.
1633  */
1634 
1635 Datum
1637 {
1638  Datum arg1 = PG_GETARG_DATUM(0);
1639  Datum arg2 = PG_GETARG_DATUM(1);
1640  bool result;
1641  Size len1,
1642  len2;
1643 
1644  /*
1645  * Since we only care about equality or not-equality, we can avoid all the
1646  * expense of strcoll() here, and just do bitwise comparison. In fact, we
1647  * don't even have to do a bitwise comparison if we can show the lengths
1648  * of the strings are unequal; which might save us from having to detoast
1649  * one or both values.
1650  */
1651  len1 = toast_raw_datum_size(arg1);
1652  len2 = toast_raw_datum_size(arg2);
1653  if (len1 != len2)
1654  result = false;
1655  else
1656  {
1657  text *targ1 = DatumGetTextPP(arg1);
1658  text *targ2 = DatumGetTextPP(arg2);
1659 
1660  result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1661  len1 - VARHDRSZ) == 0);
1662 
1663  PG_FREE_IF_COPY(targ1, 0);
1664  PG_FREE_IF_COPY(targ2, 1);
1665  }
1666 
1667  PG_RETURN_BOOL(result);
1668 }
1669 
1670 Datum
1672 {
1673  Datum arg1 = PG_GETARG_DATUM(0);
1674  Datum arg2 = PG_GETARG_DATUM(1);
1675  bool result;
1676  Size len1,
1677  len2;
1678 
1679  /* See comment in texteq() */
1680  len1 = toast_raw_datum_size(arg1);
1681  len2 = toast_raw_datum_size(arg2);
1682  if (len1 != len2)
1683  result = true;
1684  else
1685  {
1686  text *targ1 = DatumGetTextPP(arg1);
1687  text *targ2 = DatumGetTextPP(arg2);
1688 
1689  result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1690  len1 - VARHDRSZ) != 0);
1691 
1692  PG_FREE_IF_COPY(targ1, 0);
1693  PG_FREE_IF_COPY(targ2, 1);
1694  }
1695 
1696  PG_RETURN_BOOL(result);
1697 }
1698 
1699 Datum
1701 {
1702  text *arg1 = PG_GETARG_TEXT_PP(0);
1703  text *arg2 = PG_GETARG_TEXT_PP(1);
1704  bool result;
1705 
1706  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1707 
1708  PG_FREE_IF_COPY(arg1, 0);
1709  PG_FREE_IF_COPY(arg2, 1);
1710 
1711  PG_RETURN_BOOL(result);
1712 }
1713 
1714 Datum
1716 {
1717  text *arg1 = PG_GETARG_TEXT_PP(0);
1718  text *arg2 = PG_GETARG_TEXT_PP(1);
1719  bool result;
1720 
1721  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1722 
1723  PG_FREE_IF_COPY(arg1, 0);
1724  PG_FREE_IF_COPY(arg2, 1);
1725 
1726  PG_RETURN_BOOL(result);
1727 }
1728 
1729 Datum
1731 {
1732  text *arg1 = PG_GETARG_TEXT_PP(0);
1733  text *arg2 = PG_GETARG_TEXT_PP(1);
1734  bool result;
1735 
1736  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1737 
1738  PG_FREE_IF_COPY(arg1, 0);
1739  PG_FREE_IF_COPY(arg2, 1);
1740 
1741  PG_RETURN_BOOL(result);
1742 }
1743 
1744 Datum
1746 {
1747  text *arg1 = PG_GETARG_TEXT_PP(0);
1748  text *arg2 = PG_GETARG_TEXT_PP(1);
1749  bool result;
1750 
1751  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1752 
1753  PG_FREE_IF_COPY(arg1, 0);
1754  PG_FREE_IF_COPY(arg2, 1);
1755 
1756  PG_RETURN_BOOL(result);
1757 }
1758 
1759 Datum
1761 {
1762  text *arg1 = PG_GETARG_TEXT_PP(0);
1763  text *arg2 = PG_GETARG_TEXT_PP(1);
1764  int32 result;
1765 
1766  result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1767 
1768  PG_FREE_IF_COPY(arg1, 0);
1769  PG_FREE_IF_COPY(arg2, 1);
1770 
1771  PG_RETURN_INT32(result);
1772 }
1773 
1774 Datum
1776 {
1778  Oid collid = ssup->ssup_collation;
1779  MemoryContext oldcontext;
1780 
1781  oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1782 
1783  /* Use generic string SortSupport */
1784  varstr_sortsupport(ssup, collid, false);
1785 
1786  MemoryContextSwitchTo(oldcontext);
1787 
1788  PG_RETURN_VOID();
1789 }
1790 
1791 /*
1792  * Generic sortsupport interface for character type's operator classes.
1793  * Includes locale support, and support for BpChar semantics (i.e. removing
1794  * trailing spaces before comparison).
1795  *
1796  * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
1797  * same representation. Callers that always use the C collation (e.g.
1798  * non-collatable type callers like bytea) may have NUL bytes in their strings;
1799  * this will not work with any other collation, though.
1800  */
1801 void
1803 {
1804  bool abbreviate = ssup->abbreviate;
1805  bool collate_c = false;
1806  VarStringSortSupport *sss;
1807  pg_locale_t locale = 0;
1808 
1809  /*
1810  * If possible, set ssup->comparator to a function which can be used to
1811  * directly compare two datums. If we can do this, we'll avoid the
1812  * overhead of a trip through the fmgr layer for every comparison, which
1813  * can be substantial.
1814  *
1815  * Most typically, we'll set the comparator to varstrfastcmp_locale, which
1816  * uses strcoll() to perform comparisons and knows about the special
1817  * requirements of BpChar callers. However, if LC_COLLATE = C, we can
1818  * make things quite a bit faster with varstrfastcmp_c or bpcharfastcmp_c,
1819  * both of which use memcmp() rather than strcoll().
1820  *
1821  * There is a further exception on Windows. When the database encoding is
1822  * UTF-8 and we are not using the C collation, complex hacks are required.
1823  * We don't currently have a comparator that handles that case, so we fall
1824  * back on the slow method of having the sort code invoke bttextcmp() (in
1825  * the case of text) via the fmgr trampoline.
1826  */
1827  if (lc_collate_is_c(collid))
1828  {
1829  if (!bpchar)
1830  ssup->comparator = varstrfastcmp_c;
1831  else
1832  ssup->comparator = bpcharfastcmp_c;
1833 
1834  collate_c = true;
1835  }
1836 #ifdef WIN32
1837  else if (GetDatabaseEncoding() == PG_UTF8)
1838  return;
1839 #endif
1840  else
1841  {
1843 
1844  /*
1845  * We need a collation-sensitive comparison. To make things faster,
1846  * we'll figure out the collation based on the locale id and cache the
1847  * result.
1848  */
1849  if (collid != DEFAULT_COLLATION_OID)
1850  {
1851  if (!OidIsValid(collid))
1852  {
1853  /*
1854  * This typically means that the parser could not resolve a
1855  * conflict of implicit collations, so report it that way.
1856  */
1857  ereport(ERROR,
1858  (errcode(ERRCODE_INDETERMINATE_COLLATION),
1859  errmsg("could not determine which collation to use for string comparison"),
1860  errhint("Use the COLLATE clause to set the collation explicitly.")));
1861  }
1862  locale = pg_newlocale_from_collation(collid);
1863  }
1864  }
1865 
1866  /*
1867  * Unfortunately, it seems that abbreviation for non-C collations is
1868  * broken on many common platforms; testing of multiple versions of glibc
1869  * reveals that, for many locales, strcoll() and strxfrm() do not return
1870  * consistent results, which is fatal to this optimization. While no
1871  * other libc other than Cygwin has so far been shown to have a problem,
1872  * we take the conservative course of action for right now and disable
1873  * this categorically. (Users who are certain this isn't a problem on
1874  * their system can define TRUST_STRXFRM.)
1875  *
1876  * Even apart from the risk of broken locales, it's possible that there
1877  * are platforms where the use of abbreviated keys should be disabled at
1878  * compile time. Having only 4 byte datums could make worst-case
1879  * performance drastically more likely, for example. Moreover, macOS's
1880  * strxfrm() implementation is known to not effectively concentrate a
1881  * significant amount of entropy from the original string in earlier
1882  * transformed blobs. It's possible that other supported platforms are
1883  * similarly encumbered. So, if we ever get past disabling this
1884  * categorically, we may still want or need to disable it for particular
1885  * platforms.
1886  */
1887 #ifndef TRUST_STRXFRM
1888  if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
1889  abbreviate = false;
1890 #endif
1891 
1892  /*
1893  * If we're using abbreviated keys, or if we're using a locale-aware
1894  * comparison, we need to initialize a StringSortSupport object. Both
1895  * cases will make use of the temporary buffers we initialize here for
1896  * scratch space (and to detect requirement for BpChar semantics from
1897  * caller), and the abbreviation case requires additional state.
1898  */
1899  if (abbreviate || !collate_c)
1900  {
1901  sss = palloc(sizeof(VarStringSortSupport));
1902  sss->buf1 = palloc(TEXTBUFLEN);
1903  sss->buflen1 = TEXTBUFLEN;
1904  sss->buf2 = palloc(TEXTBUFLEN);
1905  sss->buflen2 = TEXTBUFLEN;
1906  /* Start with invalid values */
1907  sss->last_len1 = -1;
1908  sss->last_len2 = -1;
1909  /* Initialize */
1910  sss->last_returned = 0;
1911  sss->locale = locale;
1912 
1913  /*
1914  * To avoid somehow confusing a strxfrm() blob and an original string,
1915  * constantly keep track of the variety of data that buf1 and buf2
1916  * currently contain.
1917  *
1918  * Comparisons may be interleaved with conversion calls. Frequently,
1919  * conversions and comparisons are batched into two distinct phases,
1920  * but the correctness of caching cannot hinge upon this. For
1921  * comparison caching, buffer state is only trusted if cache_blob is
1922  * found set to false, whereas strxfrm() caching only trusts the state
1923  * when cache_blob is found set to true.
1924  *
1925  * Arbitrarily initialize cache_blob to true.
1926  */
1927  sss->cache_blob = true;
1928  sss->collate_c = collate_c;
1929  sss->bpchar = bpchar;
1930  ssup->ssup_extra = sss;
1931 
1932  /*
1933  * If possible, plan to use the abbreviated keys optimization. The
1934  * core code may switch back to authoritative comparator should
1935  * abbreviation be aborted.
1936  */
1937  if (abbreviate)
1938  {
1939  sss->prop_card = 0.20;
1940  initHyperLogLog(&sss->abbr_card, 10);
1941  initHyperLogLog(&sss->full_card, 10);
1942  ssup->abbrev_full_comparator = ssup->comparator;
1943  ssup->comparator = varstrcmp_abbrev;
1946  }
1947  }
1948 }
1949 
1950 /*
1951  * sortsupport comparison func (for C locale case)
1952  */
1953 static int
1955 {
1956  VarString *arg1 = DatumGetVarStringPP(x);
1957  VarString *arg2 = DatumGetVarStringPP(y);
1958  char *a1p,
1959  *a2p;
1960  int len1,
1961  len2,
1962  result;
1963 
1964  a1p = VARDATA_ANY(arg1);
1965  a2p = VARDATA_ANY(arg2);
1966 
1967  len1 = VARSIZE_ANY_EXHDR(arg1);
1968  len2 = VARSIZE_ANY_EXHDR(arg2);
1969 
1970  result = memcmp(a1p, a2p, Min(len1, len2));
1971  if ((result == 0) && (len1 != len2))
1972  result = (len1 < len2) ? -1 : 1;
1973 
1974  /* We can't afford to leak memory here. */
1975  if (PointerGetDatum(arg1) != x)
1976  pfree(arg1);
1977  if (PointerGetDatum(arg2) != y)
1978  pfree(arg2);
1979 
1980  return result;
1981 }
1982 
1983 /*
1984  * sortsupport comparison func (for BpChar C locale case)
1985  *
1986  * BpChar outsources its sortsupport to this module. Specialization for the
1987  * varstr_sortsupport BpChar case, modeled on
1988  * internal_bpchar_pattern_compare().
1989  */
1990 static int
1992 {
1993  BpChar *arg1 = DatumGetBpCharPP(x);
1994  BpChar *arg2 = DatumGetBpCharPP(y);
1995  char *a1p,
1996  *a2p;
1997  int len1,
1998  len2,
1999  result;
2000 
2001  a1p = VARDATA_ANY(arg1);
2002  a2p = VARDATA_ANY(arg2);
2003 
2004  len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
2005  len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
2006 
2007  result = memcmp(a1p, a2p, Min(len1, len2));
2008  if ((result == 0) && (len1 != len2))
2009  result = (len1 < len2) ? -1 : 1;
2010 
2011  /* We can't afford to leak memory here. */
2012  if (PointerGetDatum(arg1) != x)
2013  pfree(arg1);
2014  if (PointerGetDatum(arg2) != y)
2015  pfree(arg2);
2016 
2017  return result;
2018 }
2019 
2020 /*
2021  * sortsupport comparison func (for locale case)
2022  */
2023 static int
2025 {
2026  VarString *arg1 = DatumGetVarStringPP(x);
2027  VarString *arg2 = DatumGetVarStringPP(y);
2028  bool arg1_match;
2030 
2031  /* working state */
2032  char *a1p,
2033  *a2p;
2034  int len1,
2035  len2,
2036  result;
2037 
2038  a1p = VARDATA_ANY(arg1);
2039  a2p = VARDATA_ANY(arg2);
2040 
2041  len1 = VARSIZE_ANY_EXHDR(arg1);
2042  len2 = VARSIZE_ANY_EXHDR(arg2);
2043 
2044  /* Fast pre-check for equality, as discussed in varstr_cmp() */
2045  if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
2046  {
2047  /*
2048  * No change in buf1 or buf2 contents, so avoid changing last_len1 or
2049  * last_len2. Existing contents of buffers might still be used by
2050  * next call.
2051  *
2052  * It's fine to allow the comparison of BpChar padding bytes here,
2053  * even though that implies that the memcmp() will usually be
2054  * performed for BpChar callers (though multibyte characters could
2055  * still prevent that from occurring). The memcmp() is still very
2056  * cheap, and BpChar's funny semantics have us remove trailing spaces
2057  * (not limited to padding), so we need make no distinction between
2058  * padding space characters and "real" space characters.
2059  */
2060  result = 0;
2061  goto done;
2062  }
2063 
2064  if (sss->bpchar)
2065  {
2066  /* Get true number of bytes, ignoring trailing spaces */
2067  len1 = bpchartruelen(a1p, len1);
2068  len2 = bpchartruelen(a2p, len2);
2069  }
2070 
2071  if (len1 >= sss->buflen1)
2072  {
2073  pfree(sss->buf1);
2074  sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2075  sss->buf1 = MemoryContextAlloc(ssup->ssup_cxt, sss->buflen1);
2076  }
2077  if (len2 >= sss->buflen2)
2078  {
2079  pfree(sss->buf2);
2080  sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
2081  sss->buf2 = MemoryContextAlloc(ssup->ssup_cxt, sss->buflen2);
2082  }
2083 
2084  /*
2085  * We're likely to be asked to compare the same strings repeatedly, and
2086  * memcmp() is so much cheaper than strcoll() that it pays to try to cache
2087  * comparisons, even though in general there is no reason to think that
2088  * that will work out (every string datum may be unique). Caching does
2089  * not slow things down measurably when it doesn't work out, and can speed
2090  * things up by rather a lot when it does. In part, this is because the
2091  * memcmp() compares data from cachelines that are needed in L1 cache even
2092  * when the last comparison's result cannot be reused.
2093  */
2094  arg1_match = true;
2095  if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
2096  {
2097  arg1_match = false;
2098  memcpy(sss->buf1, a1p, len1);
2099  sss->buf1[len1] = '\0';
2100  sss->last_len1 = len1;
2101  }
2102 
2103  /*
2104  * If we're comparing the same two strings as last time, we can return the
2105  * same answer without calling strcoll() again. This is more likely than
2106  * it seems (at least with moderate to low cardinality sets), because
2107  * quicksort compares the same pivot against many values.
2108  */
2109  if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2110  {
2111  memcpy(sss->buf2, a2p, len2);
2112  sss->buf2[len2] = '\0';
2113  sss->last_len2 = len2;
2114  }
2115  else if (arg1_match && !sss->cache_blob)
2116  {
2117  /* Use result cached following last actual strcoll() call */
2118  result = sss->last_returned;
2119  goto done;
2120  }
2121 
2122  if (sss->locale)
2123  {
2124  if (sss->locale->provider == COLLPROVIDER_ICU)
2125  {
2126 #ifdef USE_ICU
2127 #ifdef HAVE_UCOL_STRCOLLUTF8
2128  if (GetDatabaseEncoding() == PG_UTF8)
2129  {
2130  UErrorCode status;
2131 
2132  status = U_ZERO_ERROR;
2133  result = ucol_strcollUTF8(sss->locale->info.icu.ucol,
2134  a1p, len1,
2135  a2p, len2,
2136  &status);
2137  if (U_FAILURE(status))
2138  ereport(ERROR,
2139  (errmsg("collation failed: %s", u_errorName(status))));
2140  }
2141  else
2142 #endif
2143  {
2144  int32_t ulen1, ulen2;
2145  UChar *uchar1, *uchar2;
2146 
2147  ulen1 = icu_to_uchar(&uchar1, a1p, len1);
2148  ulen2 = icu_to_uchar(&uchar2, a2p, len2);
2149 
2150  result = ucol_strcoll(sss->locale->info.icu.ucol,
2151  uchar1, ulen1,
2152  uchar2, ulen2);
2153  }
2154 #else /* not USE_ICU */
2155  /* shouldn't happen */
2156  elog(ERROR, "unsupported collprovider: %c", sss->locale->provider);
2157 #endif /* not USE_ICU */
2158  }
2159  else
2160  {
2161 #ifdef HAVE_LOCALE_T
2162  result = strcoll_l(sss->buf1, sss->buf2, sss->locale->info.lt);
2163 #else
2164  /* shouldn't happen */
2165  elog(ERROR, "unsupported collprovider: %c", sss->locale->provider);
2166 #endif
2167  }
2168  }
2169  else
2170  result = strcoll(sss->buf1, sss->buf2);
2171 
2172  /*
2173  * In some locales strcoll() can claim that nonidentical strings are
2174  * equal. Believing that would be bad news for a number of reasons, so we
2175  * follow Perl's lead and sort "equal" strings according to strcmp().
2176  */
2177  if (result == 0)
2178  result = strcmp(sss->buf1, sss->buf2);
2179 
2180  /* Cache result, perhaps saving an expensive strcoll() call next time */
2181  sss->cache_blob = false;
2182  sss->last_returned = result;
2183 done:
2184  /* We can't afford to leak memory here. */
2185  if (PointerGetDatum(arg1) != x)
2186  pfree(arg1);
2187  if (PointerGetDatum(arg2) != y)
2188  pfree(arg2);
2189 
2190  return result;
2191 }
2192 
2193 /*
2194  * Abbreviated key comparison func
2195  */
2196 static int
2198 {
2199  /*
2200  * When 0 is returned, the core system will call varstrfastcmp_c()
2201  * (bpcharfastcmp_c() in BpChar case) or varstrfastcmp_locale(). Even a
2202  * strcmp() on two non-truncated strxfrm() blobs cannot indicate *equality*
2203  * authoritatively, for the same reason that there is a strcoll()
2204  * tie-breaker call to strcmp() in varstr_cmp().
2205  */
2206  if (x > y)
2207  return 1;
2208  else if (x == y)
2209  return 0;
2210  else
2211  return -1;
2212 }
2213 
2214 /*
2215  * Conversion routine for sortsupport. Converts original to abbreviated key
2216  * representation. Our encoding strategy is simple -- pack the first 8 bytes
2217  * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2218  * stored in reverse order), and treat it as an unsigned integer. When the "C"
2219  * locale is used, or in case of bytea, just memcpy() from original instead.
2220  */
2221 static Datum
2223 {
2225  VarString *authoritative = DatumGetVarStringPP(original);
2226  char *authoritative_data = VARDATA_ANY(authoritative);
2227 
2228  /* working state */
2229  Datum res;
2230  char *pres;
2231  int len;
2232  uint32 hash;
2233 
2234  pres = (char *) &res;
2235  /* memset(), so any non-overwritten bytes are NUL */
2236  memset(pres, 0, sizeof(Datum));
2237  len = VARSIZE_ANY_EXHDR(authoritative);
2238 
2239  /* Get number of bytes, ignoring trailing spaces */
2240  if (sss->bpchar)
2241  len = bpchartruelen(authoritative_data, len);
2242 
2243  /*
2244  * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2245  * abbreviate keys. The full comparator for the C locale is always
2246  * memcmp(). It would be incorrect to allow bytea callers (callers that
2247  * always force the C collation -- bytea isn't a collatable type, but this
2248  * approach is convenient) to use strxfrm(). This is because bytea
2249  * strings may contain NUL bytes. Besides, this should be faster, too.
2250  *
2251  * More generally, it's okay that bytea callers can have NUL bytes in
2252  * strings because varstrcmp_abbrev() need not make a distinction between
2253  * terminating NUL bytes, and NUL bytes representing actual NULs in the
2254  * authoritative representation. Hopefully a comparison at or past one
2255  * abbreviated key's terminating NUL byte will resolve the comparison
2256  * without consulting the authoritative representation; specifically, some
2257  * later non-NUL byte in the longer string can resolve the comparison
2258  * against a subsequent terminating NUL in the shorter string. There will
2259  * usually be what is effectively a "length-wise" resolution there and
2260  * then.
2261  *
2262  * If that doesn't work out -- if all bytes in the longer string
2263  * positioned at or past the offset of the smaller string's (first)
2264  * terminating NUL are actually representative of NUL bytes in the
2265  * authoritative binary string (perhaps with some *terminating* NUL bytes
2266  * towards the end of the longer string iff it happens to still be small)
2267  * -- then an authoritative tie-breaker will happen, and do the right
2268  * thing: explicitly consider string length.
2269  */
2270  if (sss->collate_c)
2271  memcpy(pres, authoritative_data, Min(len, sizeof(Datum)));
2272  else
2273  {
2274  Size bsize;
2275 #ifdef USE_ICU
2276  int32_t ulen = -1;
2277  UChar *uchar;
2278 #endif
2279 
2280  /*
2281  * We're not using the C collation, so fall back on strxfrm or ICU
2282  * analogs.
2283  */
2284 
2285  /* By convention, we use buffer 1 to store and NUL-terminate */
2286  if (len >= sss->buflen1)
2287  {
2288  pfree(sss->buf1);
2289  sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2290  sss->buf1 = palloc(sss->buflen1);
2291  }
2292 
2293  /* Might be able to reuse strxfrm() blob from last call */
2294  if (sss->last_len1 == len && sss->cache_blob &&
2295  memcmp(sss->buf1, authoritative_data, len) == 0)
2296  {
2297  memcpy(pres, sss->buf2, Min(sizeof(Datum), sss->last_len2));
2298  /* No change affecting cardinality, so no hashing required */
2299  goto done;
2300  }
2301 
2302  memcpy(sss->buf1, authoritative_data, len);
2303  /* Just like strcoll(), strxfrm() expects a NUL-terminated string.
2304  * Not necessary for ICU, but doesn't hurt. */
2305  sss->buf1[len] = '\0';
2306  sss->last_len1 = len;
2307 
2308 #ifdef USE_ICU
2309  /* When using ICU and not UTF8, convert string to UChar. */
2310  if (sss->locale && sss->locale->provider == COLLPROVIDER_ICU &&
2312  ulen = icu_to_uchar(&uchar, sss->buf1, len);
2313 #endif
2314 
2315  /*
2316  * Loop: Call strxfrm() or ucol_getSortKey(), possibly enlarge buffer,
2317  * and try again. Both of these functions have the result buffer
2318  * content undefined if the result did not fit, so we need to retry
2319  * until everything fits, even though we only need the first few bytes
2320  * in the end. When using ucol_nextSortKeyPart(), however, we only
2321  * ask for as many bytes as we actually need.
2322  */
2323  for (;;)
2324  {
2325 #ifdef USE_ICU
2326  if (sss->locale && sss->locale->provider == COLLPROVIDER_ICU)
2327  {
2328  /*
2329  * When using UTF8, use the iteration interface so we only
2330  * need to produce as many bytes as we actually need.
2331  */
2332  if (GetDatabaseEncoding() == PG_UTF8)
2333  {
2334  UCharIterator iter;
2335  uint32_t state[2];
2336  UErrorCode status;
2337 
2338  uiter_setUTF8(&iter, sss->buf1, len);
2339  state[0] = state[1] = 0; /* won't need that again */
2340  status = U_ZERO_ERROR;
2341  bsize = ucol_nextSortKeyPart(sss->locale->info.icu.ucol,
2342  &iter,
2343  state,
2344  (uint8_t *) sss->buf2,
2345  Min(sizeof(Datum), sss->buflen2),
2346  &status);
2347  if (U_FAILURE(status))
2348  ereport(ERROR,
2349  (errmsg("sort key generation failed: %s", u_errorName(status))));
2350  }
2351  else
2352  bsize = ucol_getSortKey(sss->locale->info.icu.ucol,
2353  uchar, ulen,
2354  (uint8_t *) sss->buf2, sss->buflen2);
2355  }
2356  else
2357 #endif
2358 #ifdef HAVE_LOCALE_T
2359  if (sss->locale && sss->locale->provider == COLLPROVIDER_LIBC)
2360  bsize = strxfrm_l(sss->buf2, sss->buf1,
2361  sss->buflen2, sss->locale->info.lt);
2362  else
2363 #endif
2364  bsize = strxfrm(sss->buf2, sss->buf1, sss->buflen2);
2365 
2366  sss->last_len2 = bsize;
2367  if (bsize < sss->buflen2)
2368  break;
2369 
2370  /*
2371  * Grow buffer and retry.
2372  */
2373  pfree(sss->buf2);
2374  sss->buflen2 = Max(bsize + 1,
2375  Min(sss->buflen2 * 2, MaxAllocSize));
2376  sss->buf2 = palloc(sss->buflen2);
2377  }
2378 
2379  /*
2380  * Every Datum byte is always compared. This is safe because the
2381  * strxfrm() blob is itself NUL terminated, leaving no danger of
2382  * misinterpreting any NUL bytes not intended to be interpreted as
2383  * logically representing termination.
2384  *
2385  * (Actually, even if there were NUL bytes in the blob it would be
2386  * okay. See remarks on bytea case above.)
2387  */
2388  memcpy(pres, sss->buf2, Min(sizeof(Datum), bsize));
2389  }
2390 
2391  /*
2392  * Maintain approximate cardinality of both abbreviated keys and original,
2393  * authoritative keys using HyperLogLog. Used as cheap insurance against
2394  * the worst case, where we do many string transformations for no saving
2395  * in full strcoll()-based comparisons. These statistics are used by
2396  * varstr_abbrev_abort().
2397  *
2398  * First, Hash key proper, or a significant fraction of it. Mix in length
2399  * in order to compensate for cases where differences are past
2400  * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2401  */
2402  hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2403  Min(len, PG_CACHE_LINE_SIZE)));
2404 
2405  if (len > PG_CACHE_LINE_SIZE)
2406  hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2407 
2408  addHyperLogLog(&sss->full_card, hash);
2409 
2410  /* Hash abbreviated key */
2411 #if SIZEOF_DATUM == 8
2412  {
2413  uint32 lohalf,
2414  hihalf;
2415 
2416  lohalf = (uint32) res;
2417  hihalf = (uint32) (res >> 32);
2418  hash = DatumGetUInt32(hash_uint32(lohalf ^ hihalf));
2419  }
2420 #else /* SIZEOF_DATUM != 8 */
2421  hash = DatumGetUInt32(hash_uint32((uint32) res));
2422 #endif
2423 
2424  addHyperLogLog(&sss->abbr_card, hash);
2425 
2426  /* Cache result, perhaps saving an expensive strxfrm() call next time */
2427  sss->cache_blob = true;
2428 done:
2429 
2430  /*
2431  * Byteswap on little-endian machines.
2432  *
2433  * This is needed so that varstrcmp_abbrev() (an unsigned integer 3-way
2434  * comparator) works correctly on all platforms. If we didn't do this,
2435  * the comparator would have to call memcmp() with a pair of pointers to
2436  * the first byte of each abbreviated key, which is slower.
2437  */
2438  res = DatumBigEndianToNative(res);
2439 
2440  /* Don't leak memory here */
2441  if (PointerGetDatum(authoritative) != original)
2442  pfree(authoritative);
2443 
2444  return res;
2445 }
2446 
2447 /*
2448  * Callback for estimating effectiveness of abbreviated key optimization, using
2449  * heuristic rules. Returns value indicating if the abbreviation optimization
2450  * should be aborted, based on its projected effectiveness.
2451  */
2452 static bool
2453 varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2454 {
2456  double abbrev_distinct,
2457  key_distinct;
2458 
2459  Assert(ssup->abbreviate);
2460 
2461  /* Have a little patience */
2462  if (memtupcount < 100)
2463  return false;
2464 
2465  abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2466  key_distinct = estimateHyperLogLog(&sss->full_card);
2467 
2468  /*
2469  * Clamp cardinality estimates to at least one distinct value. While
2470  * NULLs are generally disregarded, if only NULL values were seen so far,
2471  * that might misrepresent costs if we failed to clamp.
2472  */
2473  if (abbrev_distinct <= 1.0)
2474  abbrev_distinct = 1.0;
2475 
2476  if (key_distinct <= 1.0)
2477  key_distinct = 1.0;
2478 
2479  /*
2480  * In the worst case all abbreviated keys are identical, while at the same
2481  * time there are differences within full key strings not captured in
2482  * abbreviations.
2483  */
2484 #ifdef TRACE_SORT
2485  if (trace_sort)
2486  {
2487  double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2488 
2489  elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2490  "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2491  memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2492  sss->prop_card);
2493  }
2494 #endif
2495 
2496  /*
2497  * If the number of distinct abbreviated keys approximately matches the
2498  * number of distinct authoritative original keys, that's reason enough to
2499  * proceed. We can win even with a very low cardinality set if most
2500  * tie-breakers only memcmp(). This is by far the most important
2501  * consideration.
2502  *
2503  * While comparisons that are resolved at the abbreviated key level are
2504  * considerably cheaper than tie-breakers resolved with memcmp(), both of
2505  * those two outcomes are so much cheaper than a full strcoll() once
2506  * sorting is underway that it doesn't seem worth it to weigh abbreviated
2507  * cardinality against the overall size of the set in order to more
2508  * accurately model costs. Assume that an abbreviated comparison, and an
2509  * abbreviated comparison with a cheap memcmp()-based authoritative
2510  * resolution are equivalent.
2511  */
2512  if (abbrev_distinct > key_distinct * sss->prop_card)
2513  {
2514  /*
2515  * When we have exceeded 10,000 tuples, decay required cardinality
2516  * aggressively for next call.
2517  *
2518  * This is useful because the number of comparisons required on
2519  * average increases at a linearithmic rate, and at roughly 10,000
2520  * tuples that factor will start to dominate over the linear costs of
2521  * string transformation (this is a conservative estimate). The decay
2522  * rate is chosen to be a little less aggressive than halving -- which
2523  * (since we're called at points at which memtupcount has doubled)
2524  * would never see the cost model actually abort past the first call
2525  * following a decay. This decay rate is mostly a precaution against
2526  * a sudden, violent swing in how well abbreviated cardinality tracks
2527  * full key cardinality. The decay also serves to prevent a marginal
2528  * case from being aborted too late, when too much has already been
2529  * invested in string transformation.
2530  *
2531  * It's possible for sets of several million distinct strings with
2532  * mere tens of thousands of distinct abbreviated keys to still
2533  * benefit very significantly. This will generally occur provided
2534  * each abbreviated key is a proxy for a roughly uniform number of the
2535  * set's full keys. If it isn't so, we hope to catch that early and
2536  * abort. If it isn't caught early, by the time the problem is
2537  * apparent it's probably not worth aborting.
2538  */
2539  if (memtupcount > 10000)
2540  sss->prop_card *= 0.65;
2541 
2542  return false;
2543  }
2544 
2545  /*
2546  * Abort abbreviation strategy.
2547  *
2548  * The worst case, where all abbreviated keys are identical while all
2549  * original strings differ will typically only see a regression of about
2550  * 10% in execution time for small to medium sized lists of strings.
2551  * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2552  * often expect very large improvements, particularly with sets of strings
2553  * of moderately high to high abbreviated cardinality. There is little to
2554  * lose but much to gain, which our strategy reflects.
2555  */
2556 #ifdef TRACE_SORT
2557  if (trace_sort)
2558  elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2559  "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2560  memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2561 #endif
2562 
2563  return true;
2564 }
2565 
2566 Datum
2568 {
2569  text *arg1 = PG_GETARG_TEXT_PP(0);
2570  text *arg2 = PG_GETARG_TEXT_PP(1);
2571  text *result;
2572 
2573  result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2574 
2575  PG_RETURN_TEXT_P(result);
2576 }
2577 
2578 Datum
2580 {
2581  text *arg1 = PG_GETARG_TEXT_PP(0);
2582  text *arg2 = PG_GETARG_TEXT_PP(1);
2583  text *result;
2584 
2585  result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2586 
2587  PG_RETURN_TEXT_P(result);
2588 }
2589 
2590 
2591 /*
2592  * The following operators support character-by-character comparison
2593  * of text datums, to allow building indexes suitable for LIKE clauses.
2594  * Note that the regular texteq/textne comparison operators, and regular
2595  * support functions 1 and 2 with "C" collation are assumed to be
2596  * compatible with these!
2597  */
2598 
2599 static int
2601 {
2602  int result;
2603  int len1,
2604  len2;
2605 
2606  len1 = VARSIZE_ANY_EXHDR(arg1);
2607  len2 = VARSIZE_ANY_EXHDR(arg2);
2608 
2609  result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2610  if (result != 0)
2611  return result;
2612  else if (len1 < len2)
2613  return -1;
2614  else if (len1 > len2)
2615  return 1;
2616  else
2617  return 0;
2618 }
2619 
2620 
2621 Datum
2623 {
2624  text *arg1 = PG_GETARG_TEXT_PP(0);
2625  text *arg2 = PG_GETARG_TEXT_PP(1);
2626  int result;
2627 
2628  result = internal_text_pattern_compare(arg1, arg2);
2629 
2630  PG_FREE_IF_COPY(arg1, 0);
2631  PG_FREE_IF_COPY(arg2, 1);
2632 
2633  PG_RETURN_BOOL(result < 0);
2634 }
2635 
2636 
2637 Datum
2639 {
2640  text *arg1 = PG_GETARG_TEXT_PP(0);
2641  text *arg2 = PG_GETARG_TEXT_PP(1);
2642  int result;
2643 
2644  result = internal_text_pattern_compare(arg1, arg2);
2645 
2646  PG_FREE_IF_COPY(arg1, 0);
2647  PG_FREE_IF_COPY(arg2, 1);
2648 
2649  PG_RETURN_BOOL(result <= 0);
2650 }
2651 
2652 
2653 Datum
2655 {
2656  text *arg1 = PG_GETARG_TEXT_PP(0);
2657  text *arg2 = PG_GETARG_TEXT_PP(1);
2658  int result;
2659 
2660  result = internal_text_pattern_compare(arg1, arg2);
2661 
2662  PG_FREE_IF_COPY(arg1, 0);
2663  PG_FREE_IF_COPY(arg2, 1);
2664 
2665  PG_RETURN_BOOL(result >= 0);
2666 }
2667 
2668 
2669 Datum
2671 {
2672  text *arg1 = PG_GETARG_TEXT_PP(0);
2673  text *arg2 = PG_GETARG_TEXT_PP(1);
2674  int result;
2675 
2676  result = internal_text_pattern_compare(arg1, arg2);
2677 
2678  PG_FREE_IF_COPY(arg1, 0);
2679  PG_FREE_IF_COPY(arg2, 1);
2680 
2681  PG_RETURN_BOOL(result > 0);
2682 }
2683 
2684 
2685 Datum
2687 {
2688  text *arg1 = PG_GETARG_TEXT_PP(0);
2689  text *arg2 = PG_GETARG_TEXT_PP(1);
2690  int result;
2691 
2692  result = internal_text_pattern_compare(arg1, arg2);
2693 
2694  PG_FREE_IF_COPY(arg1, 0);
2695  PG_FREE_IF_COPY(arg2, 1);
2696 
2697  PG_RETURN_INT32(result);
2698 }
2699 
2700 
2701 Datum
2703 {
2705  MemoryContext oldcontext;
2706 
2707  oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2708 
2709  /* Use generic string SortSupport, forcing "C" collation */
2710  varstr_sortsupport(ssup, C_COLLATION_OID, false);
2711 
2712  MemoryContextSwitchTo(oldcontext);
2713 
2714  PG_RETURN_VOID();
2715 }
2716 
2717 
2718 /*-------------------------------------------------------------
2719  * byteaoctetlen
2720  *
2721  * get the number of bytes contained in an instance of type 'bytea'
2722  *-------------------------------------------------------------
2723  */
2724 Datum
2726 {
2727  Datum str = PG_GETARG_DATUM(0);
2728 
2729  /* We need not detoast the input at all */
2731 }
2732 
2733 /*
2734  * byteacat -
2735  * takes two bytea* and returns a bytea* that is the concatenation of
2736  * the two.
2737  *
2738  * Cloned from textcat and modified as required.
2739  */
2740 Datum
2742 {
2743  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2744  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2745 
2747 }
2748 
2749 /*
2750  * bytea_catenate
2751  * Guts of byteacat(), broken out so it can be used by other functions
2752  *
2753  * Arguments can be in short-header form, but not compressed or out-of-line
2754  */
2755 static bytea *
2757 {
2758  bytea *result;
2759  int len1,
2760  len2,
2761  len;
2762  char *ptr;
2763 
2764  len1 = VARSIZE_ANY_EXHDR(t1);
2765  len2 = VARSIZE_ANY_EXHDR(t2);
2766 
2767  /* paranoia ... probably should throw error instead? */
2768  if (len1 < 0)
2769  len1 = 0;
2770  if (len2 < 0)
2771  len2 = 0;
2772 
2773  len = len1 + len2 + VARHDRSZ;
2774  result = (bytea *) palloc(len);
2775 
2776  /* Set size of result string... */
2777  SET_VARSIZE(result, len);
2778 
2779  /* Fill data field of result string... */
2780  ptr = VARDATA(result);
2781  if (len1 > 0)
2782  memcpy(ptr, VARDATA_ANY(t1), len1);
2783  if (len2 > 0)
2784  memcpy(ptr + len1, VARDATA_ANY(t2), len2);
2785 
2786  return result;
2787 }
2788 
2789 #define PG_STR_GET_BYTEA(str_) \
2790  DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
2791 
2792 /*
2793  * bytea_substr()
2794  * Return a substring starting at the specified position.
2795  * Cloned from text_substr and modified as required.
2796  *
2797  * Input:
2798  * - string
2799  * - starting position (is one-based)
2800  * - string length (optional)
2801  *
2802  * If the starting position is zero or less, then return from the start of the string
2803  * adjusting the length to be consistent with the "negative start" per SQL.
2804  * If the length is less than zero, an ERROR is thrown. If no third argument
2805  * (length) is provided, the length to the end of the string is assumed.
2806  */
2807 Datum
2809 {
2811  PG_GETARG_INT32(1),
2812  PG_GETARG_INT32(2),
2813  false));
2814 }
2815 
2816 /*
2817  * bytea_substr_no_len -
2818  * Wrapper to avoid opr_sanity failure due to
2819  * one function accepting a different number of args.
2820  */
2821 Datum
2823 {
2825  PG_GETARG_INT32(1),
2826  -1,
2827  true));
2828 }
2829 
2830 static bytea *
2832  int S,
2833  int L,
2834  bool length_not_specified)
2835 {
2836  int S1; /* adjusted start position */
2837  int L1; /* adjusted substring length */
2838 
2839  S1 = Max(S, 1);
2840 
2841  if (length_not_specified)
2842  {
2843  /*
2844  * Not passed a length - DatumGetByteaPSlice() grabs everything to the
2845  * end of the string if we pass it a negative value for length.
2846  */
2847  L1 = -1;
2848  }
2849  else
2850  {
2851  /* end position */
2852  int E = S + L;
2853 
2854  /*
2855  * A negative value for L is the only way for the end position to be
2856  * before the start. SQL99 says to throw an error.
2857  */
2858  if (E < S)
2859  ereport(ERROR,
2860  (errcode(ERRCODE_SUBSTRING_ERROR),
2861  errmsg("negative substring length not allowed")));
2862 
2863  /*
2864  * A zero or negative value for the end position can happen if the
2865  * start was negative or one. SQL99 says to return a zero-length
2866  * string.
2867  */
2868  if (E < 1)
2869  return PG_STR_GET_BYTEA("");
2870 
2871  L1 = E - S1;
2872  }
2873 
2874  /*
2875  * If the start position is past the end of the string, SQL99 says to
2876  * return a zero-length string -- DatumGetByteaPSlice() will do that for
2877  * us. Convert to zero-based starting position
2878  */
2879  return DatumGetByteaPSlice(str, S1 - 1, L1);
2880 }
2881 
2882 /*
2883  * byteaoverlay
2884  * Replace specified substring of first string with second
2885  *
2886  * The SQL standard defines OVERLAY() in terms of substring and concatenation.
2887  * This code is a direct implementation of what the standard says.
2888  */
2889 Datum
2891 {
2892  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2893  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2894  int sp = PG_GETARG_INT32(2); /* substring start position */
2895  int sl = PG_GETARG_INT32(3); /* substring length */
2896 
2897  PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
2898 }
2899 
2900 Datum
2902 {
2903  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2904  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2905  int sp = PG_GETARG_INT32(2); /* substring start position */
2906  int sl;
2907 
2908  sl = VARSIZE_ANY_EXHDR(t2); /* defaults to length(t2) */
2909  PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
2910 }
2911 
2912 static bytea *
2913 bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
2914 {
2915  bytea *result;
2916  bytea *s1;
2917  bytea *s2;
2918  int sp_pl_sl;
2919 
2920  /*
2921  * Check for possible integer-overflow cases. For negative sp, throw a
2922  * "substring length" error because that's what should be expected
2923  * according to the spec's definition of OVERLAY().
2924  */
2925  if (sp <= 0)
2926  ereport(ERROR,
2927  (errcode(ERRCODE_SUBSTRING_ERROR),
2928  errmsg("negative substring length not allowed")));
2929  sp_pl_sl = sp + sl;
2930  if (sp_pl_sl <= sl)
2931  ereport(ERROR,
2932  (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
2933  errmsg("integer out of range")));
2934 
2935  s1 = bytea_substring(PointerGetDatum(t1), 1, sp - 1, false);
2936  s2 = bytea_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
2937  result = bytea_catenate(s1, t2);
2938  result = bytea_catenate(result, s2);
2939 
2940  return result;
2941 }
2942 
2943 /*
2944  * byteapos -
2945  * Return the position of the specified substring.
2946  * Implements the SQL POSITION() function.
2947  * Cloned from textpos and modified as required.
2948  */
2949 Datum
2951 {
2952  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2953  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2954  int pos;
2955  int px,
2956  p;
2957  int len1,
2958  len2;
2959  char *p1,
2960  *p2;
2961 
2962  len1 = VARSIZE_ANY_EXHDR(t1);
2963  len2 = VARSIZE_ANY_EXHDR(t2);
2964 
2965  if (len2 <= 0)
2966  PG_RETURN_INT32(1); /* result for empty pattern */
2967 
2968  p1 = VARDATA_ANY(t1);
2969  p2 = VARDATA_ANY(t2);
2970 
2971  pos = 0;
2972  px = (len1 - len2);
2973  for (p = 0; p <= px; p++)
2974  {
2975  if ((*p2 == *p1) && (memcmp(p1, p2, len2) == 0))
2976  {
2977  pos = p + 1;
2978  break;
2979  };
2980  p1++;
2981  };
2982 
2983  PG_RETURN_INT32(pos);
2984 }
2985 
2986 /*-------------------------------------------------------------
2987  * byteaGetByte
2988  *
2989  * this routine treats "bytea" as an array of bytes.
2990  * It returns the Nth byte (a number between 0 and 255).
2991  *-------------------------------------------------------------
2992  */
2993 Datum
2995 {
2996  bytea *v = PG_GETARG_BYTEA_PP(0);
2997  int32 n = PG_GETARG_INT32(1);
2998  int len;
2999  int byte;
3000 
3001  len = VARSIZE_ANY_EXHDR(v);
3002 
3003  if (n < 0 || n >= len)
3004  ereport(ERROR,
3005  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3006  errmsg("index %d out of valid range, 0..%d",
3007  n, len - 1)));
3008 
3009  byte = ((unsigned char *) VARDATA_ANY(v))[n];
3010 
3011  PG_RETURN_INT32(byte);
3012 }
3013 
3014 /*-------------------------------------------------------------
3015  * byteaGetBit
3016  *
3017  * This routine treats a "bytea" type like an array of bits.
3018  * It returns the value of the Nth bit (0 or 1).
3019  *
3020  *-------------------------------------------------------------
3021  */
3022 Datum
3024 {
3025  bytea *v = PG_GETARG_BYTEA_PP(0);
3026  int32 n = PG_GETARG_INT32(1);
3027  int byteNo,
3028  bitNo;
3029  int len;
3030  int byte;
3031 
3032  len = VARSIZE_ANY_EXHDR(v);
3033 
3034  if (n < 0 || n >= len * 8)
3035  ereport(ERROR,
3036  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3037  errmsg("index %d out of valid range, 0..%d",
3038  n, len * 8 - 1)));
3039 
3040  byteNo = n / 8;
3041  bitNo = n % 8;
3042 
3043  byte = ((unsigned char *) VARDATA_ANY(v))[byteNo];
3044 
3045  if (byte & (1 << bitNo))
3046  PG_RETURN_INT32(1);
3047  else
3048  PG_RETURN_INT32(0);
3049 }
3050 
3051 /*-------------------------------------------------------------
3052  * byteaSetByte
3053  *
3054  * Given an instance of type 'bytea' creates a new one with
3055  * the Nth byte set to the given value.
3056  *
3057  *-------------------------------------------------------------
3058  */
3059 Datum
3061 {
3062  bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3063  int32 n = PG_GETARG_INT32(1);
3064  int32 newByte = PG_GETARG_INT32(2);
3065  int len;
3066 
3067  len = VARSIZE(res) - VARHDRSZ;
3068 
3069  if (n < 0 || n >= len)
3070  ereport(ERROR,
3071  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3072  errmsg("index %d out of valid range, 0..%d",
3073  n, len - 1)));
3074 
3075  /*
3076  * Now set the byte.
3077  */
3078  ((unsigned char *) VARDATA(res))[n] = newByte;
3079 
3080  PG_RETURN_BYTEA_P(res);
3081 }
3082 
3083 /*-------------------------------------------------------------
3084  * byteaSetBit
3085  *
3086  * Given an instance of type 'bytea' creates a new one with
3087  * the Nth bit set to the given value.
3088  *
3089  *-------------------------------------------------------------
3090  */
3091 Datum
3093 {
3094  bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3095  int32 n = PG_GETARG_INT32(1);
3096  int32 newBit = PG_GETARG_INT32(2);
3097  int len;
3098  int oldByte,
3099  newByte;
3100  int byteNo,
3101  bitNo;
3102 
3103  len = VARSIZE(res) - VARHDRSZ;
3104 
3105  if (n < 0 || n >= len * 8)
3106  ereport(ERROR,
3107  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3108  errmsg("index %d out of valid range, 0..%d",
3109  n, len * 8 - 1)));
3110 
3111  byteNo = n / 8;
3112  bitNo = n % 8;
3113 
3114  /*
3115  * sanity check!
3116  */
3117  if (newBit != 0 && newBit != 1)
3118  ereport(ERROR,
3119  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3120  errmsg("new bit must be 0 or 1")));
3121 
3122  /*
3123  * Update the byte.
3124  */
3125  oldByte = ((unsigned char *) VARDATA(res))[byteNo];
3126 
3127  if (newBit == 0)
3128  newByte = oldByte & (~(1 << bitNo));
3129  else
3130  newByte = oldByte | (1 << bitNo);
3131 
3132  ((unsigned char *) VARDATA(res))[byteNo] = newByte;
3133 
3134  PG_RETURN_BYTEA_P(res);
3135 }
3136 
3137 
3138 /* text_name()
3139  * Converts a text type to a Name type.
3140  */
3141 Datum
3143 {
3144  text *s = PG_GETARG_TEXT_PP(0);
3145  Name result;
3146  int len;
3147 
3148  len = VARSIZE_ANY_EXHDR(s);
3149 
3150  /* Truncate oversize input */
3151  if (len >= NAMEDATALEN)
3152  len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
3153 
3154  /* We use palloc0 here to ensure result is zero-padded */
3155  result = (Name) palloc0(NAMEDATALEN);
3156  memcpy(NameStr(*result), VARDATA_ANY(s), len);
3157 
3158  PG_RETURN_NAME(result);
3159 }
3160 
3161 /* name_text()
3162  * Converts a Name type to a text type.
3163  */
3164 Datum
3166 {
3167  Name s = PG_GETARG_NAME(0);
3168 
3170 }
3171 
3172 
3173 /*
3174  * textToQualifiedNameList - convert a text object to list of names
3175  *
3176  * This implements the input parsing needed by nextval() and other
3177  * functions that take a text parameter representing a qualified name.
3178  * We split the name at dots, downcase if not double-quoted, and
3179  * truncate names if they're too long.
3180  */
3181 List *
3183 {
3184  char *rawname;
3185  List *result = NIL;
3186  List *namelist;
3187  ListCell *l;
3188 
3189  /* Convert to C string (handles possible detoasting). */
3190  /* Note we rely on being able to modify rawname below. */
3191  rawname = text_to_cstring(textval);
3192 
3193  if (!SplitIdentifierString(rawname, '.', &namelist))
3194  ereport(ERROR,
3195  (errcode(ERRCODE_INVALID_NAME),
3196  errmsg("invalid name syntax")));
3197 
3198  if (namelist == NIL)
3199  ereport(ERROR,
3200  (errcode(ERRCODE_INVALID_NAME),
3201  errmsg("invalid name syntax")));
3202 
3203  foreach(l, namelist)
3204  {
3205  char *curname = (char *) lfirst(l);
3206 
3207  result = lappend(result, makeString(pstrdup(curname)));
3208  }
3209 
3210  pfree(rawname);
3211  list_free(namelist);
3212 
3213  return result;
3214 }
3215 
3216 /*
3217  * SplitIdentifierString --- parse a string containing identifiers
3218  *
3219  * This is the guts of textToQualifiedNameList, and is exported for use in
3220  * other situations such as parsing GUC variables. In the GUC case, it's
3221  * important to avoid memory leaks, so the API is designed to minimize the
3222  * amount of stuff that needs to be allocated and freed.
3223  *
3224  * Inputs:
3225  * rawstring: the input string; must be overwritable! On return, it's
3226  * been modified to contain the separated identifiers.
3227  * separator: the separator punctuation expected between identifiers
3228  * (typically '.' or ','). Whitespace may also appear around
3229  * identifiers.
3230  * Outputs:
3231  * namelist: filled with a palloc'd list of pointers to identifiers within
3232  * rawstring. Caller should list_free() this even on error return.
3233  *
3234  * Returns TRUE if okay, FALSE if there is a syntax error in the string.
3235  *
3236  * Note that an empty string is considered okay here, though not in
3237  * textToQualifiedNameList.
3238  */
3239 bool
3240 SplitIdentifierString(char *rawstring, char separator,
3241  List **namelist)
3242 {
3243  char *nextp = rawstring;
3244  bool done = false;
3245 
3246  *namelist = NIL;
3247 
3248  while (isspace((unsigned char) *nextp))
3249  nextp++; /* skip leading whitespace */
3250 
3251  if (*nextp == '\0')
3252  return true; /* allow empty string */
3253 
3254  /* At the top of the loop, we are at start of a new identifier. */
3255  do
3256  {
3257  char *curname;
3258  char *endp;
3259 
3260  if (*nextp == '"')
3261  {
3262  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3263  curname = nextp + 1;
3264  for (;;)
3265  {
3266  endp = strchr(nextp + 1, '"');
3267  if (endp == NULL)
3268  return false; /* mismatched quotes */
3269  if (endp[1] != '"')
3270  break; /* found end of quoted name */
3271  /* Collapse adjacent quotes into one quote, and look again */
3272  memmove(endp, endp + 1, strlen(endp));
3273  nextp = endp;
3274  }
3275  /* endp now points at the terminating quote */
3276  nextp = endp + 1;
3277  }
3278  else
3279  {
3280  /* Unquoted name --- extends to separator or whitespace */
3281  char *downname;
3282  int len;
3283 
3284  curname = nextp;
3285  while (*nextp && *nextp != separator &&
3286  !isspace((unsigned char) *nextp))
3287  nextp++;
3288  endp = nextp;
3289  if (curname == nextp)
3290  return false; /* empty unquoted name not allowed */
3291 
3292  /*
3293  * Downcase the identifier, using same code as main lexer does.
3294  *
3295  * XXX because we want to overwrite the input in-place, we cannot
3296  * support a downcasing transformation that increases the string
3297  * length. This is not a problem given the current implementation
3298  * of downcase_truncate_identifier, but we'll probably have to do
3299  * something about this someday.
3300  */
3301  len = endp - curname;
3302  downname = downcase_truncate_identifier(curname, len, false);
3303  Assert(strlen(downname) <= len);
3304  strncpy(curname, downname, len); /* strncpy is required here */
3305  pfree(downname);
3306  }
3307 
3308  while (isspace((unsigned char) *nextp))
3309  nextp++; /* skip trailing whitespace */
3310 
3311  if (*nextp == separator)
3312  {
3313  nextp++;
3314  while (isspace((unsigned char) *nextp))
3315  nextp++; /* skip leading whitespace for next */
3316  /* we expect another name, so done remains false */
3317  }
3318  else if (*nextp == '\0')
3319  done = true;
3320  else
3321  return false; /* invalid syntax */
3322 
3323  /* Now safe to overwrite separator with a null */
3324  *endp = '\0';
3325 
3326  /* Truncate name if it's overlength */
3327  truncate_identifier(curname, strlen(curname), false);
3328 
3329  /*
3330  * Finished isolating current name --- add it to list
3331  */
3332  *namelist = lappend(*namelist, curname);
3333 
3334  /* Loop back if we didn't reach end of string */
3335  } while (!done);
3336 
3337  return true;
3338 }
3339 
3340 
3341 /*
3342  * SplitDirectoriesString --- parse a string containing directory names
3343  *
3344  * This is similar to SplitIdentifierString, except that the parsing
3345  * rules are meant to handle pathnames instead of identifiers: there is
3346  * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
3347  * and we apply canonicalize_path() to each extracted string. Because of the
3348  * last, the returned strings are separately palloc'd rather than being
3349  * pointers into rawstring --- but we still scribble on rawstring.
3350  *
3351  * Inputs:
3352  * rawstring: the input string; must be modifiable!
3353  * separator: the separator punctuation expected between directories
3354  * (typically ',' or ';'). Whitespace may also appear around
3355  * directories.
3356  * Outputs:
3357  * namelist: filled with a palloc'd list of directory names.
3358  * Caller should list_free_deep() this even on error return.
3359  *
3360  * Returns TRUE if okay, FALSE if there is a syntax error in the string.
3361  *
3362  * Note that an empty string is considered okay here.
3363  */
3364 bool
3365 SplitDirectoriesString(char *rawstring, char separator,
3366  List **namelist)
3367 {
3368  char *nextp = rawstring;
3369  bool done = false;
3370 
3371  *namelist = NIL;
3372 
3373  while (isspace((unsigned char) *nextp))
3374  nextp++; /* skip leading whitespace */
3375 
3376  if (*nextp == '\0')
3377  return true; /* allow empty string */
3378 
3379  /* At the top of the loop, we are at start of a new directory. */
3380  do
3381  {
3382  char *curname;
3383  char *endp;
3384 
3385  if (*nextp == '"')
3386  {
3387  /* Quoted name --- collapse quote-quote pairs */
3388  curname = nextp + 1;
3389  for (;;)
3390  {
3391  endp = strchr(nextp + 1, '"');
3392  if (endp == NULL)
3393  return false; /* mismatched quotes */
3394  if (endp[1] != '"')
3395  break; /* found end of quoted name */
3396  /* Collapse adjacent quotes into one quote, and look again */
3397  memmove(endp, endp + 1, strlen(endp));
3398  nextp = endp;
3399  }
3400  /* endp now points at the terminating quote */
3401  nextp = endp + 1;
3402  }
3403  else
3404  {
3405  /* Unquoted name --- extends to separator or end of string */
3406  curname = endp = nextp;
3407  while (*nextp && *nextp != separator)
3408  {
3409  /* trailing whitespace should not be included in name */
3410  if (!isspace((unsigned char) *nextp))
3411  endp = nextp + 1;
3412  nextp++;
3413  }
3414  if (curname == endp)
3415  return false; /* empty unquoted name not allowed */
3416  }
3417 
3418  while (isspace((unsigned char) *nextp))
3419  nextp++; /* skip trailing whitespace */
3420 
3421  if (*nextp == separator)
3422  {
3423  nextp++;
3424  while (isspace((unsigned char) *nextp))
3425  nextp++; /* skip leading whitespace for next */
3426  /* we expect another name, so done remains false */
3427  }
3428  else if (*nextp == '\0')
3429  done = true;
3430  else
3431  return false; /* invalid syntax */
3432 
3433  /* Now safe to overwrite separator with a null */
3434  *endp = '\0';
3435 
3436  /* Truncate path if it's overlength */
3437  if (strlen(curname) >= MAXPGPATH)
3438  curname[MAXPGPATH - 1] = '\0';
3439 
3440  /*
3441  * Finished isolating current name --- add it to list
3442  */
3443  curname = pstrdup(curname);
3444  canonicalize_path(curname);
3445  *namelist = lappend(*namelist, curname);
3446 
3447  /* Loop back if we didn't reach end of string */
3448  } while (!done);
3449 
3450  return true;
3451 }
3452 
3453 
3454 /*****************************************************************************
3455  * Comparison Functions used for bytea
3456  *
3457  * Note: btree indexes need these routines not to leak memory; therefore,
3458  * be careful to free working copies of toasted datums. Most places don't
3459  * need to be so careful.
3460  *****************************************************************************/
3461 
3462 Datum
3464 {
3465  Datum arg1 = PG_GETARG_DATUM(0);
3466  Datum arg2 = PG_GETARG_DATUM(1);
3467  bool result;
3468  Size len1,
3469  len2;
3470 
3471  /*
3472  * We can use a fast path for unequal lengths, which might save us from
3473  * having to detoast one or both values.
3474  */
3475  len1 = toast_raw_datum_size(arg1);
3476  len2 = toast_raw_datum_size(arg2);
3477  if (len1 != len2)
3478  result = false;
3479  else
3480  {
3481  bytea *barg1 = DatumGetByteaPP(arg1);
3482  bytea *barg2 = DatumGetByteaPP(arg2);
3483 
3484  result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3485  len1 - VARHDRSZ) == 0);
3486 
3487  PG_FREE_IF_COPY(barg1, 0);
3488  PG_FREE_IF_COPY(barg2, 1);
3489  }
3490 
3491  PG_RETURN_BOOL(result);
3492 }
3493 
3494 Datum
3496 {
3497  Datum arg1 = PG_GETARG_DATUM(0);
3498  Datum arg2 = PG_GETARG_DATUM(1);
3499  bool result;
3500  Size len1,
3501  len2;
3502 
3503  /*
3504  * We can use a fast path for unequal lengths, which might save us from
3505  * having to detoast one or both values.
3506  */
3507  len1 = toast_raw_datum_size(arg1);
3508  len2 = toast_raw_datum_size(arg2);
3509  if (len1 != len2)
3510  result = true;
3511  else
3512  {
3513  bytea *barg1 = DatumGetByteaPP(arg1);
3514  bytea *barg2 = DatumGetByteaPP(arg2);
3515 
3516  result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3517  len1 - VARHDRSZ) != 0);
3518 
3519  PG_FREE_IF_COPY(barg1, 0);
3520  PG_FREE_IF_COPY(barg2, 1);
3521  }
3522 
3523  PG_RETURN_BOOL(result);
3524 }
3525 
3526 Datum
3528 {
3529  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3530  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3531  int len1,
3532  len2;
3533  int cmp;
3534 
3535  len1 = VARSIZE_ANY_EXHDR(arg1);
3536  len2 = VARSIZE_ANY_EXHDR(arg2);
3537 
3538  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3539 
3540  PG_FREE_IF_COPY(arg1, 0);
3541  PG_FREE_IF_COPY(arg2, 1);
3542 
3543  PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 < len2)));
3544 }
3545 
3546 Datum
3548 {
3549  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3550  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3551  int len1,
3552  len2;
3553  int cmp;
3554 
3555  len1 = VARSIZE_ANY_EXHDR(arg1);
3556  len2 = VARSIZE_ANY_EXHDR(arg2);
3557 
3558  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3559 
3560  PG_FREE_IF_COPY(arg1, 0);
3561  PG_FREE_IF_COPY(arg2, 1);
3562 
3563  PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 <= len2)));
3564 }
3565 
3566 Datum
3568 {
3569  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3570  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3571  int len1,
3572  len2;
3573  int cmp;
3574 
3575  len1 = VARSIZE_ANY_EXHDR(arg1);
3576  len2 = VARSIZE_ANY_EXHDR(arg2);
3577 
3578  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3579 
3580  PG_FREE_IF_COPY(arg1, 0);
3581  PG_FREE_IF_COPY(arg2, 1);
3582 
3583  PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 > len2)));
3584 }
3585 
3586 Datum
3588 {
3589  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3590  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3591  int len1,
3592  len2;
3593  int cmp;
3594 
3595  len1 = VARSIZE_ANY_EXHDR(arg1);
3596  len2 = VARSIZE_ANY_EXHDR(arg2);
3597 
3598  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3599 
3600  PG_FREE_IF_COPY(arg1, 0);
3601  PG_FREE_IF_COPY(arg2, 1);
3602 
3603  PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 >= len2)));
3604 }
3605 
3606 Datum
3608 {
3609  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3610  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3611  int len1,
3612  len2;
3613  int cmp;
3614 
3615  len1 = VARSIZE_ANY_EXHDR(arg1);
3616  len2 = VARSIZE_ANY_EXHDR(arg2);
3617 
3618  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3619  if ((cmp == 0) && (len1 != len2))
3620  cmp = (len1 < len2) ? -1 : 1;
3621 
3622  PG_FREE_IF_COPY(arg1, 0);
3623  PG_FREE_IF_COPY(arg2, 1);
3624 
3625  PG_RETURN_INT32(cmp);
3626 }
3627 
3628 Datum
3630 {
3632  MemoryContext oldcontext;
3633 
3634  oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
3635 
3636  /* Use generic string SortSupport, forcing "C" collation */
3637  varstr_sortsupport(ssup, C_COLLATION_OID, false);
3638 
3639  MemoryContextSwitchTo(oldcontext);
3640 
3641  PG_RETURN_VOID();
3642 }
3643 
3644 /*
3645  * appendStringInfoText
3646  *
3647  * Append a text to str.
3648  * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3649  */
3650 static void
3652 {
3654 }
3655 
3656 /*
3657  * replace_text
3658  * replace all occurrences of 'old_sub_str' in 'orig_str'
3659  * with 'new_sub_str' to form 'new_str'
3660  *
3661  * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3662  * otherwise returns 'new_str'
3663  */
3664 Datum
3666 {
3667  text *src_text = PG_GETARG_TEXT_PP(0);
3668  text *from_sub_text = PG_GETARG_TEXT_PP(1);
3669  text *to_sub_text = PG_GETARG_TEXT_PP(2);
3670  int src_text_len;
3671  int from_sub_text_len;
3673  text *ret_text;
3674  int start_posn;
3675  int curr_posn;
3676  int chunk_len;
3677  char *start_ptr;
3678  StringInfoData str;
3679 
3680  text_position_setup(src_text, from_sub_text, &state);
3681 
3682  /*
3683  * Note: we check the converted string length, not the original, because
3684  * they could be different if the input contained invalid encoding.
3685  */
3686  src_text_len = state.len1;
3687  from_sub_text_len = state.len2;
3688 
3689  /* Return unmodified source string if empty source or pattern */
3690  if (src_text_len < 1 || from_sub_text_len < 1)
3691  {
3692  text_position_cleanup(&state);
3693  PG_RETURN_TEXT_P(src_text);
3694  }
3695 
3696  start_posn = 1;
3697  curr_posn = text_position_next(1, &state);
3698 
3699  /* When the from_sub_text is not found, there is nothing to do. */
3700  if (curr_posn == 0)
3701  {
3702  text_position_cleanup(&state);
3703  PG_RETURN_TEXT_P(src_text);
3704  }
3705 
3706  /* start_ptr points to the start_posn'th character of src_text */
3707  start_ptr = VARDATA_ANY(src_text);
3708 
3709  initStringInfo(&str);
3710 
3711  do
3712  {
3714 
3715  /* copy the data skipped over by last text_position_next() */
3716  chunk_len = charlen_to_bytelen(start_ptr, curr_posn - start_posn);
3717  appendBinaryStringInfo(&str, start_ptr, chunk_len);
3718 
3719  appendStringInfoText(&str, to_sub_text);
3720 
3721  start_posn = curr_posn;
3722  start_ptr += chunk_len;
3723  start_posn += from_sub_text_len;
3724  start_ptr += charlen_to_bytelen(start_ptr, from_sub_text_len);
3725 
3726  curr_posn = text_position_next(start_posn, &state);
3727  }
3728  while (curr_posn > 0);
3729 
3730  /* copy trailing data */
3731  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3732  appendBinaryStringInfo(&str, start_ptr, chunk_len);
3733 
3734  text_position_cleanup(&state);
3735 
3736  ret_text = cstring_to_text_with_len(str.data, str.len);
3737  pfree(str.data);
3738 
3739  PG_RETURN_TEXT_P(ret_text);
3740 }
3741 
3742 /*
3743  * check_replace_text_has_escape_char
3744  *
3745  * check whether replace_text contains escape char.
3746  */
3747 static bool
3749 {
3750  const char *p = VARDATA_ANY(replace_text);
3751  const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3752 
3754  {
3755  for (; p < p_end; p++)
3756  {
3757  if (*p == '\\')
3758  return true;
3759  }
3760  }
3761  else
3762  {
3763  for (; p < p_end; p += pg_mblen(p))
3764  {
3765  if (*p == '\\')
3766  return true;
3767  }
3768  }
3769 
3770  return false;
3771 }
3772 
3773 /*
3774  * appendStringInfoRegexpSubstr
3775  *
3776  * Append replace_text to str, substituting regexp back references for
3777  * \n escapes. start_ptr is the start of the match in the source string,
3778  * at logical character position data_pos.
3779  */
3780 static void
3782  regmatch_t *pmatch,
3783  char *start_ptr, int data_pos)
3784 {
3785  const char *p = VARDATA_ANY(replace_text);
3786  const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3787  int eml = pg_database_encoding_max_length();
3788 
3789  for (;;)
3790  {
3791  const char *chunk_start = p;
3792  int so;
3793  int eo;
3794 
3795  /* Find next escape char. */
3796  if (eml == 1)
3797  {
3798  for (; p < p_end && *p != '\\'; p++)
3799  /* nothing */ ;
3800  }
3801  else
3802  {
3803  for (; p < p_end && *p != '\\'; p += pg_mblen(p))
3804  /* nothing */ ;
3805  }
3806 
3807  /* Copy the text we just scanned over, if any. */
3808  if (p > chunk_start)
3809  appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3810 
3811  /* Done if at end of string, else advance over escape char. */
3812  if (p >= p_end)
3813  break;
3814  p++;
3815 
3816  if (p >= p_end)
3817  {
3818  /* Escape at very end of input. Treat same as unexpected char */
3819  appendStringInfoChar(str, '\\');
3820  break;
3821  }
3822 
3823  if (*p >= '1' && *p <= '9')
3824  {
3825  /* Use the back reference of regexp. */
3826  int idx = *p - '0';
3827 
3828  so = pmatch[idx].rm_so;
3829  eo = pmatch[idx].rm_eo;
3830  p++;
3831  }
3832  else if (*p == '&')
3833  {
3834  /* Use the entire matched string. */
3835  so = pmatch[0].rm_so;
3836  eo = pmatch[0].rm_eo;
3837  p++;
3838  }
3839  else if (*p == '\\')
3840  {
3841  /* \\ means transfer one \ to output. */
3842  appendStringInfoChar(str, '\\');
3843  p++;
3844  continue;
3845  }
3846  else
3847  {
3848  /*
3849  * If escape char is not followed by any expected char, just treat
3850  * it as ordinary data to copy. (XXX would it be better to throw
3851  * an error?)
3852  */
3853  appendStringInfoChar(str, '\\');
3854  continue;
3855  }
3856 
3857  if (so != -1 && eo != -1)
3858  {
3859  /*
3860  * Copy the text that is back reference of regexp. Note so and eo
3861  * are counted in characters not bytes.
3862  */
3863  char *chunk_start;
3864  int chunk_len;
3865 
3866  Assert(so >= data_pos);
3867  chunk_start = start_ptr;
3868  chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3869  chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3870  appendBinaryStringInfo(str, chunk_start, chunk_len);
3871  }
3872  }
3873 }
3874 
3875 #define REGEXP_REPLACE_BACKREF_CNT 10
3876 
3877 /*
3878  * replace_text_regexp
3879  *
3880  * replace text that matches to regexp in src_text to replace_text.
3881  *
3882  * Note: to avoid having to include regex.h in builtins.h, we declare
3883  * the regexp argument as void *, but really it's regex_t *.
3884  */
3885 text *
3886 replace_text_regexp(text *src_text, void *regexp,
3887  text *replace_text, bool glob)
3888 {
3889  text *ret_text;
3890  regex_t *re = (regex_t *) regexp;
3891  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3894  pg_wchar *data;
3895  size_t data_len;
3896  int search_start;
3897  int data_pos;
3898  char *start_ptr;
3899  bool have_escape;
3900 
3901  initStringInfo(&buf);
3902 
3903  /* Convert data string to wide characters. */
3904  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3905  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3906 
3907  /* Check whether replace_text has escape char. */
3908  have_escape = check_replace_text_has_escape_char(replace_text);
3909 
3910  /* start_ptr points to the data_pos'th character of src_text */
3911  start_ptr = (char *) VARDATA_ANY(src_text);
3912  data_pos = 0;
3913 
3914  search_start = 0;
3915  while (search_start <= data_len)
3916  {
3917  int regexec_result;
3918 
3920 
3921  regexec_result = pg_regexec(re,
3922  data,
3923  data_len,
3924  search_start,
3925  NULL, /* no details */
3927  pmatch,
3928  0);
3929 
3930  if (regexec_result == REG_NOMATCH)
3931  break;
3932 
3933  if (regexec_result != REG_OKAY)
3934  {
3935  char errMsg[100];
3936 
3938  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3939  ereport(ERROR,
3940  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3941  errmsg("regular expression failed: %s", errMsg)));
3942  }
3943 
3944  /*
3945  * Copy the text to the left of the match position. Note we are given
3946  * character not byte indexes.
3947  */
3948  if (pmatch[0].rm_so - data_pos > 0)
3949  {
3950  int chunk_len;
3951 
3952  chunk_len = charlen_to_bytelen(start_ptr,
3953  pmatch[0].rm_so - data_pos);
3954  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3955 
3956  /*
3957  * Advance start_ptr over that text, to avoid multiple rescans of
3958  * it if the replace_text contains multiple back-references.
3959  */
3960  start_ptr += chunk_len;
3961  data_pos = pmatch[0].rm_so;
3962  }
3963 
3964  /*
3965  * Copy the replace_text. Process back references when the
3966  * replace_text has escape characters.
3967  */
3968  if (have_escape)
3969  appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3970  start_ptr, data_pos);
3971  else
3972  appendStringInfoText(&buf, replace_text);
3973 
3974  /* Advance start_ptr and data_pos over the matched text. */
3975  start_ptr += charlen_to_bytelen(start_ptr,
3976  pmatch[0].rm_eo - data_pos);
3977  data_pos = pmatch[0].rm_eo;
3978 
3979  /*
3980  * When global option is off, replace the first instance only.
3981  */
3982  if (!glob)
3983  break;
3984 
3985  /*
3986  * Advance search position. Normally we start the next search at the
3987  * end of the previous match; but if the match was of zero length, we
3988  * have to advance by one character, or we'd just find the same match
3989  * again.
3990  */
3991  search_start = data_pos;
3992  if (pmatch[0].rm_so == pmatch[0].rm_eo)
3993  search_start++;
3994  }
3995 
3996  /*
3997  * Copy the text to the right of the last match.
3998  */
3999  if (data_pos < data_len)
4000  {
4001  int chunk_len;
4002 
4003  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4004  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4005  }
4006 
4007  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4008  pfree(buf.data);
4009  pfree(data);
4010 
4011  return ret_text;
4012 }
4013 
4014 /*
4015  * split_text
4016  * parse input string
4017  * return ord item (1 based)
4018  * based on provided field separator
4019  */
4020 Datum
4022 {
4023  text *inputstring = PG_GETARG_TEXT_PP(0);
4024  text *fldsep = PG_GETARG_TEXT_PP(1);
4025  int fldnum = PG_GETARG_INT32(2);
4026  int inputstring_len;
4027  int fldsep_len;
4029  int start_posn;
4030  int end_posn;
4031  text *result_text;
4032 
4033  /* field number is 1 based */
4034  if (fldnum < 1)
4035  ereport(ERROR,
4036  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4037  errmsg("field position must be greater than zero")));
4038 
4039  text_position_setup(inputstring, fldsep, &state);
4040 
4041  /*
4042  * Note: we check the converted string length, not the original, because
4043  * they could be different if the input contained invalid encoding.
4044  */
4045  inputstring_len = state.len1;
4046  fldsep_len = state.len2;
4047 
4048  /* return empty string for empty input string */
4049  if (inputstring_len < 1)
4050  {
4051  text_position_cleanup(&state);
4053  }
4054 
4055  /* empty field separator */
4056  if (fldsep_len < 1)
4057  {
4058  text_position_cleanup(&state);
4059  /* if first field, return input string, else empty string */
4060  if (fldnum == 1)
4061  PG_RETURN_TEXT_P(inputstring);
4062  else
4064  }
4065 
4066  /* identify bounds of first field */
4067  start_posn = 1;
4068  end_posn = text_position_next(1, &state);
4069 
4070  /* special case if fldsep not found at all */
4071  if (end_posn == 0)
4072  {
4073  text_position_cleanup(&state);
4074  /* if field 1 requested, return input string, else empty string */
4075  if (fldnum == 1)
4076  PG_RETURN_TEXT_P(inputstring);
4077  else
4079  }
4080 
4081  while (end_posn > 0 && --fldnum > 0)
4082  {
4083  /* identify bounds of next field */
4084  start_posn = end_posn + fldsep_len;
4085  end_posn = text_position_next(start_posn, &state);
4086  }
4087 
4088  text_position_cleanup(&state);
4089 
4090  if (fldnum > 0)
4091  {
4092  /* N'th field separator not found */
4093  /* if last field requested, return it, else empty string */
4094  if (fldnum == 1)
4095  result_text = text_substring(PointerGetDatum(inputstring),
4096  start_posn,
4097  -1,
4098  true);
4099  else
4100  result_text = cstring_to_text("");
4101  }
4102  else
4103  {
4104  /* non-last field requested */
4105  result_text = text_substring(PointerGetDatum(inputstring),
4106  start_posn,
4107  end_posn - start_posn,
4108  false);
4109  }
4110 
4111  PG_RETURN_TEXT_P(result_text);
4112 }
4113 
4114 /*
4115  * Convenience function to return true when two text params are equal.
4116  */
4117 static bool
4118 text_isequal(text *txt1, text *txt2)
4119 {
4121  PointerGetDatum(txt1),
4122  PointerGetDatum(txt2)));
4123 }
4124 
4125 /*
4126  * text_to_array
4127  * parse input string and return text array of elements,
4128  * based on provided field separator
4129  */
4130 Datum
4132 {
4133  return text_to_array_internal(fcinfo);
4134 }
4135 
4136 /*
4137  * text_to_array_null
4138  * parse input string and return text array of elements,
4139  * based on provided field separator and null string
4140  *
4141  * This is a separate entry point only to prevent the regression tests from
4142  * complaining about different argument sets for the same internal function.
4143  */
4144 Datum
4146 {
4147  return text_to_array_internal(fcinfo);
4148 }
4149 
4150 /*
4151  * common code for text_to_array and text_to_array_null functions
4152  *
4153  * These are not strict so we have to test for null inputs explicitly.
4154  */
4155 static Datum
4157 {
4158  text *inputstring;
4159  text *fldsep;
4160  text *null_string;
4161  int inputstring_len;
4162  int fldsep_len;
4163  char *start_ptr;
4164  text *result_text;
4165  bool is_null;
4166  ArrayBuildState *astate = NULL;
4167 
4168  /* when input string is NULL, then result is NULL too */
4169  if (PG_ARGISNULL(0))
4170  PG_RETURN_NULL();
4171 
4172  inputstring = PG_GETARG_TEXT_PP(0);
4173 
4174  /* fldsep can be NULL */
4175  if (!PG_ARGISNULL(1))
4176  fldsep = PG_GETARG_TEXT_PP(1);
4177  else
4178  fldsep = NULL;
4179 
4180  /* null_string can be NULL or omitted */
4181  if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
4182  null_string = PG_GETARG_TEXT_PP(2);
4183  else
4184  null_string = NULL;
4185 
4186  if (fldsep != NULL)
4187  {
4188  /*
4189  * Normal case with non-null fldsep. Use the text_position machinery
4190  * to search for occurrences of fldsep.
4191  */
4193  int fldnum;
4194  int start_posn;
4195  int end_posn;
4196  int chunk_len;
4197 
4198  text_position_setup(inputstring, fldsep, &state);
4199 
4200  /*
4201  * Note: we check the converted string length, not the original,
4202  * because they could be different if the input contained invalid
4203  * encoding.
4204  */
4205  inputstring_len = state.len1;
4206  fldsep_len = state.len2;
4207 
4208  /* return empty array for empty input string */
4209  if (inputstring_len < 1)
4210  {
4211  text_position_cleanup(&state);
4213  }
4214 
4215  /*
4216  * empty field separator: return the input string as a one-element
4217  * array
4218  */
4219  if (fldsep_len < 1)
4220  {
4221  text_position_cleanup(&state);
4222  /* single element can be a NULL too */
4223  is_null = null_string ? text_isequal(inputstring, null_string) : false;
4225  PointerGetDatum(inputstring),
4226  is_null, 1));
4227  }
4228 
4229  start_posn = 1;
4230  /* start_ptr points to the start_posn'th character of inputstring */
4231  start_ptr = VARDATA_ANY(inputstring);
4232 
4233  for (fldnum = 1;; fldnum++) /* field number is 1 based */
4234  {
4236 
4237  end_posn = text_position_next(start_posn, &state);
4238 
4239  if (end_posn == 0)
4240  {
4241  /* fetch last field */
4242  chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
4243  }
4244  else
4245  {
4246  /* fetch non-last field */
4247  chunk_len = charlen_to_bytelen(start_ptr, end_posn - start_posn);
4248  }
4249 
4250  /* must build a temp text datum to pass to accumArrayResult */
4251  result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4252  is_null = null_string ? text_isequal(result_text, null_string) : false;
4253 
4254  /* stash away this field */
4255  astate = accumArrayResult(astate,
4256  PointerGetDatum(result_text),
4257  is_null,
4258  TEXTOID,
4260 
4261  pfree(result_text);
4262 
4263  if (end_posn == 0)
4264  break;
4265 
4266  start_posn = end_posn;
4267  start_ptr += chunk_len;
4268  start_posn += fldsep_len;
4269  start_ptr += charlen_to_bytelen(start_ptr, fldsep_len);
4270  }
4271 
4272  text_position_cleanup(&state);
4273  }
4274  else
4275  {
4276  /*
4277  * When fldsep is NULL, each character in the inputstring becomes an
4278  * element in the result array. The separator is effectively the
4279  * space between characters.
4280  */
4281  inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4282 
4283  /* return empty array for empty input string */
4284  if (inputstring_len < 1)
4286 
4287  start_ptr = VARDATA_ANY(inputstring);
4288 
4289  while (inputstring_len > 0)
4290  {
4291  int chunk_len = pg_mblen(start_ptr);
4292 
4294 
4295  /* must build a temp text datum to pass to accumArrayResult */
4296  result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4297  is_null = null_string ? text_isequal(result_text, null_string) : false;
4298 
4299  /* stash away this field */
4300  astate = accumArrayResult(astate,
4301  PointerGetDatum(result_text),
4302  is_null,
4303  TEXTOID,
4305 
4306  pfree(result_text);
4307 
4308  start_ptr += chunk_len;
4309  inputstring_len -= chunk_len;
4310  }
4311  }
4312 
4315 }
4316 
4317 /*
4318  * array_to_text
4319  * concatenate Cstring representation of input array elements
4320  * using provided field separator
4321  */
4322 Datum
4324 {
4326  char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4327 
4328  PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
4329 }
4330 
4331 /*
4332  * array_to_text_null
4333  * concatenate Cstring representation of input array elements
4334  * using provided field separator and null string
4335  *
4336  * This version is not strict so we have to test for null inputs explicitly.
4337  */
4338 Datum
4340 {
4341  ArrayType *v;
4342  char *fldsep;
4343  char *null_string;
4344 
4345  /* returns NULL when first or second parameter is NULL */
4346  if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
4347  PG_RETURN_NULL();
4348 
4349  v = PG_GETARG_ARRAYTYPE_P(0);
4350  fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4351 
4352  /* NULL null string is passed through as a null pointer */
4353  if (!PG_ARGISNULL(2))
4354  null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
4355  else
4356  null_string = NULL;
4357 
4358  PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
4359 }
4360 
4361 /*
4362  * common code for array_to_text and array_to_text_null functions
4363  */
4364 static text *
4366  const char *fldsep, const char *null_string)
4367 {
4368  text *result;
4369  int nitems,
4370  *dims,
4371  ndims;
4372  Oid element_type;
4373  int typlen;
4374  bool typbyval;
4375  char typalign;
4377  bool printed = false;
4378  char *p;
4379  bits8 *bitmap;
4380  int bitmask;
4381  int i;
4382  ArrayMetaState *my_extra;
4383 
4384  ndims = ARR_NDIM(v);
4385  dims = ARR_DIMS(v);
4386  nitems = ArrayGetNItems(ndims, dims);
4387 
4388  /* if there are no elements, return an empty string */
4389  if (nitems == 0)
4390  return cstring_to_text_with_len("", 0);
4391 
4392  element_type = ARR_ELEMTYPE(v);
4393  initStringInfo(&buf);
4394 
4395  /*
4396  * We arrange to look up info about element type, including its output
4397  * conversion proc, only once per series of calls, assuming the element
4398  * type doesn't change underneath us.
4399  */
4400  my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4401  if (my_extra == NULL)
4402  {
4403  fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4404  sizeof(ArrayMetaState));
4405  my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4406  my_extra->element_type = ~element_type;
4407  }
4408 
4409  if (my_extra->element_type != element_type)
4410  {
4411  /*
4412  * Get info about element type, including its output conversion proc
4413  */
4414  get_type_io_data(element_type, IOFunc_output,
4415  &my_extra->typlen, &my_extra->typbyval,
4416  &my_extra->typalign, &my_extra->typdelim,
4417  &my_extra->typioparam, &my_extra->typiofunc);
4418  fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
4419  fcinfo->flinfo->fn_mcxt);
4420  my_extra->element_type = element_type;
4421  }
4422  typlen = my_extra->typlen;
4423  typbyval = my_extra->typbyval;
4424  typalign = my_extra->typalign;
4425 
4426  p = ARR_DATA_PTR(v);
4427  bitmap = ARR_NULLBITMAP(v);
4428  bitmask = 1;
4429 
4430  for (i = 0; i < nitems; i++)
4431  {
4432  Datum itemvalue;
4433  char *value;
4434 
4435  /* Get source element, checking for NULL */
4436  if (bitmap && (*bitmap & bitmask) == 0)
4437  {
4438  /* if null_string is NULL, we just ignore null elements */
4439  if (null_string != NULL)
4440  {
4441  if (printed)
4442  appendStringInfo(&buf, "%s%s", fldsep, null_string);
4443  else
4444  appendStringInfoString(&buf, null_string);
4445  printed = true;
4446  }
4447  }
4448  else
4449  {
4450  itemvalue = fetch_att(p, typbyval, typlen);
4451 
4452  value = OutputFunctionCall(&my_extra->proc, itemvalue);
4453 
4454  if (printed)
4455  appendStringInfo(&buf, "%s%s", fldsep, value);
4456  else
4457  appendStringInfoString(&buf, value);
4458  printed = true;
4459 
4460  p = att_addlength_pointer(p, typlen, p);
4461  p = (char *) att_align_nominal(p, typalign);
4462  }
4463 
4464  /* advance bitmap pointer if any */
4465  if (bitmap)
4466  {
4467  bitmask <<= 1;
4468  if (bitmask == 0x100)
4469  {
4470  bitmap++;
4471  bitmask = 1;
4472  }
4473  }
4474  }
4475 
4476  result = cstring_to_text_with_len(buf.data, buf.len);
4477  pfree(buf.data);
4478 
4479  return result;
4480 }
4481 
4482 #define HEXBASE 16
4483 /*
4484  * Convert an int32 to a string containing a base 16 (hex) representation of
4485  * the number.
4486  */
4487 Datum
4489 {
4491  char *ptr;
4492  const char *digits = "0123456789abcdef";
4493  char buf[32]; /* bigger than needed, but reasonable */
4494 
4495  ptr = buf + sizeof(buf) - 1;
4496  *ptr = '\0';
4497 
4498  do
4499  {
4500  *--ptr = digits[value % HEXBASE];
4501  value /= HEXBASE;
4502  } while (ptr > buf && value);
4503 
4505 }
4506 
4507 /*
4508  * Convert an int64 to a string containing a base 16 (hex) representation of
4509  * the number.
4510  */
4511 Datum
4513 {
4514  uint64 value = (uint64) PG_GETARG_INT64(0);
4515  char *ptr;
4516  const char *digits = "0123456789abcdef";
4517  char buf[32]; /* bigger than needed, but reasonable */
4518 
4519  ptr = buf + sizeof(buf) - 1;
4520  *ptr = '\0';
4521 
4522  do
4523  {
4524  *--ptr = digits[value % HEXBASE];
4525  value /= HEXBASE;
4526  } while (ptr > buf && value);
4527 
4529 }
4530 
4531 /*
4532  * Create an md5 hash of a text string and return it as hex
4533  *
4534  * md5 produces a 16 byte (128 bit) hash; double it for hex
4535  */
4536 #define MD5_HASH_LEN 32
4537 
4538 Datum
4540 {
4541  text *in_text = PG_GETARG_TEXT_PP(0);
4542  size_t len;
4543  char hexsum[MD5_HASH_LEN + 1];
4544 
4545  /* Calculate the length of the buffer using varlena metadata */
4546  len = VARSIZE_ANY_EXHDR(in_text);
4547 
4548  /* get the hash result */
4549  if (pg_md5_hash(VARDATA_ANY(in_text), len, hexsum) == false)
4550  ereport(ERROR,
4551  (errcode(ERRCODE_OUT_OF_MEMORY),
4552  errmsg("out of memory")));
4553 
4554  /* convert to text and return it */
4556 }
4557 
4558 /*
4559  * Create an md5 hash of a bytea field and return it as a hex string:
4560  * 16-byte md5 digest is represented in 32 hex characters.
4561  */
4562 Datum
4564 {
4565  bytea *in = PG_GETARG_BYTEA_PP(0);
4566  size_t len;
4567  char hexsum[MD5_HASH_LEN + 1];
4568 
4569  len = VARSIZE_ANY_EXHDR(in);
4570  if (pg_md5_hash(VARDATA_ANY(in), len, hexsum) == false)
4571  ereport(ERROR,
4572  (errcode(ERRCODE_OUT_OF_MEMORY),
4573  errmsg("out of memory")));
4574 
4576 }
4577 
4578 /*
4579  * Return the size of a datum, possibly compressed
4580  *
4581  * Works on any data type
4582  */
4583 Datum
4585 {
4587  int32 result;
4588  int typlen;
4589 
4590  /* On first call, get the input type's typlen, and save at *fn_extra */
4591  if (fcinfo->flinfo->fn_extra == NULL)
4592  {
4593  /* Lookup the datatype of the supplied argument */
4594  Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4595 
4596  typlen = get_typlen(argtypeid);
4597  if (typlen == 0) /* should not happen */
4598  elog(ERROR, "cache lookup failed for type %u", argtypeid);
4599 
4600  fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4601  sizeof(int));
4602  *((int *) fcinfo->flinfo->fn_extra) = typlen;
4603  }
4604  else
4605  typlen = *((int *) fcinfo->flinfo->fn_extra);
4606 
4607  if (typlen == -1)
4608  {
4609  /* varlena type, possibly toasted */
4610  result = toast_datum_size(value);
4611  }
4612  else if (typlen == -2)
4613  {
4614  /* cstring */
4615  result = strlen(DatumGetCString(value)) + 1;
4616  }
4617  else
4618  {
4619  /* ordinary fixed-width type */
4620  result = typlen;
4621  }
4622 
4623  PG_RETURN_INT32(result);
4624 }
4625 
4626 /*
4627  * string_agg - Concatenates values and returns string.
4628  *
4629  * Syntax: string_agg(value text, delimiter text) RETURNS text
4630  *
4631  * Note: Any NULL values are ignored. The first-call delimiter isn't
4632  * actually used at all, and on subsequent calls the delimiter precedes
4633  * the associated value.
4634  */
4635 
4636 /* subroutine to initialize state */
4637 static StringInfo
4639 {
4640  StringInfo state;
4641  MemoryContext aggcontext;
4642  MemoryContext oldcontext;
4643 
4644  if (!AggCheckCallContext(fcinfo, &aggcontext))
4645  {
4646  /* cannot be called directly because of internal-type argument */
4647  elog(ERROR, "string_agg_transfn called in non-aggregate context");
4648  }
4649 
4650  /*
4651  * Create state in aggregate context. It'll stay there across subsequent
4652  * calls.
4653  */
4654  oldcontext = MemoryContextSwitchTo(aggcontext);
4655  state = makeStringInfo();
4656  MemoryContextSwitchTo(oldcontext);
4657 
4658  return state;
4659 }
4660 
4661 Datum
4663 {
4664  StringInfo state;
4665 
4666  state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4667 
4668  /* Append the value unless null. */
4669  if (!PG_ARGISNULL(1))
4670  {
4671  /* On the first time through, we ignore the delimiter. */
4672  if (state == NULL)
4673  state = makeStringAggState(fcinfo);
4674  else if (!PG_ARGISNULL(2))
4675  appendStringInfoText(state, PG_GETARG_TEXT_PP(2)); /* delimiter */
4676 
4677  appendStringInfoText(state, PG_GETARG_TEXT_PP(1)); /* value */
4678  }
4679 
4680  /*
4681  * The transition type for string_agg() is declared to be "internal",
4682  * which is a pass-by-value type the same size as a pointer.
4683  */
4684  PG_RETURN_POINTER(state);
4685 }
4686 
4687 Datum
4689 {
4690  StringInfo state;
4691 
4692  /* cannot be called directly because of internal-type argument */
4693  Assert(AggCheckCallContext(fcinfo, NULL));
4694 
4695  state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4696 
4697  if (state != NULL)
4699  else
4700  PG_RETURN_NULL();
4701 }
4702 
4703 /*
4704  * Implementation of both concat() and concat_ws().
4705  *
4706  * sepstr is the separator string to place between values.
4707  * argidx identifies the first argument to concatenate (counting from zero).
4708  * Returns NULL if result should be NULL, else text value.
4709  */
4710 static text *
4711 concat_internal(const char *sepstr, int argidx,
4712  FunctionCallInfo fcinfo)
4713 {
4714  text *result;
4715  StringInfoData str;
4716  bool first_arg = true;
4717  int i;
4718 
4719  /*
4720  * concat(VARIADIC some-array) is essentially equivalent to
4721  * array_to_text(), ie concat the array elements with the given separator.
4722  * So we just pass the case off to that code.
4723  */
4724  if (get_fn_expr_variadic(fcinfo->flinfo))
4725  {
4726  ArrayType *arr;
4727 
4728  /* Should have just the one argument */
4729  Assert(argidx == PG_NARGS() - 1);
4730 
4731  /* concat(VARIADIC NULL) is defined as NULL */
4732  if (PG_ARGISNULL(argidx))
4733  return NULL;
4734 
4735  /*
4736  * Non-null argument had better be an array. We assume that any call
4737  * context that could let get_fn_expr_variadic return true will have
4738  * checked that a VARIADIC-labeled parameter actually is an array. So
4739  * it should be okay to just Assert that it's an array rather than
4740  * doing a full-fledged error check.
4741  */
4743 
4744  /* OK, safe to fetch the array value */
4745  arr = PG_GETARG_ARRAYTYPE_P(argidx);
4746 
4747  /*
4748  * And serialize the array. We tell array_to_text to ignore null
4749  * elements, which matches the behavior of the loop below.
4750  */
4751  return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4752  }
4753 
4754  /* Normal case without explicit VARIADIC marker */
4755  initStringInfo(&str);
4756 
4757  for (i = argidx; i < PG_NARGS(); i++)
4758  {
4759  if (!PG_ARGISNULL(i))
4760  {
4762  Oid valtype;
4763  Oid typOutput;
4764  bool typIsVarlena;
4765 
4766  /* add separator if appropriate */
4767  if (first_arg)
4768  first_arg = false;
4769  else
4770  appendStringInfoString(&str, sepstr);
4771 
4772  /* call the appropriate type output function, append the result */
4773  valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4774  if (!OidIsValid(valtype))
4775  elog(ERROR, "could not determine data type of concat() input");
4776  getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4778  OidOutputFunctionCall(typOutput, value));
4779  }
4780  }
4781 
4782  result = cstring_to_text_with_len(str.data, str.len);
4783  pfree(str.data);
4784 
4785  return result;
4786 }
4787 
4788 /*
4789  * Concatenate all arguments. NULL arguments are ignored.
4790  */
4791 Datum
4793 {
4794  text *result;
4795 
4796  result = concat_internal("", 0, fcinfo);
4797  if (result == NULL)
4798  PG_RETURN_NULL();
4799  PG_RETURN_TEXT_P(result);
4800 }
4801 
4802 /*
4803  * Concatenate all but first argument value with separators. The first
4804  * parameter is used as the separator. NULL arguments are ignored.
4805  */
4806 Datum
4808 {
4809  char *sep;
4810  text *result;
4811 
4812  /* return NULL when separator is NULL */
4813  if (PG_ARGISNULL(0))
4814  PG_RETURN_NULL();
4816 
4817  result = concat_internal(sep, 1, fcinfo);
4818  if (result == NULL)
4819  PG_RETURN_NULL();
4820  PG_RETURN_TEXT_P(result);
4821 }
4822 
4823 /*
4824  * Return first n characters in the string. When n is negative,
4825  * return all but last |n| characters.
4826  */
4827 Datum
4829 {
4830  text *str = PG_GETARG_TEXT_PP(0);
4831  const char *p = VARDATA_ANY(str);
4832  int len = VARSIZE_ANY_EXHDR(str);
4833  int n = PG_GETARG_INT32(1);
4834  int rlen;
4835 
4836  if (n < 0)
4837  n = pg_mbstrlen_with_len(p, len) + n;
4838  rlen = pg_mbcharcliplen(p, len, n);
4839 
4841 }
4842 
4843 /*
4844  * Return last n characters in the string. When n is negative,
4845  * return all but first |n| characters.
4846  */
4847 Datum
4849 {
4850  text *str = PG_GETARG_TEXT_PP(0);
4851  const char *p = VARDATA_ANY(str);
4852  int len = VARSIZE_ANY_EXHDR(str);
4853  int n = PG_GETARG_INT32(1);
4854  int off;
4855 
4856  if (n < 0)
4857  n = -n;
4858  else
4859  n = pg_mbstrlen_with_len(p, len) - n;
4860  off = pg_mbcharcliplen(p, len, n);
4861 
4862  PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4863 }
4864 
4865 /*
4866  * Return reversed string
4867  */
4868 Datum
4870 {
4871  text *str = PG_GETARG_TEXT_PP(0);
4872  const char *p = VARDATA_ANY(str);
4873  int len = VARSIZE_ANY_EXHDR(str);
4874  const char *endp = p + len;
4875  text *result;
4876  char *dst;
4877 
4878  result = palloc(len + VARHDRSZ);
4879  dst = (char *) VARDATA(result) + len;
4880  SET_VARSIZE(result, len + VARHDRSZ);
4881 
4883  {
4884  /* multibyte version */
4885  while (p < endp)
4886  {
4887  int sz;
4888 
4889  sz = pg_mblen(p);
4890  dst -= sz;
4891  memcpy(dst, p, sz);
4892  p += sz;
4893  }
4894  }
4895  else
4896  {
4897  /* single byte version */
4898  while (p < endp)
4899  *(--dst) = *p++;
4900  }
4901 
4902  PG_RETURN_TEXT_P(result);
4903 }
4904 
4905 
4906 /*
4907  * Support macros for text_format()
4908  */
4909 #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4910 
4911 #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4912  do { \
4913  if (++(ptr) >= (end_ptr)) \
4914  ereport(ERROR, \
4915  (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4916  errmsg("unterminated format() type specifier"), \
4917  errhint("For a single \"%%\" use \"%%%%\"."))); \
4918  } while (0)
4919 
4920 /*
4921  * Returns a formatted string
4922  */
4923 Datum
4925 {
4926  text *fmt;
4927  StringInfoData str;
4928  const char *cp;
4929  const char *start_ptr;
4930  const char *end_ptr;
4931  text *result;
4932  int arg;
4933  bool funcvariadic;
4934  int nargs;
4935  Datum *elements = NULL;
4936  bool *nulls = NULL;
4937  Oid element_type = InvalidOid;
4938  Oid prev_type = InvalidOid;
4939  Oid prev_width_type = InvalidOid;
4940  FmgrInfo typoutputfinfo;
4941  FmgrInfo typoutputinfo_width;
4942 
4943  /* When format string is null, immediately return null */
4944  if (PG_ARGISNULL(0))
4945  PG_RETURN_NULL();
4946 
4947  /* If argument is marked VARIADIC, expand array into elements */
4948  if (get_fn_expr_variadic(fcinfo->flinfo))
4949  {
4950  ArrayType *arr;
4951  int16 elmlen;
4952  bool elmbyval;
4953  char elmalign;
4954  int nitems;
4955 
4956  /* Should have just the one argument */
4957  Assert(PG_NARGS() == 2);
4958 
4959  /* If argument is NULL, we treat it as zero-length array */
4960  if (PG_ARGISNULL(1))
4961  nitems = 0;
4962  else
4963  {
4964  /*
4965  * Non-null argument had better be an array. We assume that any
4966  * call context that could let get_fn_expr_variadic return true
4967  * will have checked that a VARIADIC-labeled parameter actually is
4968  * an array. So it should be okay to just Assert that it's an
4969  * array rather than doing a full-fledged error check.
4970  */
4972 
4973  /* OK, safe to fetch the array value */
4974  arr = PG_GETARG_ARRAYTYPE_P(1);
4975 
4976  /* Get info about array element type */
4977  element_type = ARR_ELEMTYPE(arr);
4978  get_typlenbyvalalign(element_type,
4979  &elmlen, &elmbyval, &elmalign);
4980 
4981  /* Extract all array elements */
4982  deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
4983  &elements, &nulls, &nitems);
4984  }
4985 
4986  nargs = nitems + 1;
4987  funcvariadic = true;
4988  }
4989  else
4990  {
4991  /* Non-variadic case, we'll process the arguments individually */
4992  nargs = PG_NARGS();
4993  funcvariadic = false;
4994  }
4995 
4996  /* Setup for main loop. */
4997  fmt = PG_GETARG_TEXT_PP(0);
4998  start_ptr = VARDATA_ANY(fmt);
4999  end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
5000  initStringInfo(&str);
5001  arg = 1; /* next argument position to print */
5002 
5003  /* Scan format string, looking for conversion specifiers. */
5004  for (cp = start_ptr; cp < end_ptr; cp++)
5005  {
5006  int argpos;
5007  int widthpos;
5008  int flags;
5009  int width;
5010  Datum value;
5011  bool isNull;
5012  Oid typid;
5013 
5014  /*
5015  * If it's not the start of a conversion specifier, just copy it to
5016  * the output buffer.
5017  */
5018  if (*cp != '%')
5019  {
5020  appendStringInfoCharMacro(&str, *cp);
5021  continue;
5022  }
5023 
5024  ADVANCE_PARSE_POINTER(cp, end_ptr);
5025 
5026  /* Easy case: %% outputs a single % */
5027  if (*cp == '%')
5028  {
5029  appendStringInfoCharMacro(&str, *cp);
5030  continue;
5031  }
5032 
5033  /* Parse the optional portions of the format specifier */
5034  cp = text_format_parse_format(cp, end_ptr,
5035  &argpos, &widthpos,
5036  &flags, &width);
5037 
5038  /*
5039  * Next we should see the main conversion specifier. Whether or not
5040  * an argument position was present, it's known that at least one
5041  * character remains in the string at this point. Experience suggests
5042  * that it's worth checking that that character is one of the expected
5043  * ones before we try to fetch arguments, so as to produce the least
5044  * confusing response to a mis-formatted specifier.
5045  */
5046  if (strchr("sIL", *cp) == NULL)
5047  ereport(ERROR,
5048  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5049  errmsg("unrecognized format() type specifier \"%c\"",
5050  *cp),
5051  errhint("For a single \"%%\" use \"%%%%\".")));
5052 
5053  /* If indirect width was specified, get its value */
5054  if (widthpos >= 0)
5055  {
5056  /* Collect the specified or next argument position */
5057  if (widthpos > 0)
5058  arg = widthpos;
5059  if (arg >= nargs)
5060  ereport(ERROR,
5061  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5062  errmsg("too few arguments for format()")));
5063 
5064  /* Get the value and type of the selected argument */
5065  if (!funcvariadic)
5066  {
5067  value = PG_GETARG_DATUM(arg);
5068  isNull = PG_ARGISNULL(arg);
5069  typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5070  }
5071  else
5072  {
5073  value = elements[arg - 1];
5074  isNull = nulls[arg - 1];
5075  typid = element_type;
5076  }
5077  if (!OidIsValid(typid))
5078  elog(ERROR, "could not determine data type of format() input");
5079 
5080  arg++;
5081 
5082  /* We can treat NULL width the same as zero */
5083  if (isNull)
5084  width = 0;
5085  else if (typid == INT4OID)
5086  width = DatumGetInt32(value);
5087  else if (typid == INT2OID)
5088  width = DatumGetInt16(value);
5089  else
5090  {
5091  /* For less-usual datatypes, convert to text then to int */
5092  char *str;
5093 
5094  if (typid != prev_width_type)
5095  {
5096  Oid typoutputfunc;
5097  bool typIsVarlena;
5098 
5099  getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5100  fmgr_info(typoutputfunc, &typoutputinfo_width);
5101  prev_width_type = typid;
5102  }
5103 
5104  str = OutputFunctionCall(&typoutputinfo_width, value);
5105 
5106  /* pg_atoi will complain about bad data or overflow */
5107  width = pg_atoi(str, sizeof(int), '\0');
5108 
5109  pfree(str);
5110  }
5111  }
5112 
5113  /* Collect the specified or next argument position */
5114  if (argpos > 0)
5115  arg = argpos;
5116  if (arg >= nargs)
5117  ereport(ERROR,
5118  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5119  errmsg("too few arguments for format()")));
5120 
5121  /* Get the value and type of the selected argument */
5122  if (!funcvariadic)
5123  {
5124  value = PG_GETARG_DATUM(arg);
5125  isNull = PG_ARGISNULL(arg);
5126  typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5127  }
5128  else
5129  {
5130  value = elements[arg - 1];
5131  isNull = nulls[arg - 1];
5132  typid = element_type;
5133  }
5134  if (!OidIsValid(typid))
5135  elog(ERROR, "could not determine data type of format() input");
5136 
5137  arg++;
5138 
5139  /*
5140  * Get the appropriate typOutput function, reusing previous one if
5141  * same type as previous argument. That's particularly useful in the
5142  * variadic-array case, but often saves work even for ordinary calls.
5143  */
5144  if (typid != prev_type)
5145  {
5146  Oid typoutputfunc;
5147  bool typIsVarlena;
5148 
5149  getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5150  fmgr_info(typoutputfunc, &typoutputfinfo);
5151  prev_type = typid;
5152  }
5153 
5154  /*
5155  * And now we can format the value.
5156  */
5157  switch (*cp)
5158  {
5159  case 's':
5160  case 'I':
5161  case 'L':
5162  text_format_string_conversion(&str, *cp, &typoutputfinfo,
5163  value, isNull,
5164  flags, width);
5165  break;
5166  default:
5167  /* should not get here, because of previous check */
5168  ereport(ERROR,
5169  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5170  errmsg("unrecognized format() type specifier \"%c\"",
5171  *cp),
5172  errhint("For a single \"%%\" use \"%%%%\".")));
5173  break;
5174  }
5175  }
5176 
5177  /* Don't need deconstruct_array results anymore. */
5178  if (elements != NULL)
5179  pfree(elements);
5180  if (nulls != NULL)
5181  pfree(nulls);
5182 
5183  /* Generate results. */
5184  result = cstring_to_text_with_len(str.data, str.len);
5185  pfree(str.data);
5186 
5187  PG_RETURN_TEXT_P(result);
5188 }
5189 
5190 /*
5191  * Parse contiguous digits as a decimal number.
5192  *
5193  * Returns true if some digits could be parsed.
5194  * The value is returned into *value, and *ptr is advanced to the next
5195  * character to be parsed.
5196  *
5197  * Note parsing invariant: at least one character is known available before
5198  * string end (end_ptr) at entry, and this is still true at exit.
5199  */
5200 static bool
5201 text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5202 {
5203  bool found = false;
5204  const char *cp = *ptr;
5205  int val = 0;
5206 
5207  while (*cp >= '0' && *cp <= '9')
5208  {
5209  int newval = val * 10 + (*cp - '0');
5210 
5211  if (newval / 10 != val) /* overflow? */
5212  ereport(ERROR,
5213  (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5214  errmsg("number is out of range")));
5215  val = newval;
5216  ADVANCE_PARSE_POINTER(cp, end_ptr);
5217  found = true;
5218  }
5219 
5220  *ptr = cp;
5221  *value = val;
5222 
5223  return found;
5224 }
5225 
5226 /*
5227  * Parse a format specifier (generally following the SUS printf spec).
5228  *
5229  * We have already advanced over the initial '%', and we are looking for
5230  * [argpos][flags][width]type (but the type character is not consumed here).
5231  *
5232  * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5233  * Output parameters:
5234  * argpos: argument position for value to be printed. -1 means unspecified.
5235  * widthpos: argument position for width. Zero means the argument position
5236  * was unspecified (ie, take the next arg) and -1 means no width
5237  * argument (width was omitted or specified as a constant).
5238  * flags: bitmask of flags.
5239  * width: directly-specified width value. Zero means the width was omitted
5240  * (note it's not necessary to distinguish this case from an explicit
5241  * zero width value).
5242  *
5243  * The function result is the next character position to be parsed, ie, the
5244  * location where the type character is/should be.
5245  *
5246  * Note parsing invariant: at least one character is known available before
5247  * string end (end_ptr) at entry, and this is still true at exit.
5248  */
5249 static const char *
5250 text_format_parse_format(const char *start_ptr, const char *end_ptr,
5251  int *argpos, int *widthpos,
5252  int *flags, int *width)
5253 {
5254  const char *cp = start_ptr;
5255  int n;
5256 
5257  /* set defaults for output parameters */
5258  *argpos = -1;
5259  *widthpos = -1;
5260  *flags = 0;
5261  *width = 0;
5262 
5263  /* try to identify first number */
5264  if (text_format_parse_digits(&cp, end_ptr, &n))
5265  {
5266  if (*cp != '$')
5267  {
5268  /* Must be just a width and a type, so we're done */
5269  *width = n;
5270  return cp;
5271  }
5272  /* The number was argument position */
5273  *argpos = n;
5274  /* Explicit 0 for argument index is immediately refused */
5275  if (n == 0)
5276  ereport(ERROR,
5277  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5278  errmsg("format specifies argument 0, but arguments are numbered from 1")));
5279  ADVANCE_PARSE_POINTER(cp, end_ptr);
5280  }
5281 
5282  /* Handle flags (only minus is supported now) */
5283  while (*cp == '-')
5284  {
5285  *flags |= TEXT_FORMAT_FLAG_MINUS;
5286  ADVANCE_PARSE_POINTER(cp, end_ptr);
5287  }
5288 
5289  if (*cp == '*')
5290  {
5291  /* Handle indirect width */
5292  ADVANCE_PARSE_POINTER(cp, end_ptr);
5293  if (text_format_parse_digits(&cp, end_ptr, &n))
5294  {
5295  /* number in this position must be closed by $ */
5296  if (*cp != '$')
5297  ereport(ERROR,
5298  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5299  errmsg("width argument position must be ended by \"$\"")));
5300  /* The number was width argument position */
5301  *widthpos = n;
5302  /* Explicit 0 for argument index is immediately refused */
5303  if (n == 0)
5304  ereport(ERROR,
5305  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5306  errmsg("format specifies argument 0, but arguments are numbered from 1")));
5307  ADVANCE_PARSE_POINTER(cp, end_ptr);
5308  }
5309  else
5310  *widthpos = 0; /* width's argument position is unspecified */
5311  }
5312  else
5313  {
5314  /* Check for direct width specification */
5315  if (text_format_parse_digits(&cp, end_ptr, &n))
5316  *width = n;
5317  }
5318 
5319  /* cp should now be pointing at type character */
5320  return cp;
5321 }
5322 
5323 /*
5324  * Format a %s, %I, or %L conversion
5325  */
5326 static void
5328  FmgrInfo *typOutputInfo,
5329  Datum value, bool isNull,
5330  int flags, int width)
5331 {
5332  char *str;
5333 
5334  /* Handle NULL arguments before trying to stringify the value. */
5335  if (isNull)
5336  {
5337  if (conversion == 's')
5338  text_format_append_string(buf, "", flags, width);
5339  else if (conversion == 'L')
5340  text_format_append_string(buf, "NULL", flags, width);
5341  else if (conversion == 'I')
5342  ereport(ERROR,
5343  (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5344  errmsg("null values cannot be formatted as an SQL identifier")));
5345  return;
5346  }
5347 
5348  /* Stringify. */
5349  str = OutputFunctionCall(typOutputInfo, value);
5350 
5351  /* Escape. */
5352  if (conversion == 'I')
5353  {
5354  /* quote_identifier may or may not allocate a new string. */
5355  text_format_append_string(buf, quote_identifier(str), flags, width);
5356  }
5357  else if (conversion == 'L')
5358  {
5359  char *qstr = quote_literal_cstr(str);
5360 
5361  text_format_append_string(buf, qstr, flags, width);
5362  /* quote_literal_cstr() always allocates a new string */
5363  pfree(qstr);
5364  }
5365  else
5366  text_format_append_string(buf, str, flags, width);
5367 
5368  /* Cleanup. */
5369  pfree(str);
5370 }
5371 
5372 /*
5373  * Append str to buf, padding as directed by flags/width
5374  */
5375 static void
5377  int flags, int width)
5378 {
5379  bool align_to_left = false;
5380  int len;
5381 
5382  /* fast path for typical easy case */
5383  if (width == 0)
5384  {
5385  appendStringInfoString(buf, str);
5386  return;
5387  }
5388 
5389  if (width < 0)
5390  {
5391  /* Negative width: implicit '-' flag, then take absolute value */
5392  align_to_left = true;
5393  /* -INT_MIN is undefined */
5394  if (width <= INT_MIN)
5395  ereport(ERROR,
5396  (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5397  errmsg("number is out of range")));
5398  width = -width;
5399  }
5400  else if (flags & TEXT_FORMAT_FLAG_MINUS)
5401  align_to_left = true;
5402 
5403  len = pg_mbstrlen(str);
5404  if (align_to_left)
5405  {
5406  /* left justify */
5407  appendStringInfoString(buf, str);
5408  if (len < width)
5409  appendStringInfoSpaces(buf, width - len);
5410  }
5411  else
5412  {
5413  /* right justify */
5414  if (len < width)
5415  appendStringInfoSpaces(buf, width - len);
5416  appendStringInfoString(buf, str);
5417  }
5418 }
5419 
5420 /*
5421  * text_format_nv - nonvariadic wrapper for text_format function.
5422  *
5423  * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5424  * which checks that all built-in functions that share the implementing C
5425  * function take the same number of arguments.
5426  */
5427 Datum
5429 {
5430  return text_format(fcinfo);
5431 }
5432 
5433 /*
5434  * Helper function for Levenshtein distance functions. Faster than memcmp(),
5435  * for this use case.
5436  */
5437 static inline bool
5438 rest_of_char_same(const char *s1, const char *s2, int len)
5439 {
5440  while (len > 0)
5441  {
5442  len--;
5443  if (s1[len] != s2[len])
5444  return false;
5445  }
5446  return true;
5447 }
5448 
5449 /* Expand each Levenshtein distance variant */
5450 #include "levenshtein.c"
5451 #define LEVENSHTEIN_LESS_EQUAL
5452 #include "levenshtein.c"
Datum bttext_pattern_cmp(PG_FUNCTION_ARGS)
Definition: varlena.c:2686
#define PG_CACHE_LINE_SIZE
Datum text_to_array(PG_FUNCTION_ARGS)
Definition: varlena.c:4131
Datum bytea_substr_no_len(PG_FUNCTION_ARGS)
Definition: varlena.c:2822
struct SortSupportData * SortSupport
Definition: sortsupport.h:58
static int varstrfastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2024
Value * makeString(char *str)
Definition: value.c:53
#define COLLPROVIDER_ICU
Definition: pg_collation.h:84
signed short int16
Definition: c.h:255
int(* comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:107
Datum byteaout(PG_FUNCTION_ARGS)
Definition: varlena.c:351
#define PG_RETURN_POINTER(x)
Definition: fmgr.h:321
#define DatumGetUInt32(X)
Definition: postgres.h:492
#define NIL
Definition: pg_list.h:69
Datum text_format(PG_FUNCTION_ARGS)
Definition: varlena.c:4924
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:1991
int length(const List *list)
Definition: list.c:1271
#define PG_GETARG_INT32(n)
Definition: fmgr.h:234
#define ADVANCE_PARSE_POINTER(ptr, end_ptr)
Definition: varlena.c:4911
Definition: fmgr.h:56
text * replace_text_regexp(text *src_text, void *regexp, text *replace_text, bool glob)
Definition: varlena.c:3886
#define VARATT_IS_COMPRESSED(PTR)
Definition: postgres.h:313
Datum byteaSetBit(PG_FUNCTION_ARGS)
Definition: varlena.c:3092
int pg_mbcharcliplen(const char *mbstr, int len, int limit)
Definition: mbutils.c:873
Datum split_text(PG_FUNCTION_ARGS)
Definition: varlena.c:4021
int errhint(const char *fmt,...)
Definition: elog.c:987
Datum textoverlay_no_len(PG_FUNCTION_ARGS)
Definition: varlena.c:1022
void getTypeOutputInfo(Oid type, Oid *typOutput, bool *typIsVarlena)
Definition: lsyscache.c:2632
#define VARDATA_ANY(PTR)
Definition: postgres.h:347
#define VARDATA(PTR)
Definition: postgres.h:303
char * quote_literal_cstr(const char *rawstr)
Definition: quote.c:102
MemoryContext fn_mcxt
Definition: fmgr.h:65
#define att_align_nominal(cur_offset, attalign)
Definition: tupmacs.h:144
#define MD5_HASH_LEN
Definition: varlena.c:4536
const char * quote_identifier(const char *ident)
Definition: ruleutils.c:10246
Datum text_lt(PG_FUNCTION_ARGS)
Definition: varlena.c:1700
Datum text_pattern_le(PG_FUNCTION_ARGS)
Definition: varlena.c:2638
#define DatumGetTextPSlice(X, m, n)
Definition: fmgr.h:268
#define DatumGetInt32(X)
Definition: postgres.h:478
Datum text_pattern_gt(PG_FUNCTION_ARGS)
Definition: varlena.c:2670
#define HEXBASE
Definition: varlena.c:4482
#define TEXTOID
Definition: pg_type.h:324
#define VARSIZE(PTR)
Definition: postgres.h:304
Datum replace_text(PG_FUNCTION_ARGS)
Definition: varlena.c:3665
Datum byteagt(PG_FUNCTION_ARGS)
Definition: varlena.c:3567
static void text_format_string_conversion(StringInfo buf, char conversion, FmgrInfo *typOutputInfo, Datum value, bool isNull, int flags, int width)
Definition: varlena.c:5327
void get_typlenbyvalalign(Oid typid, int16 *typlen, bool *typbyval, char *typalign)
Definition: lsyscache.c:2021
#define PointerGetDatum(X)
Definition: postgres.h:562
char * downcase_truncate_identifier(const char *ident, int len, bool warn)
Definition: scansup.c:131
Datum textrecv(PG_FUNCTION_ARGS)
Definition: varlena.c:530
#define PG_GETARG_DATUM(n)
Definition: fmgr.h:233
static void text_position_setup(text *t1, text *t2, TextPositionState *state)
Definition: varlena.c:1119
void pq_begintypsend(StringInfo buf)
Definition: pqformat.c:359
#define VARHDRSZ
Definition: c.h:445
Datum md5_bytea(PG_FUNCTION_ARGS)
Definition: varlena.c:4563
char * pstrdup(const char *in)
Definition: mcxt.c:1077
Datum textout(PG_FUNCTION_ARGS)
Definition: varlena.c:519
regoff_t rm_so
Definition: regex.h:85
#define DatumGetTextPP(X)
Definition: fmgr.h:256
StringInfo makeStringInfo(void)
Definition: stringinfo.c:29
StringInfoData * StringInfo
Definition: stringinfo.h:46
#define Min(x, y)
Definition: c.h:806
int ArrayGetNItems(int ndim, const int *dims)
Definition: arrayutils.c:75
#define PG_GETARG_BYTEA_P_COPY(n)
Definition: fmgr.h:278
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2222
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2197
#define PG_RETURN_INT32(x)
Definition: fmgr.h:314
#define DatumGetByteaPSlice(X, m, n)
Definition: fmgr.h:267
static bytea * bytea_catenate(bytea *t1, bytea *t2)
Definition: varlena.c:2756
#define INT4OID
Definition: pg_type.h:316
void canonicalize_path(char *path)
Definition: path.c:254
bool get_fn_expr_variadic(FmgrInfo *flinfo)
Definition: fmgr.c:2051
int errcode(int sqlerrcode)
Definition: elog.c:575
void pq_sendtext(StringInfo buf, const char *str, int slen)
Definition: pqformat.c:163
Datum idx(PG_FUNCTION_ARGS)
Definition: _int_op.c:264
return result
Definition: formatting.c:1618
#define DatumGetByteaPP(X)
Definition: fmgr.h:255
#define PG_GETARG_POINTER(n)
Definition: fmgr.h:241
Datum byteaSetByte(PG_FUNCTION_ARGS)
Definition: varlena.c:3060
pg_wchar * wstr2
Definition: varlena.c:52
union pg_locale_struct::@120 info
Datum bytea_string_agg_finalfn(PG_FUNCTION_ARGS)
Definition: varlena.c:482
Datum string_agg_transfn(PG_FUNCTION_ARGS)
Definition: varlena.c:4662
Datum md5_text(PG_FUNCTION_ARGS)
Definition: varlena.c:4539
#define PG_RETURN_BYTEA_P(x)
Definition: fmgr.h:329
ArrayType * construct_empty_array(Oid elmtype)
Definition: arrayfuncs.c:3424
static bytea * bytea_substring(Datum str, int S, int L, bool length_not_specified)
Definition: varlena.c:2831
#define LOG
Definition: elog.h:26
unsigned int Oid
Definition: postgres_ext.h:31
Datum byteaeq(PG_FUNCTION_ARGS)
Definition: varlena.c:3463
Datum textlen(PG_FUNCTION_ARGS)
Definition: varlena.c:622
#define OidIsValid(objectId)
Definition: c.h:538
Datum bttextsortsupport(PG_FUNCTION_ARGS)
Definition: varlena.c:1775
bytea * pq_endtypsend(StringInfo buf)
Definition: pqformat.c:379
unsigned hex_decode(const char *src, unsigned len, char *dst)
Definition: encode.c:156
void text_to_cstring_buffer(const text *src, char *dst, size_t dst_len)
Definition: varlena.c:213
static text * text_overlay(text *t1, text *t2, int sp, int sl)
Definition: varlena.c:1034
bool trace_sort
Definition: tuplesort.c:155
#define PG_GET_COLLATION()
Definition: fmgr.h:163
Datum byteaoverlay_no_len(PG_FUNCTION_ARGS)
Definition: varlena.c:2901
Datum text_concat(PG_FUNCTION_ARGS)
Definition: varlena.c:4792
Datum textoctetlen(PG_FUNCTION_ARGS)
Definition: varlena.c:660
static void text_format_append_string(StringInfo buf, const char *str, int flags, int width)
Definition: varlena.c:5376
Datum array_to_text_null(PG_FUNCTION_ARGS)
Definition: varlena.c:4339
Datum text_concat_ws(PG_FUNCTION_ARGS)
Definition: varlena.c:4807
regoff_t rm_eo
Definition: regex.h:86
signed int int32
Definition: c.h:256
#define PG_STR_GET_BYTEA(str_)
Definition: varlena.c:2789
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:1954
#define PG_GETARG_TEXT_PP(n)
Definition: fmgr.h:273
char * OutputFunctionCall(FmgrInfo *flinfo, Datum val)
Definition: fmgr.c:1667
static int32 text_length(Datum str)
Definition: varlena.c:640
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:805
bool typbyval
Definition: array.h:221
#define NAMEDATALEN
void truncate_identifier(char *ident, int len, bool warn)
Definition: scansup.c:187
Datum to_hex64(PG_FUNCTION_ARGS)
Definition: varlena.c:4512
#define VARATT_IS_EXTERNAL(PTR)
Definition: postgres.h:314
static Datum text_to_array_internal(PG_FUNCTION_ARGS)
Definition: varlena.c:4156
Datum bytealt(PG_FUNCTION_ARGS)
Definition: varlena.c:3527
bool SplitDirectoriesString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3365
FmgrInfo * flinfo
Definition: fmgr.h:79
#define PG_GETARG_ARRAYTYPE_P(n)
Definition: array.h:244
#define wcscoll_l
Definition: win32.h:348
#define appendStringInfoCharMacro(str, ch)
Definition: stringinfo.h:135
void initHyperLogLog(hyperLogLogState *cState, uint8 bwidth)
Definition: hyperloglog.c:65
unsigned hex_encode(const char *src, unsigned len, char *dst)
Definition: encode.c:126
Datum array_to_text(PG_FUNCTION_ARGS)
Definition: varlena.c:4323
void pfree(void *pointer)
Definition: mcxt.c:950
Size toast_raw_datum_size(Datum value)
Definition: tuptoaster.c:353
#define REG_OKAY
Definition: regex.h:137
void appendStringInfo(StringInfo str, const char *fmt,...)
Definition: stringinfo.c:110
Datum string_agg_finalfn(PG_FUNCTION_ARGS)
Definition: varlena.c:4688
Datum textoverlay(PG_FUNCTION_ARGS)
Definition: varlena.c:1011
#define ERROR
Definition: elog.h:43
char * s1
static bool check_replace_text_has_escape_char(const text *replace_text)
Definition: varlena.c:3748
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1128
#define DatumGetCString(X)
Definition: postgres.h:572
Size toast_datum_size(Datum value)
Definition: tuptoaster.c:409
Oid get_fn_expr_argtype(FmgrInfo *flinfo, int argnum)
Definition: fmgr.c:1909
Datum byteage(PG_FUNCTION_ARGS)
Definition: varlena.c:3587
#define ARR_DIMS(a)
Definition: array.h:275
void fmgr_info(Oid functionId, FmgrInfo *finfo)
Definition: fmgr.c:127
MemoryContext ssup_cxt
Definition: sortsupport.h:66
struct varlena * pg_detoast_datum_packed(struct varlena *datum)
Definition: fmgr.c:1863
static int text_position_next(int start_pos, TextPositionState *state)
Definition: varlena.c:1231
Datum text_to_array_null(PG_FUNCTION_ARGS)
Definition: varlena.c:4145
#define MAXPGPATH
int pg_mbcliplen(const char *mbstr, int len, int limit)
Definition: mbutils.c:831
Datum byteain(PG_FUNCTION_ARGS)
Definition: varlena.c:255
static int charlen_to_bytelen(const char *p, int n)
Definition: varlena.c:735
static text * text_substring(Datum str, int32 start, int32 length, bool length_not_specified)
Definition: varlena.c:814
Datum unknownrecv(PG_FUNCTION_ARGS)
Definition: varlena.c:587
static text * array_to_text_internal(FunctionCallInfo fcinfo, ArrayType *v, const char *fldsep, const char *null_string)
Definition: varlena.c:4365
Definition: c.h:493
static void appendStringInfoText(StringInfo str, const text *t)
Definition: varlena.c:3651
Datum text_larger(PG_FUNCTION_ARGS)
Definition: varlena.c:2567
#define INT2OID
Definition: pg_type.h:308
Datum texteq(PG_FUNCTION_ARGS)
Definition: varlena.c:1636
void appendStringInfoString(StringInfo str, const char *s)
Definition: stringinfo.c:189
#define ARR_DATA_PTR(a)
Definition: array.h:303
hyperLogLogState abbr_card
Definition: varlena.c:73
Datum text_smaller(PG_FUNCTION_ARGS)
Definition: varlena.c:2579
Datum textne(PG_FUNCTION_ARGS)
Definition: varlena.c:1671
int16 typlen
Definition: array.h:220
pg_locale_t locale
Definition: varlena.c:76
static char * buf
Definition: pg_test_fsync.c:66
#define DatumBigEndianToNative(x)
Definition: pg_bswap.h:65
#define memmove(d, s, c)
Definition: c.h:1058
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:161
bool SplitIdentifierString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3240
#define DEFAULT_COLLATION_OID
Definition: pg_collation.h:74
char typdelim
Definition: array.h:223
int pg_database_encoding_max_length(void)
Definition: wchar.c:1833
static bytea * bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
Definition: varlena.c:2913
Datum text_name(PG_FUNCTION_ARGS)
Definition: varlena.c:3142
static text * text_catenate(text *t1, text *t2)
Definition: varlena.c:694
#define DatumGetInt16(X)
Definition: postgres.h:450
#define DatumGetBool(X)
Definition: postgres.h:399
void px(PlannerInfo *root, Gene *tour1, Gene *tour2, Gene *offspring, int num_gene, City *city_table)
Definition: geqo_px.c:46
unsigned int uint32
Definition: c.h:268
int(* abbrev_full_comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:192
void * ssup_extra
Definition: sortsupport.h:87
ArrayType * create_singleton_array(FunctionCallInfo fcinfo, Oid element_type, Datum element, bool isNull, int ndims)
Datum textpos(PG_FUNCTION_ARGS)
Definition: varlena.c:1072
size_t pg_regerror(int errcode, const regex_t *preg, char *errbuf, size_t errbuf_size)
Definition: regerror.c:60
MemoryContext CurrentMemoryContext
Definition: mcxt.c:37
Datum text_substr_no_len(PG_FUNCTION_ARGS)
Definition: varlena.c:795
int bytea_output
Definition: varlena.c:41
void fmgr_info_cxt(Oid functionId, FmgrInfo *finfo, MemoryContext mcxt)
Definition: fmgr.c:137
static int text_cmp(text *arg1, text *arg2, Oid collid)
Definition: varlena.c:1611
Datum byteaGetByte(PG_FUNCTION_ARGS)
Definition: varlena.c:2994
#define S(n, x)
Definition: sha1.c:55
#define PG_RETURN_ARRAYTYPE_P(x)
Definition: array.h:246
Datum pg_column_size(PG_FUNCTION_ARGS)
Definition: varlena.c:4584
Datum text_gt(PG_FUNCTION_ARGS)
Definition: varlena.c:1730
#define att_addlength_pointer(cur_offset, attlen, attptr)
Definition: tupmacs.h:172
#define ereport(elevel, rest)
Definition: elog.h:122
static int internal_text_pattern_compare(text *arg1, text *arg2)
Definition: varlena.c:2600
Datum makeArrayResult(ArrayBuildState *astate, MemoryContext rcontext)
Definition: arrayfuncs.c:5055
static bool text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
Definition: varlena.c:5201
unsigned int pg_wchar
Definition: mbprint.c:31
#define DatumGetVarStringPP(X)
Definition: varlena.c:92
List * textToQualifiedNameList(text *textval)
Definition: varlena.c:3182
#define byte(x, n)
Definition: rijndael.c:68
Datum textcat(PG_FUNCTION_ARGS)
Definition: varlena.c:679
List * lappend(List *list, void *datum)
Definition: list.c:128
Datum name_text(PG_FUNCTION_ARGS)
Definition: varlena.c:3165
char * pq_getmsgtext(StringInfo msg, int rawbytes, int *nbytes)
Definition: pqformat.c:588
#define MaxAllocSize
Definition: memutils.h:40
int skiptable[256]
Definition: varlena.c:57
void appendStringInfoChar(StringInfo str, char ch)
Definition: stringinfo.c:201
void initStringInfo(StringInfo str)
Definition: stringinfo.c:65
Datum byteasend(PG_FUNCTION_ARGS)
Definition: varlena.c:442
void varstr_sortsupport(SortSupport ssup, Oid collid, bool bpchar)
Definition: varlena.c:1802
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1265
Datum text_le(PG_FUNCTION_ARGS)
Definition: varlena.c:1715
Datum hash_uint32(uint32 k)
Definition: hashfunc.c:512
uint8 bits8
Definition: c.h:275
Datum text_substr(PG_FUNCTION_ARGS)
Definition: varlena.c:781
#define TextDatumGetCString(d)
Definition: builtins.h:92
void * palloc0(Size size)
Definition: mcxt.c:878
Datum text_format_nv(PG_FUNCTION_ARGS)
Definition: varlena.c:5428
char * s2
#define PG_RETURN_BOOL(x)
Definition: fmgr.h:319
uintptr_t Datum
Definition: postgres.h:372
int GetDatabaseEncoding(void)
Definition: mbutils.c:1015
Datum text_reverse(PG_FUNCTION_ARGS)
Definition: varlena.c:4869
Datum bytea_substr(PG_FUNCTION_ARGS)
Definition: varlena.c:2808
int bpchartruelen(char *s, int len)
Definition: varchar.c:660
#define REGEXP_REPLACE_BACKREF_CNT
Definition: varlena.c:3875
void appendStringInfoSpaces(StringInfo str, int count)
Definition: stringinfo.c:219
int pg_mbstrlen(const char *mbstr)
Definition: mbutils.c:785
double estimateHyperLogLog(hyperLogLogState *cState)
Definition: hyperloglog.c:185
Datum text_ge(PG_FUNCTION_ARGS)
Definition: varlena.c:1745
#define VARSIZE_ANY(PTR)
Definition: postgres.h:334
#define strxfrm_l
Definition: win32.h:347
static void text_position_cleanup(TextPositionState *state)
Definition: varlena.c:1365
Datum byteacmp(PG_FUNCTION_ARGS)
Definition: varlena.c:3607
#define InvalidOid
Definition: postgres_ext.h:36
int pg_mb2wchar_with_len(const char *from, pg_wchar *to, int len)
Definition: mbutils.c:734
Datum to_hex32(PG_FUNCTION_ARGS)
Definition: varlena.c:4488
hyperLogLogState full_card
Definition: varlena.c:74
#define PG_RETURN_VOID()
Definition: fmgr.h:309
#define PG_RETURN_TEXT_P(x)
Definition: fmgr.h:330
#define Max(x, y)
Definition: c.h:800
text * cstring_to_text(const char *s)
Definition: varlena.c:149
Datum unknownsend(PG_FUNCTION_ARGS)
Definition: varlena.c:602
#define PG_ARGISNULL(n)
Definition: fmgr.h:174
#define NULL
Definition: c.h:229
bool pg_md5_hash(const void *buff, size_t len, char *hexsum)
Definition: md5.c:293
#define Assert(condition)
Definition: c.h:675
#define lfirst(lc)
Definition: pg_list.h:106
Definition: regguts.h:298
Datum hash_any(register const unsigned char *k, register int keylen)
Definition: hashfunc.c:307
void pq_copymsgbytes(StringInfo msg, char *buf, int datalen)
Definition: pqformat.c:570
Datum text_right(PG_FUNCTION_ARGS)
Definition: varlena.c:4848
static text * concat_internal(const char *sepstr, int argidx, FunctionCallInfo fcinfo)
Definition: varlena.c:4711
int varstr_cmp(char *arg1, int len1, char *arg2, int len2, Oid collid)
Definition: varlena.c:1382
Oid typioparam
Definition: array.h:224
#define PG_RETURN_CSTRING(x)
Definition: fmgr.h:322
Datum unknownin(PG_FUNCTION_ARGS)
Definition: varlena.c:563
size_t Size
Definition: c.h:356
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2453
static bool rest_of_char_same(const char *s1, const char *s2, int len)
Definition: varlena.c:5438
Datum text_pattern_lt(PG_FUNCTION_ARGS)
Definition: varlena.c:2622
Datum textsend(PG_FUNCTION_ARGS)
Definition: varlena.c:548
#define newval
#define PG_GETARG_BYTEA_PP(n)
Definition: fmgr.h:272
Datum byteane(PG_FUNCTION_ARGS)
Definition: varlena.c:3495
void addHyperLogLog(hyperLogLogState *cState, uint32 hash)
Definition: hyperloglog.c:166
int AggCheckCallContext(FunctionCallInfo fcinfo, MemoryContext *aggcontext)
Definition: nodeAgg.c:4018
Datum textin(PG_FUNCTION_ARGS)
Definition: varlena.c:508
#define PG_FREE_IF_COPY(ptr, n)
Definition: fmgr.h:225
#define PG_NARGS()
Definition: fmgr.h:168
#define C_COLLATION_OID
Definition: pg_collation.h:77
void * fn_extra
Definition: fmgr.h:64
int pg_mblen(const char *mbstr)
Definition: mbutils.c:771
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:3781
#define ARR_NDIM(a)
Definition: array.h:271
Datum byteapos(PG_FUNCTION_ARGS)
Definition: varlena.c:2950
#define TEXTBUFLEN
Definition: varlena.c:83
Oid typiofunc
Definition: array.h:225
#define DatumGetPointer(X)
Definition: postgres.h:555
char typalign
Definition: array.h:222
void deconstruct_array(ArrayType *array, Oid elmtype, int elmlen, bool elmbyval, char elmalign, Datum **elemsp, bool **nullsp, int *nelemsp)
Definition: arrayfuncs.c:3475
char * text_to_cstring(const text *t)
Definition: varlena.c:182
pg_wchar * wstr1
Definition: varlena.c:51
ArrayBuildState * accumArrayResult(ArrayBuildState *astate, Datum dvalue, bool disnull, Oid element_type, MemoryContext rcontext)
Definition: arrayfuncs.c:4991
#define DatumGetBpCharPP(X)
Definition: fmgr.h:257
Oid get_base_element_type(Oid typid)
Definition: lsyscache.c:2557
Datum bttextcmp(PG_FUNCTION_ARGS)
Definition: varlena.c:1760
Datum unknownout(PG_FUNCTION_ARGS)
Definition: varlena.c:575
int pg_regexec(regex_t *re, const chr *string, size_t len, size_t search_start, rm_detail_t *details, size_t nmatch, regmatch_t pmatch[], int flags)
Definition: regexec.c:172
int16 get_typlen(Oid typid)
Definition: lsyscache.c:1947
char * OidOutputFunctionCall(Oid functionId, Datum val)
Definition: fmgr.c:1747
Datum bytearecv(PG_FUNCTION_ARGS)
Definition: varlena.c:423
#define VARSIZE_ANY_EXHDR(PTR)
Definition: postgres.h:340
void * palloc(Size size)
Definition: mcxt.c:849
int errmsg(const char *fmt,...)
Definition: elog.c:797
#define fetch_att(T, attbyval, attlen)
Definition: tupmacs.h:71
static StringInfo makeStringAggState(FunctionCallInfo fcinfo)
Definition: varlena.c:4638
FmgrInfo proc
Definition: array.h:226
Datum bytea_string_agg_transfn(PG_FUNCTION_ARGS)
Definition: varlena.c:450
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:707
Datum byteaoverlay(PG_FUNCTION_ARGS)
Definition: varlena.c:2890
Datum bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
Definition: varlena.c:2702
Datum byteaoctetlen(PG_FUNCTION_ARGS)
Definition: varlena.c:2725
void list_free(List *list)
Definition: list.c:1133
int i
Oid element_type
Definition: array.h:219
#define REG_NOMATCH
Definition: regex.h:138
#define NameStr(name)
Definition: c.h:499
static char * locale
Definition: initdb.c:123
void * arg
static bool text_isequal(text *txt1, text *txt2)
Definition: varlena.c:4118
#define PG_GETARG_CSTRING(n)
Definition: fmgr.h:242
#define VAL(CH)
Definition: varlena.c:240