PostgreSQL Source Code  git master
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);
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 {
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 {
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(const char *arg1, int len1, const 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  && (!mylocale || mylocale->provider == COLLPROVIDER_LIBC))
1438  {
1439  int a1len;
1440  int a2len;
1441  int r;
1442 
1443  if (len1 >= TEXTBUFLEN / 2)
1444  {
1445  a1len = len1 * 2 + 2;
1446  a1p = palloc(a1len);
1447  }
1448  else
1449  {
1450  a1len = TEXTBUFLEN;
1451  a1p = a1buf;
1452  }
1453  if (len2 >= TEXTBUFLEN / 2)
1454  {
1455  a2len = len2 * 2 + 2;
1456  a2p = palloc(a2len);
1457  }
1458  else
1459  {
1460  a2len = TEXTBUFLEN;
1461  a2p = a2buf;
1462  }
1463 
1464  /* stupid Microsloth API does not work for zero-length input */
1465  if (len1 == 0)
1466  r = 0;
1467  else
1468  {
1469  r = MultiByteToWideChar(CP_UTF8, 0, arg1, len1,
1470  (LPWSTR) a1p, a1len / 2);
1471  if (!r)
1472  ereport(ERROR,
1473  (errmsg("could not convert string to UTF-16: error code %lu",
1474  GetLastError())));
1475  }
1476  ((LPWSTR) a1p)[r] = 0;
1477 
1478  if (len2 == 0)
1479  r = 0;
1480  else
1481  {
1482  r = MultiByteToWideChar(CP_UTF8, 0, arg2, len2,
1483  (LPWSTR) a2p, a2len / 2);
1484  if (!r)
1485  ereport(ERROR,
1486  (errmsg("could not convert string to UTF-16: error code %lu",
1487  GetLastError())));
1488  }
1489  ((LPWSTR) a2p)[r] = 0;
1490 
1491  errno = 0;
1492 #ifdef HAVE_LOCALE_T
1493  if (mylocale)
1494  result = wcscoll_l((LPWSTR) a1p, (LPWSTR) a2p, mylocale->info.lt);
1495  else
1496 #endif
1497  result = wcscoll((LPWSTR) a1p, (LPWSTR) a2p);
1498  if (result == 2147483647) /* _NLSCMPERROR; missing from mingw
1499  * headers */
1500  ereport(ERROR,
1501  (errmsg("could not compare Unicode strings: %m")));
1502 
1503  /*
1504  * In some locales wcscoll() can claim that nonidentical strings
1505  * are equal. Believing that would be bad news for a number of
1506  * reasons, so we follow Perl's lead and sort "equal" strings
1507  * according to strcmp (on the UTF-8 representation).
1508  */
1509  if (result == 0)
1510  {
1511  result = memcmp(arg1, arg2, Min(len1, len2));
1512  if ((result == 0) && (len1 != len2))
1513  result = (len1 < len2) ? -1 : 1;
1514  }
1515 
1516  if (a1p != a1buf)
1517  pfree(a1p);
1518  if (a2p != a2buf)
1519  pfree(a2p);
1520 
1521  return result;
1522  }
1523 #endif /* WIN32 */
1524 
1525  if (len1 >= TEXTBUFLEN)
1526  a1p = (char *) palloc(len1 + 1);
1527  else
1528  a1p = a1buf;
1529  if (len2 >= TEXTBUFLEN)
1530  a2p = (char *) palloc(len2 + 1);
1531  else
1532  a2p = a2buf;
1533 
1534  memcpy(a1p, arg1, len1);
1535  a1p[len1] = '\0';
1536  memcpy(a2p, arg2, len2);
1537  a2p[len2] = '\0';
1538 
1539  if (mylocale)
1540  {
1541  if (mylocale->provider == COLLPROVIDER_ICU)
1542  {
1543 #ifdef USE_ICU
1544 #ifdef HAVE_UCOL_STRCOLLUTF8
1545  if (GetDatabaseEncoding() == PG_UTF8)
1546  {
1547  UErrorCode status;
1548 
1549  status = U_ZERO_ERROR;
1550  result = ucol_strcollUTF8(mylocale->info.icu.ucol,
1551  arg1, len1,
1552  arg2, len2,
1553  &status);
1554  if (U_FAILURE(status))
1555  ereport(ERROR,
1556  (errmsg("collation failed: %s", u_errorName(status))));
1557  }
1558  else
1559 #endif
1560  {
1561  int32_t ulen1,
1562  ulen2;
1563  UChar *uchar1,
1564  *uchar2;
1565 
1566  ulen1 = icu_to_uchar(&uchar1, arg1, len1);
1567  ulen2 = icu_to_uchar(&uchar2, arg2, len2);
1568 
1569  result = ucol_strcoll(mylocale->info.icu.ucol,
1570  uchar1, ulen1,
1571  uchar2, ulen2);
1572 
1573  pfree(uchar1);
1574  pfree(uchar2);
1575  }
1576 #else /* not USE_ICU */
1577  /* shouldn't happen */
1578  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1579 #endif /* not USE_ICU */
1580  }
1581  else
1582  {
1583 #ifdef HAVE_LOCALE_T
1584  result = strcoll_l(a1p, a2p, mylocale->info.lt);
1585 #else
1586  /* shouldn't happen */
1587  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1588 #endif
1589  }
1590  }
1591  else
1592  result = strcoll(a1p, a2p);
1593 
1594  /*
1595  * In some locales strcoll() can claim that nonidentical strings are
1596  * equal. Believing that would be bad news for a number of reasons,
1597  * so we follow Perl's lead and sort "equal" strings according to
1598  * strcmp().
1599  */
1600  if (result == 0)
1601  result = strcmp(a1p, a2p);
1602 
1603  if (a1p != a1buf)
1604  pfree(a1p);
1605  if (a2p != a2buf)
1606  pfree(a2p);
1607  }
1608 
1609  return result;
1610 }
1611 
1612 /* text_cmp()
1613  * Internal comparison function for text strings.
1614  * Returns -1, 0 or 1
1615  */
1616 static int
1617 text_cmp(text *arg1, text *arg2, Oid collid)
1618 {
1619  char *a1p,
1620  *a2p;
1621  int len1,
1622  len2;
1623 
1624  a1p = VARDATA_ANY(arg1);
1625  a2p = VARDATA_ANY(arg2);
1626 
1627  len1 = VARSIZE_ANY_EXHDR(arg1);
1628  len2 = VARSIZE_ANY_EXHDR(arg2);
1629 
1630  return varstr_cmp(a1p, len1, a2p, len2, collid);
1631 }
1632 
1633 /*
1634  * Comparison functions for text strings.
1635  *
1636  * Note: btree indexes need these routines not to leak memory; therefore,
1637  * be careful to free working copies of toasted datums. Most places don't
1638  * need to be so careful.
1639  */
1640 
1641 Datum
1643 {
1644  Datum arg1 = PG_GETARG_DATUM(0);
1645  Datum arg2 = PG_GETARG_DATUM(1);
1646  bool result;
1647  Size len1,
1648  len2;
1649 
1650  /*
1651  * Since we only care about equality or not-equality, we can avoid all the
1652  * expense of strcoll() here, and just do bitwise comparison. In fact, we
1653  * don't even have to do a bitwise comparison if we can show the lengths
1654  * of the strings are unequal; which might save us from having to detoast
1655  * one or both values.
1656  */
1657  len1 = toast_raw_datum_size(arg1);
1658  len2 = toast_raw_datum_size(arg2);
1659  if (len1 != len2)
1660  result = false;
1661  else
1662  {
1663  text *targ1 = DatumGetTextPP(arg1);
1664  text *targ2 = DatumGetTextPP(arg2);
1665 
1666  result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1667  len1 - VARHDRSZ) == 0);
1668 
1669  PG_FREE_IF_COPY(targ1, 0);
1670  PG_FREE_IF_COPY(targ2, 1);
1671  }
1672 
1673  PG_RETURN_BOOL(result);
1674 }
1675 
1676 Datum
1678 {
1679  Datum arg1 = PG_GETARG_DATUM(0);
1680  Datum arg2 = PG_GETARG_DATUM(1);
1681  bool result;
1682  Size len1,
1683  len2;
1684 
1685  /* See comment in texteq() */
1686  len1 = toast_raw_datum_size(arg1);
1687  len2 = toast_raw_datum_size(arg2);
1688  if (len1 != len2)
1689  result = true;
1690  else
1691  {
1692  text *targ1 = DatumGetTextPP(arg1);
1693  text *targ2 = DatumGetTextPP(arg2);
1694 
1695  result = (memcmp(VARDATA_ANY(targ1), VARDATA_ANY(targ2),
1696  len1 - VARHDRSZ) != 0);
1697 
1698  PG_FREE_IF_COPY(targ1, 0);
1699  PG_FREE_IF_COPY(targ2, 1);
1700  }
1701 
1702  PG_RETURN_BOOL(result);
1703 }
1704 
1705 Datum
1707 {
1708  text *arg1 = PG_GETARG_TEXT_PP(0);
1709  text *arg2 = PG_GETARG_TEXT_PP(1);
1710  bool result;
1711 
1712  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0);
1713 
1714  PG_FREE_IF_COPY(arg1, 0);
1715  PG_FREE_IF_COPY(arg2, 1);
1716 
1717  PG_RETURN_BOOL(result);
1718 }
1719 
1720 Datum
1722 {
1723  text *arg1 = PG_GETARG_TEXT_PP(0);
1724  text *arg2 = PG_GETARG_TEXT_PP(1);
1725  bool result;
1726 
1727  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) <= 0);
1728 
1729  PG_FREE_IF_COPY(arg1, 0);
1730  PG_FREE_IF_COPY(arg2, 1);
1731 
1732  PG_RETURN_BOOL(result);
1733 }
1734 
1735 Datum
1737 {
1738  text *arg1 = PG_GETARG_TEXT_PP(0);
1739  text *arg2 = PG_GETARG_TEXT_PP(1);
1740  bool result;
1741 
1742  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0);
1743 
1744  PG_FREE_IF_COPY(arg1, 0);
1745  PG_FREE_IF_COPY(arg2, 1);
1746 
1747  PG_RETURN_BOOL(result);
1748 }
1749 
1750 Datum
1752 {
1753  text *arg1 = PG_GETARG_TEXT_PP(0);
1754  text *arg2 = PG_GETARG_TEXT_PP(1);
1755  bool result;
1756 
1757  result = (text_cmp(arg1, arg2, PG_GET_COLLATION()) >= 0);
1758 
1759  PG_FREE_IF_COPY(arg1, 0);
1760  PG_FREE_IF_COPY(arg2, 1);
1761 
1762  PG_RETURN_BOOL(result);
1763 }
1764 
1765 Datum
1767 {
1768  text *arg1 = PG_GETARG_TEXT_PP(0);
1769  text *arg2 = PG_GETARG_TEXT_PP(1);
1770  int32 result;
1771 
1772  result = text_cmp(arg1, arg2, PG_GET_COLLATION());
1773 
1774  PG_FREE_IF_COPY(arg1, 0);
1775  PG_FREE_IF_COPY(arg2, 1);
1776 
1777  PG_RETURN_INT32(result);
1778 }
1779 
1780 Datum
1782 {
1784  Oid collid = ssup->ssup_collation;
1785  MemoryContext oldcontext;
1786 
1787  oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
1788 
1789  /* Use generic string SortSupport */
1790  varstr_sortsupport(ssup, collid, false);
1791 
1792  MemoryContextSwitchTo(oldcontext);
1793 
1794  PG_RETURN_VOID();
1795 }
1796 
1797 /*
1798  * Generic sortsupport interface for character type's operator classes.
1799  * Includes locale support, and support for BpChar semantics (i.e. removing
1800  * trailing spaces before comparison).
1801  *
1802  * Relies on the assumption that text, VarChar, BpChar, and bytea all have the
1803  * same representation. Callers that always use the C collation (e.g.
1804  * non-collatable type callers like bytea) may have NUL bytes in their strings;
1805  * this will not work with any other collation, though.
1806  */
1807 void
1809 {
1810  bool abbreviate = ssup->abbreviate;
1811  bool collate_c = false;
1812  VarStringSortSupport *sss;
1813  pg_locale_t locale = 0;
1814 
1815  /*
1816  * If possible, set ssup->comparator to a function which can be used to
1817  * directly compare two datums. If we can do this, we'll avoid the
1818  * overhead of a trip through the fmgr layer for every comparison, which
1819  * can be substantial.
1820  *
1821  * Most typically, we'll set the comparator to varstrfastcmp_locale, which
1822  * uses strcoll() to perform comparisons and knows about the special
1823  * requirements of BpChar callers. However, if LC_COLLATE = C, we can
1824  * make things quite a bit faster with varstrfastcmp_c or bpcharfastcmp_c,
1825  * both of which use memcmp() rather than strcoll().
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  else
1837  {
1838  /*
1839  * We need a collation-sensitive comparison. To make things faster,
1840  * we'll figure out the collation based on the locale id and cache the
1841  * result.
1842  */
1843  if (collid != DEFAULT_COLLATION_OID)
1844  {
1845  if (!OidIsValid(collid))
1846  {
1847  /*
1848  * This typically means that the parser could not resolve a
1849  * conflict of implicit collations, so report it that way.
1850  */
1851  ereport(ERROR,
1852  (errcode(ERRCODE_INDETERMINATE_COLLATION),
1853  errmsg("could not determine which collation to use for string comparison"),
1854  errhint("Use the COLLATE clause to set the collation explicitly.")));
1855  }
1856  locale = pg_newlocale_from_collation(collid);
1857  }
1858 
1859  /*
1860  * There is a further exception on Windows. When the database
1861  * encoding is UTF-8 and we are not using the C collation, complex
1862  * hacks are required. We don't currently have a comparator that
1863  * handles that case, so we fall back on the slow method of having the
1864  * sort code invoke bttextcmp() (in the case of text) via the fmgr
1865  * trampoline. ICU locales work just the same on Windows, however.
1866  */
1867 #ifdef WIN32
1868  if (GetDatabaseEncoding() == PG_UTF8 &&
1869  !(locale && locale->provider == COLLPROVIDER_ICU))
1870  return;
1871 #endif
1872 
1874  }
1875 
1876  /*
1877  * Unfortunately, it seems that abbreviation for non-C collations is
1878  * broken on many common platforms; testing of multiple versions of glibc
1879  * reveals that, for many locales, strcoll() and strxfrm() do not return
1880  * consistent results, which is fatal to this optimization. While no
1881  * other libc other than Cygwin has so far been shown to have a problem,
1882  * we take the conservative course of action for right now and disable
1883  * this categorically. (Users who are certain this isn't a problem on
1884  * their system can define TRUST_STRXFRM.)
1885  *
1886  * Even apart from the risk of broken locales, it's possible that there
1887  * are platforms where the use of abbreviated keys should be disabled at
1888  * compile time. Having only 4 byte datums could make worst-case
1889  * performance drastically more likely, for example. Moreover, macOS's
1890  * strxfrm() implementation is known to not effectively concentrate a
1891  * significant amount of entropy from the original string in earlier
1892  * transformed blobs. It's possible that other supported platforms are
1893  * similarly encumbered. So, if we ever get past disabling this
1894  * categorically, we may still want or need to disable it for particular
1895  * platforms.
1896  */
1897 #ifndef TRUST_STRXFRM
1898  if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
1899  abbreviate = false;
1900 #endif
1901 
1902  /*
1903  * If we're using abbreviated keys, or if we're using a locale-aware
1904  * comparison, we need to initialize a StringSortSupport object. Both
1905  * cases will make use of the temporary buffers we initialize here for
1906  * scratch space (and to detect requirement for BpChar semantics from
1907  * caller), and the abbreviation case requires additional state.
1908  */
1909  if (abbreviate || !collate_c)
1910  {
1911  sss = palloc(sizeof(VarStringSortSupport));
1912  sss->buf1 = palloc(TEXTBUFLEN);
1913  sss->buflen1 = TEXTBUFLEN;
1914  sss->buf2 = palloc(TEXTBUFLEN);
1915  sss->buflen2 = TEXTBUFLEN;
1916  /* Start with invalid values */
1917  sss->last_len1 = -1;
1918  sss->last_len2 = -1;
1919  /* Initialize */
1920  sss->last_returned = 0;
1921  sss->locale = locale;
1922 
1923  /*
1924  * To avoid somehow confusing a strxfrm() blob and an original string,
1925  * constantly keep track of the variety of data that buf1 and buf2
1926  * currently contain.
1927  *
1928  * Comparisons may be interleaved with conversion calls. Frequently,
1929  * conversions and comparisons are batched into two distinct phases,
1930  * but the correctness of caching cannot hinge upon this. For
1931  * comparison caching, buffer state is only trusted if cache_blob is
1932  * found set to false, whereas strxfrm() caching only trusts the state
1933  * when cache_blob is found set to true.
1934  *
1935  * Arbitrarily initialize cache_blob to true.
1936  */
1937  sss->cache_blob = true;
1938  sss->collate_c = collate_c;
1939  sss->bpchar = bpchar;
1940  ssup->ssup_extra = sss;
1941 
1942  /*
1943  * If possible, plan to use the abbreviated keys optimization. The
1944  * core code may switch back to authoritative comparator should
1945  * abbreviation be aborted.
1946  */
1947  if (abbreviate)
1948  {
1949  sss->prop_card = 0.20;
1950  initHyperLogLog(&sss->abbr_card, 10);
1951  initHyperLogLog(&sss->full_card, 10);
1952  ssup->abbrev_full_comparator = ssup->comparator;
1953  ssup->comparator = varstrcmp_abbrev;
1956  }
1957  }
1958 }
1959 
1960 /*
1961  * sortsupport comparison func (for C locale case)
1962  */
1963 static int
1965 {
1966  VarString *arg1 = DatumGetVarStringPP(x);
1967  VarString *arg2 = DatumGetVarStringPP(y);
1968  char *a1p,
1969  *a2p;
1970  int len1,
1971  len2,
1972  result;
1973 
1974  a1p = VARDATA_ANY(arg1);
1975  a2p = VARDATA_ANY(arg2);
1976 
1977  len1 = VARSIZE_ANY_EXHDR(arg1);
1978  len2 = VARSIZE_ANY_EXHDR(arg2);
1979 
1980  result = memcmp(a1p, a2p, Min(len1, len2));
1981  if ((result == 0) && (len1 != len2))
1982  result = (len1 < len2) ? -1 : 1;
1983 
1984  /* We can't afford to leak memory here. */
1985  if (PointerGetDatum(arg1) != x)
1986  pfree(arg1);
1987  if (PointerGetDatum(arg2) != y)
1988  pfree(arg2);
1989 
1990  return result;
1991 }
1992 
1993 /*
1994  * sortsupport comparison func (for BpChar C locale case)
1995  *
1996  * BpChar outsources its sortsupport to this module. Specialization for the
1997  * varstr_sortsupport BpChar case, modeled on
1998  * internal_bpchar_pattern_compare().
1999  */
2000 static int
2002 {
2003  BpChar *arg1 = DatumGetBpCharPP(x);
2004  BpChar *arg2 = DatumGetBpCharPP(y);
2005  char *a1p,
2006  *a2p;
2007  int len1,
2008  len2,
2009  result;
2010 
2011  a1p = VARDATA_ANY(arg1);
2012  a2p = VARDATA_ANY(arg2);
2013 
2014  len1 = bpchartruelen(a1p, VARSIZE_ANY_EXHDR(arg1));
2015  len2 = bpchartruelen(a2p, VARSIZE_ANY_EXHDR(arg2));
2016 
2017  result = memcmp(a1p, a2p, Min(len1, len2));
2018  if ((result == 0) && (len1 != len2))
2019  result = (len1 < len2) ? -1 : 1;
2020 
2021  /* We can't afford to leak memory here. */
2022  if (PointerGetDatum(arg1) != x)
2023  pfree(arg1);
2024  if (PointerGetDatum(arg2) != y)
2025  pfree(arg2);
2026 
2027  return result;
2028 }
2029 
2030 /*
2031  * sortsupport comparison func (for locale case)
2032  */
2033 static int
2035 {
2036  VarString *arg1 = DatumGetVarStringPP(x);
2037  VarString *arg2 = DatumGetVarStringPP(y);
2038  bool arg1_match;
2040 
2041  /* working state */
2042  char *a1p,
2043  *a2p;
2044  int len1,
2045  len2,
2046  result;
2047 
2048  a1p = VARDATA_ANY(arg1);
2049  a2p = VARDATA_ANY(arg2);
2050 
2051  len1 = VARSIZE_ANY_EXHDR(arg1);
2052  len2 = VARSIZE_ANY_EXHDR(arg2);
2053 
2054  /* Fast pre-check for equality, as discussed in varstr_cmp() */
2055  if (len1 == len2 && memcmp(a1p, a2p, len1) == 0)
2056  {
2057  /*
2058  * No change in buf1 or buf2 contents, so avoid changing last_len1 or
2059  * last_len2. Existing contents of buffers might still be used by
2060  * next call.
2061  *
2062  * It's fine to allow the comparison of BpChar padding bytes here,
2063  * even though that implies that the memcmp() will usually be
2064  * performed for BpChar callers (though multibyte characters could
2065  * still prevent that from occurring). The memcmp() is still very
2066  * cheap, and BpChar's funny semantics have us remove trailing spaces
2067  * (not limited to padding), so we need make no distinction between
2068  * padding space characters and "real" space characters.
2069  */
2070  result = 0;
2071  goto done;
2072  }
2073 
2074  if (sss->bpchar)
2075  {
2076  /* Get true number of bytes, ignoring trailing spaces */
2077  len1 = bpchartruelen(a1p, len1);
2078  len2 = bpchartruelen(a2p, len2);
2079  }
2080 
2081  if (len1 >= sss->buflen1)
2082  {
2083  pfree(sss->buf1);
2084  sss->buflen1 = Max(len1 + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2085  sss->buf1 = MemoryContextAlloc(ssup->ssup_cxt, sss->buflen1);
2086  }
2087  if (len2 >= sss->buflen2)
2088  {
2089  pfree(sss->buf2);
2090  sss->buflen2 = Max(len2 + 1, Min(sss->buflen2 * 2, MaxAllocSize));
2091  sss->buf2 = MemoryContextAlloc(ssup->ssup_cxt, sss->buflen2);
2092  }
2093 
2094  /*
2095  * We're likely to be asked to compare the same strings repeatedly, and
2096  * memcmp() is so much cheaper than strcoll() that it pays to try to cache
2097  * comparisons, even though in general there is no reason to think that
2098  * that will work out (every string datum may be unique). Caching does
2099  * not slow things down measurably when it doesn't work out, and can speed
2100  * things up by rather a lot when it does. In part, this is because the
2101  * memcmp() compares data from cachelines that are needed in L1 cache even
2102  * when the last comparison's result cannot be reused.
2103  */
2104  arg1_match = true;
2105  if (len1 != sss->last_len1 || memcmp(sss->buf1, a1p, len1) != 0)
2106  {
2107  arg1_match = false;
2108  memcpy(sss->buf1, a1p, len1);
2109  sss->buf1[len1] = '\0';
2110  sss->last_len1 = len1;
2111  }
2112 
2113  /*
2114  * If we're comparing the same two strings as last time, we can return the
2115  * same answer without calling strcoll() again. This is more likely than
2116  * it seems (at least with moderate to low cardinality sets), because
2117  * quicksort compares the same pivot against many values.
2118  */
2119  if (len2 != sss->last_len2 || memcmp(sss->buf2, a2p, len2) != 0)
2120  {
2121  memcpy(sss->buf2, a2p, len2);
2122  sss->buf2[len2] = '\0';
2123  sss->last_len2 = len2;
2124  }
2125  else if (arg1_match && !sss->cache_blob)
2126  {
2127  /* Use result cached following last actual strcoll() call */
2128  result = sss->last_returned;
2129  goto done;
2130  }
2131 
2132  if (sss->locale)
2133  {
2134  if (sss->locale->provider == COLLPROVIDER_ICU)
2135  {
2136 #ifdef USE_ICU
2137 #ifdef HAVE_UCOL_STRCOLLUTF8
2138  if (GetDatabaseEncoding() == PG_UTF8)
2139  {
2140  UErrorCode status;
2141 
2142  status = U_ZERO_ERROR;
2143  result = ucol_strcollUTF8(sss->locale->info.icu.ucol,
2144  a1p, len1,
2145  a2p, len2,
2146  &status);
2147  if (U_FAILURE(status))
2148  ereport(ERROR,
2149  (errmsg("collation failed: %s", u_errorName(status))));
2150  }
2151  else
2152 #endif
2153  {
2154  int32_t ulen1,
2155  ulen2;
2156  UChar *uchar1,
2157  *uchar2;
2158 
2159  ulen1 = icu_to_uchar(&uchar1, a1p, len1);
2160  ulen2 = icu_to_uchar(&uchar2, a2p, len2);
2161 
2162  result = ucol_strcoll(sss->locale->info.icu.ucol,
2163  uchar1, ulen1,
2164  uchar2, ulen2);
2165 
2166  pfree(uchar1);
2167  pfree(uchar2);
2168  }
2169 #else /* not USE_ICU */
2170  /* shouldn't happen */
2171  elog(ERROR, "unsupported collprovider: %c", sss->locale->provider);
2172 #endif /* not USE_ICU */
2173  }
2174  else
2175  {
2176 #ifdef HAVE_LOCALE_T
2177  result = strcoll_l(sss->buf1, sss->buf2, sss->locale->info.lt);
2178 #else
2179  /* shouldn't happen */
2180  elog(ERROR, "unsupported collprovider: %c", sss->locale->provider);
2181 #endif
2182  }
2183  }
2184  else
2185  result = strcoll(sss->buf1, sss->buf2);
2186 
2187  /*
2188  * In some locales strcoll() can claim that nonidentical strings are
2189  * equal. Believing that would be bad news for a number of reasons, so we
2190  * follow Perl's lead and sort "equal" strings according to strcmp().
2191  */
2192  if (result == 0)
2193  result = strcmp(sss->buf1, sss->buf2);
2194 
2195  /* Cache result, perhaps saving an expensive strcoll() call next time */
2196  sss->cache_blob = false;
2197  sss->last_returned = result;
2198 done:
2199  /* We can't afford to leak memory here. */
2200  if (PointerGetDatum(arg1) != x)
2201  pfree(arg1);
2202  if (PointerGetDatum(arg2) != y)
2203  pfree(arg2);
2204 
2205  return result;
2206 }
2207 
2208 /*
2209  * Abbreviated key comparison func
2210  */
2211 static int
2213 {
2214  /*
2215  * When 0 is returned, the core system will call varstrfastcmp_c()
2216  * (bpcharfastcmp_c() in BpChar case) or varstrfastcmp_locale(). Even a
2217  * strcmp() on two non-truncated strxfrm() blobs cannot indicate *equality*
2218  * authoritatively, for the same reason that there is a strcoll()
2219  * tie-breaker call to strcmp() in varstr_cmp().
2220  */
2221  if (x > y)
2222  return 1;
2223  else if (x == y)
2224  return 0;
2225  else
2226  return -1;
2227 }
2228 
2229 /*
2230  * Conversion routine for sortsupport. Converts original to abbreviated key
2231  * representation. Our encoding strategy is simple -- pack the first 8 bytes
2232  * of a strxfrm() blob into a Datum (on little-endian machines, the 8 bytes are
2233  * stored in reverse order), and treat it as an unsigned integer. When the "C"
2234  * locale is used, or in case of bytea, just memcpy() from original instead.
2235  */
2236 static Datum
2238 {
2240  VarString *authoritative = DatumGetVarStringPP(original);
2241  char *authoritative_data = VARDATA_ANY(authoritative);
2242 
2243  /* working state */
2244  Datum res;
2245  char *pres;
2246  int len;
2247  uint32 hash;
2248 
2249  pres = (char *) &res;
2250  /* memset(), so any non-overwritten bytes are NUL */
2251  memset(pres, 0, sizeof(Datum));
2252  len = VARSIZE_ANY_EXHDR(authoritative);
2253 
2254  /* Get number of bytes, ignoring trailing spaces */
2255  if (sss->bpchar)
2256  len = bpchartruelen(authoritative_data, len);
2257 
2258  /*
2259  * If we're using the C collation, use memcpy(), rather than strxfrm(), to
2260  * abbreviate keys. The full comparator for the C locale is always
2261  * memcmp(). It would be incorrect to allow bytea callers (callers that
2262  * always force the C collation -- bytea isn't a collatable type, but this
2263  * approach is convenient) to use strxfrm(). This is because bytea
2264  * strings may contain NUL bytes. Besides, this should be faster, too.
2265  *
2266  * More generally, it's okay that bytea callers can have NUL bytes in
2267  * strings because varstrcmp_abbrev() need not make a distinction between
2268  * terminating NUL bytes, and NUL bytes representing actual NULs in the
2269  * authoritative representation. Hopefully a comparison at or past one
2270  * abbreviated key's terminating NUL byte will resolve the comparison
2271  * without consulting the authoritative representation; specifically, some
2272  * later non-NUL byte in the longer string can resolve the comparison
2273  * against a subsequent terminating NUL in the shorter string. There will
2274  * usually be what is effectively a "length-wise" resolution there and
2275  * then.
2276  *
2277  * If that doesn't work out -- if all bytes in the longer string
2278  * positioned at or past the offset of the smaller string's (first)
2279  * terminating NUL are actually representative of NUL bytes in the
2280  * authoritative binary string (perhaps with some *terminating* NUL bytes
2281  * towards the end of the longer string iff it happens to still be small)
2282  * -- then an authoritative tie-breaker will happen, and do the right
2283  * thing: explicitly consider string length.
2284  */
2285  if (sss->collate_c)
2286  memcpy(pres, authoritative_data, Min(len, sizeof(Datum)));
2287  else
2288  {
2289  Size bsize;
2290 #ifdef USE_ICU
2291  int32_t ulen = -1;
2292  UChar *uchar = NULL;
2293 #endif
2294 
2295  /*
2296  * We're not using the C collation, so fall back on strxfrm or ICU
2297  * analogs.
2298  */
2299 
2300  /* By convention, we use buffer 1 to store and NUL-terminate */
2301  if (len >= sss->buflen1)
2302  {
2303  pfree(sss->buf1);
2304  sss->buflen1 = Max(len + 1, Min(sss->buflen1 * 2, MaxAllocSize));
2305  sss->buf1 = palloc(sss->buflen1);
2306  }
2307 
2308  /* Might be able to reuse strxfrm() blob from last call */
2309  if (sss->last_len1 == len && sss->cache_blob &&
2310  memcmp(sss->buf1, authoritative_data, len) == 0)
2311  {
2312  memcpy(pres, sss->buf2, Min(sizeof(Datum), sss->last_len2));
2313  /* No change affecting cardinality, so no hashing required */
2314  goto done;
2315  }
2316 
2317  memcpy(sss->buf1, authoritative_data, len);
2318 
2319  /*
2320  * Just like strcoll(), strxfrm() expects a NUL-terminated string. Not
2321  * necessary for ICU, but doesn't hurt.
2322  */
2323  sss->buf1[len] = '\0';
2324  sss->last_len1 = len;
2325 
2326 #ifdef USE_ICU
2327  /* When using ICU and not UTF8, convert string to UChar. */
2328  if (sss->locale && sss->locale->provider == COLLPROVIDER_ICU &&
2330  ulen = icu_to_uchar(&uchar, sss->buf1, len);
2331 #endif
2332 
2333  /*
2334  * Loop: Call strxfrm() or ucol_getSortKey(), possibly enlarge buffer,
2335  * and try again. Both of these functions have the result buffer
2336  * content undefined if the result did not fit, so we need to retry
2337  * until everything fits, even though we only need the first few bytes
2338  * in the end. When using ucol_nextSortKeyPart(), however, we only
2339  * ask for as many bytes as we actually need.
2340  */
2341  for (;;)
2342  {
2343 #ifdef USE_ICU
2344  if (sss->locale && sss->locale->provider == COLLPROVIDER_ICU)
2345  {
2346  /*
2347  * When using UTF8, use the iteration interface so we only
2348  * need to produce as many bytes as we actually need.
2349  */
2350  if (GetDatabaseEncoding() == PG_UTF8)
2351  {
2352  UCharIterator iter;
2353  uint32_t state[2];
2354  UErrorCode status;
2355 
2356  uiter_setUTF8(&iter, sss->buf1, len);
2357  state[0] = state[1] = 0; /* won't need that again */
2358  status = U_ZERO_ERROR;
2359  bsize = ucol_nextSortKeyPart(sss->locale->info.icu.ucol,
2360  &iter,
2361  state,
2362  (uint8_t *) sss->buf2,
2363  Min(sizeof(Datum), sss->buflen2),
2364  &status);
2365  if (U_FAILURE(status))
2366  ereport(ERROR,
2367  (errmsg("sort key generation failed: %s",
2368  u_errorName(status))));
2369  }
2370  else
2371  bsize = ucol_getSortKey(sss->locale->info.icu.ucol,
2372  uchar, ulen,
2373  (uint8_t *) sss->buf2, sss->buflen2);
2374  }
2375  else
2376 #endif
2377 #ifdef HAVE_LOCALE_T
2378  if (sss->locale && sss->locale->provider == COLLPROVIDER_LIBC)
2379  bsize = strxfrm_l(sss->buf2, sss->buf1,
2380  sss->buflen2, sss->locale->info.lt);
2381  else
2382 #endif
2383  bsize = strxfrm(sss->buf2, sss->buf1, sss->buflen2);
2384 
2385  sss->last_len2 = bsize;
2386  if (bsize < sss->buflen2)
2387  break;
2388 
2389  /*
2390  * Grow buffer and retry.
2391  */
2392  pfree(sss->buf2);
2393  sss->buflen2 = Max(bsize + 1,
2394  Min(sss->buflen2 * 2, MaxAllocSize));
2395  sss->buf2 = palloc(sss->buflen2);
2396  }
2397 
2398  /*
2399  * Every Datum byte is always compared. This is safe because the
2400  * strxfrm() blob is itself NUL terminated, leaving no danger of
2401  * misinterpreting any NUL bytes not intended to be interpreted as
2402  * logically representing termination.
2403  *
2404  * (Actually, even if there were NUL bytes in the blob it would be
2405  * okay. See remarks on bytea case above.)
2406  */
2407  memcpy(pres, sss->buf2, Min(sizeof(Datum), bsize));
2408 
2409 #ifdef USE_ICU
2410  if (uchar)
2411  pfree(uchar);
2412 #endif
2413  }
2414 
2415  /*
2416  * Maintain approximate cardinality of both abbreviated keys and original,
2417  * authoritative keys using HyperLogLog. Used as cheap insurance against
2418  * the worst case, where we do many string transformations for no saving
2419  * in full strcoll()-based comparisons. These statistics are used by
2420  * varstr_abbrev_abort().
2421  *
2422  * First, Hash key proper, or a significant fraction of it. Mix in length
2423  * in order to compensate for cases where differences are past
2424  * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
2425  */
2426  hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
2427  Min(len, PG_CACHE_LINE_SIZE)));
2428 
2429  if (len > PG_CACHE_LINE_SIZE)
2430  hash ^= DatumGetUInt32(hash_uint32((uint32) len));
2431 
2432  addHyperLogLog(&sss->full_card, hash);
2433 
2434  /* Hash abbreviated key */
2435 #if SIZEOF_DATUM == 8
2436  {
2437  uint32 lohalf,
2438  hihalf;
2439 
2440  lohalf = (uint32) res;
2441  hihalf = (uint32) (res >> 32);
2442  hash = DatumGetUInt32(hash_uint32(lohalf ^ hihalf));
2443  }
2444 #else /* SIZEOF_DATUM != 8 */
2445  hash = DatumGetUInt32(hash_uint32((uint32) res));
2446 #endif
2447 
2448  addHyperLogLog(&sss->abbr_card, hash);
2449 
2450  /* Cache result, perhaps saving an expensive strxfrm() call next time */
2451  sss->cache_blob = true;
2452 done:
2453 
2454  /*
2455  * Byteswap on little-endian machines.
2456  *
2457  * This is needed so that varstrcmp_abbrev() (an unsigned integer 3-way
2458  * comparator) works correctly on all platforms. If we didn't do this,
2459  * the comparator would have to call memcmp() with a pair of pointers to
2460  * the first byte of each abbreviated key, which is slower.
2461  */
2462  res = DatumBigEndianToNative(res);
2463 
2464  /* Don't leak memory here */
2465  if (PointerGetDatum(authoritative) != original)
2466  pfree(authoritative);
2467 
2468  return res;
2469 }
2470 
2471 /*
2472  * Callback for estimating effectiveness of abbreviated key optimization, using
2473  * heuristic rules. Returns value indicating if the abbreviation optimization
2474  * should be aborted, based on its projected effectiveness.
2475  */
2476 static bool
2477 varstr_abbrev_abort(int memtupcount, SortSupport ssup)
2478 {
2480  double abbrev_distinct,
2481  key_distinct;
2482 
2483  Assert(ssup->abbreviate);
2484 
2485  /* Have a little patience */
2486  if (memtupcount < 100)
2487  return false;
2488 
2489  abbrev_distinct = estimateHyperLogLog(&sss->abbr_card);
2490  key_distinct = estimateHyperLogLog(&sss->full_card);
2491 
2492  /*
2493  * Clamp cardinality estimates to at least one distinct value. While
2494  * NULLs are generally disregarded, if only NULL values were seen so far,
2495  * that might misrepresent costs if we failed to clamp.
2496  */
2497  if (abbrev_distinct <= 1.0)
2498  abbrev_distinct = 1.0;
2499 
2500  if (key_distinct <= 1.0)
2501  key_distinct = 1.0;
2502 
2503  /*
2504  * In the worst case all abbreviated keys are identical, while at the same
2505  * time there are differences within full key strings not captured in
2506  * abbreviations.
2507  */
2508 #ifdef TRACE_SORT
2509  if (trace_sort)
2510  {
2511  double norm_abbrev_card = abbrev_distinct / (double) memtupcount;
2512 
2513  elog(LOG, "varstr_abbrev: abbrev_distinct after %d: %f "
2514  "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
2515  memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
2516  sss->prop_card);
2517  }
2518 #endif
2519 
2520  /*
2521  * If the number of distinct abbreviated keys approximately matches the
2522  * number of distinct authoritative original keys, that's reason enough to
2523  * proceed. We can win even with a very low cardinality set if most
2524  * tie-breakers only memcmp(). This is by far the most important
2525  * consideration.
2526  *
2527  * While comparisons that are resolved at the abbreviated key level are
2528  * considerably cheaper than tie-breakers resolved with memcmp(), both of
2529  * those two outcomes are so much cheaper than a full strcoll() once
2530  * sorting is underway that it doesn't seem worth it to weigh abbreviated
2531  * cardinality against the overall size of the set in order to more
2532  * accurately model costs. Assume that an abbreviated comparison, and an
2533  * abbreviated comparison with a cheap memcmp()-based authoritative
2534  * resolution are equivalent.
2535  */
2536  if (abbrev_distinct > key_distinct * sss->prop_card)
2537  {
2538  /*
2539  * When we have exceeded 10,000 tuples, decay required cardinality
2540  * aggressively for next call.
2541  *
2542  * This is useful because the number of comparisons required on
2543  * average increases at a linearithmic rate, and at roughly 10,000
2544  * tuples that factor will start to dominate over the linear costs of
2545  * string transformation (this is a conservative estimate). The decay
2546  * rate is chosen to be a little less aggressive than halving -- which
2547  * (since we're called at points at which memtupcount has doubled)
2548  * would never see the cost model actually abort past the first call
2549  * following a decay. This decay rate is mostly a precaution against
2550  * a sudden, violent swing in how well abbreviated cardinality tracks
2551  * full key cardinality. The decay also serves to prevent a marginal
2552  * case from being aborted too late, when too much has already been
2553  * invested in string transformation.
2554  *
2555  * It's possible for sets of several million distinct strings with
2556  * mere tens of thousands of distinct abbreviated keys to still
2557  * benefit very significantly. This will generally occur provided
2558  * each abbreviated key is a proxy for a roughly uniform number of the
2559  * set's full keys. If it isn't so, we hope to catch that early and
2560  * abort. If it isn't caught early, by the time the problem is
2561  * apparent it's probably not worth aborting.
2562  */
2563  if (memtupcount > 10000)
2564  sss->prop_card *= 0.65;
2565 
2566  return false;
2567  }
2568 
2569  /*
2570  * Abort abbreviation strategy.
2571  *
2572  * The worst case, where all abbreviated keys are identical while all
2573  * original strings differ will typically only see a regression of about
2574  * 10% in execution time for small to medium sized lists of strings.
2575  * Whereas on modern CPUs where cache stalls are the dominant cost, we can
2576  * often expect very large improvements, particularly with sets of strings
2577  * of moderately high to high abbreviated cardinality. There is little to
2578  * lose but much to gain, which our strategy reflects.
2579  */
2580 #ifdef TRACE_SORT
2581  if (trace_sort)
2582  elog(LOG, "varstr_abbrev: aborted abbreviation at %d "
2583  "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
2584  memtupcount, abbrev_distinct, key_distinct, sss->prop_card);
2585 #endif
2586 
2587  return true;
2588 }
2589 
2590 Datum
2592 {
2593  text *arg1 = PG_GETARG_TEXT_PP(0);
2594  text *arg2 = PG_GETARG_TEXT_PP(1);
2595  text *result;
2596 
2597  result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) > 0) ? arg1 : arg2);
2598 
2599  PG_RETURN_TEXT_P(result);
2600 }
2601 
2602 Datum
2604 {
2605  text *arg1 = PG_GETARG_TEXT_PP(0);
2606  text *arg2 = PG_GETARG_TEXT_PP(1);
2607  text *result;
2608 
2609  result = ((text_cmp(arg1, arg2, PG_GET_COLLATION()) < 0) ? arg1 : arg2);
2610 
2611  PG_RETURN_TEXT_P(result);
2612 }
2613 
2614 
2615 /*
2616  * The following operators support character-by-character comparison
2617  * of text datums, to allow building indexes suitable for LIKE clauses.
2618  * Note that the regular texteq/textne comparison operators, and regular
2619  * support functions 1 and 2 with "C" collation are assumed to be
2620  * compatible with these!
2621  */
2622 
2623 static int
2625 {
2626  int result;
2627  int len1,
2628  len2;
2629 
2630  len1 = VARSIZE_ANY_EXHDR(arg1);
2631  len2 = VARSIZE_ANY_EXHDR(arg2);
2632 
2633  result = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
2634  if (result != 0)
2635  return result;
2636  else if (len1 < len2)
2637  return -1;
2638  else if (len1 > len2)
2639  return 1;
2640  else
2641  return 0;
2642 }
2643 
2644 
2645 Datum
2647 {
2648  text *arg1 = PG_GETARG_TEXT_PP(0);
2649  text *arg2 = PG_GETARG_TEXT_PP(1);
2650  int result;
2651 
2652  result = internal_text_pattern_compare(arg1, arg2);
2653 
2654  PG_FREE_IF_COPY(arg1, 0);
2655  PG_FREE_IF_COPY(arg2, 1);
2656 
2657  PG_RETURN_BOOL(result < 0);
2658 }
2659 
2660 
2661 Datum
2663 {
2664  text *arg1 = PG_GETARG_TEXT_PP(0);
2665  text *arg2 = PG_GETARG_TEXT_PP(1);
2666  int result;
2667 
2668  result = internal_text_pattern_compare(arg1, arg2);
2669 
2670  PG_FREE_IF_COPY(arg1, 0);
2671  PG_FREE_IF_COPY(arg2, 1);
2672 
2673  PG_RETURN_BOOL(result <= 0);
2674 }
2675 
2676 
2677 Datum
2679 {
2680  text *arg1 = PG_GETARG_TEXT_PP(0);
2681  text *arg2 = PG_GETARG_TEXT_PP(1);
2682  int result;
2683 
2684  result = internal_text_pattern_compare(arg1, arg2);
2685 
2686  PG_FREE_IF_COPY(arg1, 0);
2687  PG_FREE_IF_COPY(arg2, 1);
2688 
2689  PG_RETURN_BOOL(result >= 0);
2690 }
2691 
2692 
2693 Datum
2695 {
2696  text *arg1 = PG_GETARG_TEXT_PP(0);
2697  text *arg2 = PG_GETARG_TEXT_PP(1);
2698  int result;
2699 
2700  result = internal_text_pattern_compare(arg1, arg2);
2701 
2702  PG_FREE_IF_COPY(arg1, 0);
2703  PG_FREE_IF_COPY(arg2, 1);
2704 
2705  PG_RETURN_BOOL(result > 0);
2706 }
2707 
2708 
2709 Datum
2711 {
2712  text *arg1 = PG_GETARG_TEXT_PP(0);
2713  text *arg2 = PG_GETARG_TEXT_PP(1);
2714  int result;
2715 
2716  result = internal_text_pattern_compare(arg1, arg2);
2717 
2718  PG_FREE_IF_COPY(arg1, 0);
2719  PG_FREE_IF_COPY(arg2, 1);
2720 
2721  PG_RETURN_INT32(result);
2722 }
2723 
2724 
2725 Datum
2727 {
2729  MemoryContext oldcontext;
2730 
2731  oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
2732 
2733  /* Use generic string SortSupport, forcing "C" collation */
2734  varstr_sortsupport(ssup, C_COLLATION_OID, false);
2735 
2736  MemoryContextSwitchTo(oldcontext);
2737 
2738  PG_RETURN_VOID();
2739 }
2740 
2741 
2742 /*-------------------------------------------------------------
2743  * byteaoctetlen
2744  *
2745  * get the number of bytes contained in an instance of type 'bytea'
2746  *-------------------------------------------------------------
2747  */
2748 Datum
2750 {
2751  Datum str = PG_GETARG_DATUM(0);
2752 
2753  /* We need not detoast the input at all */
2755 }
2756 
2757 /*
2758  * byteacat -
2759  * takes two bytea* and returns a bytea* that is the concatenation of
2760  * the two.
2761  *
2762  * Cloned from textcat and modified as required.
2763  */
2764 Datum
2766 {
2767  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2768  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2769 
2771 }
2772 
2773 /*
2774  * bytea_catenate
2775  * Guts of byteacat(), broken out so it can be used by other functions
2776  *
2777  * Arguments can be in short-header form, but not compressed or out-of-line
2778  */
2779 static bytea *
2781 {
2782  bytea *result;
2783  int len1,
2784  len2,
2785  len;
2786  char *ptr;
2787 
2788  len1 = VARSIZE_ANY_EXHDR(t1);
2789  len2 = VARSIZE_ANY_EXHDR(t2);
2790 
2791  /* paranoia ... probably should throw error instead? */
2792  if (len1 < 0)
2793  len1 = 0;
2794  if (len2 < 0)
2795  len2 = 0;
2796 
2797  len = len1 + len2 + VARHDRSZ;
2798  result = (bytea *) palloc(len);
2799 
2800  /* Set size of result string... */
2801  SET_VARSIZE(result, len);
2802 
2803  /* Fill data field of result string... */
2804  ptr = VARDATA(result);
2805  if (len1 > 0)
2806  memcpy(ptr, VARDATA_ANY(t1), len1);
2807  if (len2 > 0)
2808  memcpy(ptr + len1, VARDATA_ANY(t2), len2);
2809 
2810  return result;
2811 }
2812 
2813 #define PG_STR_GET_BYTEA(str_) \
2814  DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
2815 
2816 /*
2817  * bytea_substr()
2818  * Return a substring starting at the specified position.
2819  * Cloned from text_substr and modified as required.
2820  *
2821  * Input:
2822  * - string
2823  * - starting position (is one-based)
2824  * - string length (optional)
2825  *
2826  * If the starting position is zero or less, then return from the start of the string
2827  * adjusting the length to be consistent with the "negative start" per SQL.
2828  * If the length is less than zero, an ERROR is thrown. If no third argument
2829  * (length) is provided, the length to the end of the string is assumed.
2830  */
2831 Datum
2833 {
2835  PG_GETARG_INT32(1),
2836  PG_GETARG_INT32(2),
2837  false));
2838 }
2839 
2840 /*
2841  * bytea_substr_no_len -
2842  * Wrapper to avoid opr_sanity failure due to
2843  * one function accepting a different number of args.
2844  */
2845 Datum
2847 {
2849  PG_GETARG_INT32(1),
2850  -1,
2851  true));
2852 }
2853 
2854 static bytea *
2856  int S,
2857  int L,
2858  bool length_not_specified)
2859 {
2860  int S1; /* adjusted start position */
2861  int L1; /* adjusted substring length */
2862 
2863  S1 = Max(S, 1);
2864 
2865  if (length_not_specified)
2866  {
2867  /*
2868  * Not passed a length - DatumGetByteaPSlice() grabs everything to the
2869  * end of the string if we pass it a negative value for length.
2870  */
2871  L1 = -1;
2872  }
2873  else
2874  {
2875  /* end position */
2876  int E = S + L;
2877 
2878  /*
2879  * A negative value for L is the only way for the end position to be
2880  * before the start. SQL99 says to throw an error.
2881  */
2882  if (E < S)
2883  ereport(ERROR,
2884  (errcode(ERRCODE_SUBSTRING_ERROR),
2885  errmsg("negative substring length not allowed")));
2886 
2887  /*
2888  * A zero or negative value for the end position can happen if the
2889  * start was negative or one. SQL99 says to return a zero-length
2890  * string.
2891  */
2892  if (E < 1)
2893  return PG_STR_GET_BYTEA("");
2894 
2895  L1 = E - S1;
2896  }
2897 
2898  /*
2899  * If the start position is past the end of the string, SQL99 says to
2900  * return a zero-length string -- DatumGetByteaPSlice() will do that for
2901  * us. Convert to zero-based starting position
2902  */
2903  return DatumGetByteaPSlice(str, S1 - 1, L1);
2904 }
2905 
2906 /*
2907  * byteaoverlay
2908  * Replace specified substring of first string with second
2909  *
2910  * The SQL standard defines OVERLAY() in terms of substring and concatenation.
2911  * This code is a direct implementation of what the standard says.
2912  */
2913 Datum
2915 {
2916  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2917  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2918  int sp = PG_GETARG_INT32(2); /* substring start position */
2919  int sl = PG_GETARG_INT32(3); /* substring length */
2920 
2921  PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
2922 }
2923 
2924 Datum
2926 {
2927  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2928  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2929  int sp = PG_GETARG_INT32(2); /* substring start position */
2930  int sl;
2931 
2932  sl = VARSIZE_ANY_EXHDR(t2); /* defaults to length(t2) */
2933  PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
2934 }
2935 
2936 static bytea *
2937 bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
2938 {
2939  bytea *result;
2940  bytea *s1;
2941  bytea *s2;
2942  int sp_pl_sl;
2943 
2944  /*
2945  * Check for possible integer-overflow cases. For negative sp, throw a
2946  * "substring length" error because that's what should be expected
2947  * according to the spec's definition of OVERLAY().
2948  */
2949  if (sp <= 0)
2950  ereport(ERROR,
2951  (errcode(ERRCODE_SUBSTRING_ERROR),
2952  errmsg("negative substring length not allowed")));
2953  sp_pl_sl = sp + sl;
2954  if (sp_pl_sl <= sl)
2955  ereport(ERROR,
2956  (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
2957  errmsg("integer out of range")));
2958 
2959  s1 = bytea_substring(PointerGetDatum(t1), 1, sp - 1, false);
2960  s2 = bytea_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
2961  result = bytea_catenate(s1, t2);
2962  result = bytea_catenate(result, s2);
2963 
2964  return result;
2965 }
2966 
2967 /*
2968  * byteapos -
2969  * Return the position of the specified substring.
2970  * Implements the SQL POSITION() function.
2971  * Cloned from textpos and modified as required.
2972  */
2973 Datum
2975 {
2976  bytea *t1 = PG_GETARG_BYTEA_PP(0);
2977  bytea *t2 = PG_GETARG_BYTEA_PP(1);
2978  int pos;
2979  int px,
2980  p;
2981  int len1,
2982  len2;
2983  char *p1,
2984  *p2;
2985 
2986  len1 = VARSIZE_ANY_EXHDR(t1);
2987  len2 = VARSIZE_ANY_EXHDR(t2);
2988 
2989  if (len2 <= 0)
2990  PG_RETURN_INT32(1); /* result for empty pattern */
2991 
2992  p1 = VARDATA_ANY(t1);
2993  p2 = VARDATA_ANY(t2);
2994 
2995  pos = 0;
2996  px = (len1 - len2);
2997  for (p = 0; p <= px; p++)
2998  {
2999  if ((*p2 == *p1) && (memcmp(p1, p2, len2) == 0))
3000  {
3001  pos = p + 1;
3002  break;
3003  };
3004  p1++;
3005  };
3006 
3007  PG_RETURN_INT32(pos);
3008 }
3009 
3010 /*-------------------------------------------------------------
3011  * byteaGetByte
3012  *
3013  * this routine treats "bytea" as an array of bytes.
3014  * It returns the Nth byte (a number between 0 and 255).
3015  *-------------------------------------------------------------
3016  */
3017 Datum
3019 {
3020  bytea *v = PG_GETARG_BYTEA_PP(0);
3021  int32 n = PG_GETARG_INT32(1);
3022  int len;
3023  int byte;
3024 
3025  len = VARSIZE_ANY_EXHDR(v);
3026 
3027  if (n < 0 || n >= len)
3028  ereport(ERROR,
3029  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3030  errmsg("index %d out of valid range, 0..%d",
3031  n, len - 1)));
3032 
3033  byte = ((unsigned char *) VARDATA_ANY(v))[n];
3034 
3035  PG_RETURN_INT32(byte);
3036 }
3037 
3038 /*-------------------------------------------------------------
3039  * byteaGetBit
3040  *
3041  * This routine treats a "bytea" type like an array of bits.
3042  * It returns the value of the Nth bit (0 or 1).
3043  *
3044  *-------------------------------------------------------------
3045  */
3046 Datum
3048 {
3049  bytea *v = PG_GETARG_BYTEA_PP(0);
3050  int32 n = PG_GETARG_INT32(1);
3051  int byteNo,
3052  bitNo;
3053  int len;
3054  int byte;
3055 
3056  len = VARSIZE_ANY_EXHDR(v);
3057 
3058  if (n < 0 || n >= len * 8)
3059  ereport(ERROR,
3060  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3061  errmsg("index %d out of valid range, 0..%d",
3062  n, len * 8 - 1)));
3063 
3064  byteNo = n / 8;
3065  bitNo = n % 8;
3066 
3067  byte = ((unsigned char *) VARDATA_ANY(v))[byteNo];
3068 
3069  if (byte & (1 << bitNo))
3070  PG_RETURN_INT32(1);
3071  else
3072  PG_RETURN_INT32(0);
3073 }
3074 
3075 /*-------------------------------------------------------------
3076  * byteaSetByte
3077  *
3078  * Given an instance of type 'bytea' creates a new one with
3079  * the Nth byte set to the given value.
3080  *
3081  *-------------------------------------------------------------
3082  */
3083 Datum
3085 {
3086  bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3087  int32 n = PG_GETARG_INT32(1);
3088  int32 newByte = PG_GETARG_INT32(2);
3089  int len;
3090 
3091  len = VARSIZE(res) - VARHDRSZ;
3092 
3093  if (n < 0 || n >= len)
3094  ereport(ERROR,
3095  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3096  errmsg("index %d out of valid range, 0..%d",
3097  n, len - 1)));
3098 
3099  /*
3100  * Now set the byte.
3101  */
3102  ((unsigned char *) VARDATA(res))[n] = newByte;
3103 
3104  PG_RETURN_BYTEA_P(res);
3105 }
3106 
3107 /*-------------------------------------------------------------
3108  * byteaSetBit
3109  *
3110  * Given an instance of type 'bytea' creates a new one with
3111  * the Nth bit set to the given value.
3112  *
3113  *-------------------------------------------------------------
3114  */
3115 Datum
3117 {
3118  bytea *res = PG_GETARG_BYTEA_P_COPY(0);
3119  int32 n = PG_GETARG_INT32(1);
3120  int32 newBit = PG_GETARG_INT32(2);
3121  int len;
3122  int oldByte,
3123  newByte;
3124  int byteNo,
3125  bitNo;
3126 
3127  len = VARSIZE(res) - VARHDRSZ;
3128 
3129  if (n < 0 || n >= len * 8)
3130  ereport(ERROR,
3131  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
3132  errmsg("index %d out of valid range, 0..%d",
3133  n, len * 8 - 1)));
3134 
3135  byteNo = n / 8;
3136  bitNo = n % 8;
3137 
3138  /*
3139  * sanity check!
3140  */
3141  if (newBit != 0 && newBit != 1)
3142  ereport(ERROR,
3143  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3144  errmsg("new bit must be 0 or 1")));
3145 
3146  /*
3147  * Update the byte.
3148  */
3149  oldByte = ((unsigned char *) VARDATA(res))[byteNo];
3150 
3151  if (newBit == 0)
3152  newByte = oldByte & (~(1 << bitNo));
3153  else
3154  newByte = oldByte | (1 << bitNo);
3155 
3156  ((unsigned char *) VARDATA(res))[byteNo] = newByte;
3157 
3158  PG_RETURN_BYTEA_P(res);
3159 }
3160 
3161 
3162 /* text_name()
3163  * Converts a text type to a Name type.
3164  */
3165 Datum
3167 {
3168  text *s = PG_GETARG_TEXT_PP(0);
3169  Name result;
3170  int len;
3171 
3172  len = VARSIZE_ANY_EXHDR(s);
3173 
3174  /* Truncate oversize input */
3175  if (len >= NAMEDATALEN)
3176  len = pg_mbcliplen(VARDATA_ANY(s), len, NAMEDATALEN - 1);
3177 
3178  /* We use palloc0 here to ensure result is zero-padded */
3179  result = (Name) palloc0(NAMEDATALEN);
3180  memcpy(NameStr(*result), VARDATA_ANY(s), len);
3181 
3182  PG_RETURN_NAME(result);
3183 }
3184 
3185 /* name_text()
3186  * Converts a Name type to a text type.
3187  */
3188 Datum
3190 {
3191  Name s = PG_GETARG_NAME(0);
3192 
3194 }
3195 
3196 
3197 /*
3198  * textToQualifiedNameList - convert a text object to list of names
3199  *
3200  * This implements the input parsing needed by nextval() and other
3201  * functions that take a text parameter representing a qualified name.
3202  * We split the name at dots, downcase if not double-quoted, and
3203  * truncate names if they're too long.
3204  */
3205 List *
3207 {
3208  char *rawname;
3209  List *result = NIL;
3210  List *namelist;
3211  ListCell *l;
3212 
3213  /* Convert to C string (handles possible detoasting). */
3214  /* Note we rely on being able to modify rawname below. */
3215  rawname = text_to_cstring(textval);
3216 
3217  if (!SplitIdentifierString(rawname, '.', &namelist))
3218  ereport(ERROR,
3219  (errcode(ERRCODE_INVALID_NAME),
3220  errmsg("invalid name syntax")));
3221 
3222  if (namelist == NIL)
3223  ereport(ERROR,
3224  (errcode(ERRCODE_INVALID_NAME),
3225  errmsg("invalid name syntax")));
3226 
3227  foreach(l, namelist)
3228  {
3229  char *curname = (char *) lfirst(l);
3230 
3231  result = lappend(result, makeString(pstrdup(curname)));
3232  }
3233 
3234  pfree(rawname);
3235  list_free(namelist);
3236 
3237  return result;
3238 }
3239 
3240 /*
3241  * SplitIdentifierString --- parse a string containing identifiers
3242  *
3243  * This is the guts of textToQualifiedNameList, and is exported for use in
3244  * other situations such as parsing GUC variables. In the GUC case, it's
3245  * important to avoid memory leaks, so the API is designed to minimize the
3246  * amount of stuff that needs to be allocated and freed.
3247  *
3248  * Inputs:
3249  * rawstring: the input string; must be overwritable! On return, it's
3250  * been modified to contain the separated identifiers.
3251  * separator: the separator punctuation expected between identifiers
3252  * (typically '.' or ','). Whitespace may also appear around
3253  * identifiers.
3254  * Outputs:
3255  * namelist: filled with a palloc'd list of pointers to identifiers within
3256  * rawstring. Caller should list_free() this even on error return.
3257  *
3258  * Returns true if okay, false if there is a syntax error in the string.
3259  *
3260  * Note that an empty string is considered okay here, though not in
3261  * textToQualifiedNameList.
3262  */
3263 bool
3264 SplitIdentifierString(char *rawstring, char separator,
3265  List **namelist)
3266 {
3267  char *nextp = rawstring;
3268  bool done = false;
3269 
3270  *namelist = NIL;
3271 
3272  while (scanner_isspace(*nextp))
3273  nextp++; /* skip leading whitespace */
3274 
3275  if (*nextp == '\0')
3276  return true; /* allow empty string */
3277 
3278  /* At the top of the loop, we are at start of a new identifier. */
3279  do
3280  {
3281  char *curname;
3282  char *endp;
3283 
3284  if (*nextp == '"')
3285  {
3286  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3287  curname = nextp + 1;
3288  for (;;)
3289  {
3290  endp = strchr(nextp + 1, '"');
3291  if (endp == NULL)
3292  return false; /* mismatched quotes */
3293  if (endp[1] != '"')
3294  break; /* found end of quoted name */
3295  /* Collapse adjacent quotes into one quote, and look again */
3296  memmove(endp, endp + 1, strlen(endp));
3297  nextp = endp;
3298  }
3299  /* endp now points at the terminating quote */
3300  nextp = endp + 1;
3301  }
3302  else
3303  {
3304  /* Unquoted name --- extends to separator or whitespace */
3305  char *downname;
3306  int len;
3307 
3308  curname = nextp;
3309  while (*nextp && *nextp != separator &&
3310  !scanner_isspace(*nextp))
3311  nextp++;
3312  endp = nextp;
3313  if (curname == nextp)
3314  return false; /* empty unquoted name not allowed */
3315 
3316  /*
3317  * Downcase the identifier, using same code as main lexer does.
3318  *
3319  * XXX because we want to overwrite the input in-place, we cannot
3320  * support a downcasing transformation that increases the string
3321  * length. This is not a problem given the current implementation
3322  * of downcase_truncate_identifier, but we'll probably have to do
3323  * something about this someday.
3324  */
3325  len = endp - curname;
3326  downname = downcase_truncate_identifier(curname, len, false);
3327  Assert(strlen(downname) <= len);
3328  strncpy(curname, downname, len); /* strncpy is required here */
3329  pfree(downname);
3330  }
3331 
3332  while (scanner_isspace(*nextp))
3333  nextp++; /* skip trailing whitespace */
3334 
3335  if (*nextp == separator)
3336  {
3337  nextp++;
3338  while (scanner_isspace(*nextp))
3339  nextp++; /* skip leading whitespace for next */
3340  /* we expect another name, so done remains false */
3341  }
3342  else if (*nextp == '\0')
3343  done = true;
3344  else
3345  return false; /* invalid syntax */
3346 
3347  /* Now safe to overwrite separator with a null */
3348  *endp = '\0';
3349 
3350  /* Truncate name if it's overlength */
3351  truncate_identifier(curname, strlen(curname), false);
3352 
3353  /*
3354  * Finished isolating current name --- add it to list
3355  */
3356  *namelist = lappend(*namelist, curname);
3357 
3358  /* Loop back if we didn't reach end of string */
3359  } while (!done);
3360 
3361  return true;
3362 }
3363 
3364 
3365 /*
3366  * SplitDirectoriesString --- parse a string containing file/directory names
3367  *
3368  * This works fine on file names too; the function name is historical.
3369  *
3370  * This is similar to SplitIdentifierString, except that the parsing
3371  * rules are meant to handle pathnames instead of identifiers: there is
3372  * no downcasing, embedded spaces are allowed, the max length is MAXPGPATH-1,
3373  * and we apply canonicalize_path() to each extracted string. Because of the
3374  * last, the returned strings are separately palloc'd rather than being
3375  * pointers into rawstring --- but we still scribble on rawstring.
3376  *
3377  * Inputs:
3378  * rawstring: the input string; must be modifiable!
3379  * separator: the separator punctuation expected between directories
3380  * (typically ',' or ';'). Whitespace may also appear around
3381  * directories.
3382  * Outputs:
3383  * namelist: filled with a palloc'd list of directory names.
3384  * Caller should list_free_deep() this even on error return.
3385  *
3386  * Returns true if okay, false if there is a syntax error in the string.
3387  *
3388  * Note that an empty string is considered okay here.
3389  */
3390 bool
3391 SplitDirectoriesString(char *rawstring, char separator,
3392  List **namelist)
3393 {
3394  char *nextp = rawstring;
3395  bool done = false;
3396 
3397  *namelist = NIL;
3398 
3399  while (scanner_isspace(*nextp))
3400  nextp++; /* skip leading whitespace */
3401 
3402  if (*nextp == '\0')
3403  return true; /* allow empty string */
3404 
3405  /* At the top of the loop, we are at start of a new directory. */
3406  do
3407  {
3408  char *curname;
3409  char *endp;
3410 
3411  if (*nextp == '"')
3412  {
3413  /* Quoted name --- collapse quote-quote pairs */
3414  curname = nextp + 1;
3415  for (;;)
3416  {
3417  endp = strchr(nextp + 1, '"');
3418  if (endp == NULL)
3419  return false; /* mismatched quotes */
3420  if (endp[1] != '"')
3421  break; /* found end of quoted name */
3422  /* Collapse adjacent quotes into one quote, and look again */
3423  memmove(endp, endp + 1, strlen(endp));
3424  nextp = endp;
3425  }
3426  /* endp now points at the terminating quote */
3427  nextp = endp + 1;
3428  }
3429  else
3430  {
3431  /* Unquoted name --- extends to separator or end of string */
3432  curname = endp = nextp;
3433  while (*nextp && *nextp != separator)
3434  {
3435  /* trailing whitespace should not be included in name */
3436  if (!scanner_isspace(*nextp))
3437  endp = nextp + 1;
3438  nextp++;
3439  }
3440  if (curname == endp)
3441  return false; /* empty unquoted name not allowed */
3442  }
3443 
3444  while (scanner_isspace(*nextp))
3445  nextp++; /* skip trailing whitespace */
3446 
3447  if (*nextp == separator)
3448  {
3449  nextp++;
3450  while (scanner_isspace(*nextp))
3451  nextp++; /* skip leading whitespace for next */
3452  /* we expect another name, so done remains false */
3453  }
3454  else if (*nextp == '\0')
3455  done = true;
3456  else
3457  return false; /* invalid syntax */
3458 
3459  /* Now safe to overwrite separator with a null */
3460  *endp = '\0';
3461 
3462  /* Truncate path if it's overlength */
3463  if (strlen(curname) >= MAXPGPATH)
3464  curname[MAXPGPATH - 1] = '\0';
3465 
3466  /*
3467  * Finished isolating current name --- add it to list
3468  */
3469  curname = pstrdup(curname);
3470  canonicalize_path(curname);
3471  *namelist = lappend(*namelist, curname);
3472 
3473  /* Loop back if we didn't reach end of string */
3474  } while (!done);
3475 
3476  return true;
3477 }
3478 
3479 
3480 /*****************************************************************************
3481  * Comparison Functions used for bytea
3482  *
3483  * Note: btree indexes need these routines not to leak memory; therefore,
3484  * be careful to free working copies of toasted datums. Most places don't
3485  * need to be so careful.
3486  *****************************************************************************/
3487 
3488 Datum
3490 {
3491  Datum arg1 = PG_GETARG_DATUM(0);
3492  Datum arg2 = PG_GETARG_DATUM(1);
3493  bool result;
3494  Size len1,
3495  len2;
3496 
3497  /*
3498  * We can use a fast path for unequal lengths, which might save us from
3499  * having to detoast one or both values.
3500  */
3501  len1 = toast_raw_datum_size(arg1);
3502  len2 = toast_raw_datum_size(arg2);
3503  if (len1 != len2)
3504  result = false;
3505  else
3506  {
3507  bytea *barg1 = DatumGetByteaPP(arg1);
3508  bytea *barg2 = DatumGetByteaPP(arg2);
3509 
3510  result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3511  len1 - VARHDRSZ) == 0);
3512 
3513  PG_FREE_IF_COPY(barg1, 0);
3514  PG_FREE_IF_COPY(barg2, 1);
3515  }
3516 
3517  PG_RETURN_BOOL(result);
3518 }
3519 
3520 Datum
3522 {
3523  Datum arg1 = PG_GETARG_DATUM(0);
3524  Datum arg2 = PG_GETARG_DATUM(1);
3525  bool result;
3526  Size len1,
3527  len2;
3528 
3529  /*
3530  * We can use a fast path for unequal lengths, which might save us from
3531  * having to detoast one or both values.
3532  */
3533  len1 = toast_raw_datum_size(arg1);
3534  len2 = toast_raw_datum_size(arg2);
3535  if (len1 != len2)
3536  result = true;
3537  else
3538  {
3539  bytea *barg1 = DatumGetByteaPP(arg1);
3540  bytea *barg2 = DatumGetByteaPP(arg2);
3541 
3542  result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
3543  len1 - VARHDRSZ) != 0);
3544 
3545  PG_FREE_IF_COPY(barg1, 0);
3546  PG_FREE_IF_COPY(barg2, 1);
3547  }
3548 
3549  PG_RETURN_BOOL(result);
3550 }
3551 
3552 Datum
3554 {
3555  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3556  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3557  int len1,
3558  len2;
3559  int cmp;
3560 
3561  len1 = VARSIZE_ANY_EXHDR(arg1);
3562  len2 = VARSIZE_ANY_EXHDR(arg2);
3563 
3564  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3565 
3566  PG_FREE_IF_COPY(arg1, 0);
3567  PG_FREE_IF_COPY(arg2, 1);
3568 
3569  PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 < len2)));
3570 }
3571 
3572 Datum
3574 {
3575  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3576  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3577  int len1,
3578  len2;
3579  int cmp;
3580 
3581  len1 = VARSIZE_ANY_EXHDR(arg1);
3582  len2 = VARSIZE_ANY_EXHDR(arg2);
3583 
3584  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3585 
3586  PG_FREE_IF_COPY(arg1, 0);
3587  PG_FREE_IF_COPY(arg2, 1);
3588 
3589  PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 <= len2)));
3590 }
3591 
3592 Datum
3594 {
3595  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3596  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3597  int len1,
3598  len2;
3599  int cmp;
3600 
3601  len1 = VARSIZE_ANY_EXHDR(arg1);
3602  len2 = VARSIZE_ANY_EXHDR(arg2);
3603 
3604  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3605 
3606  PG_FREE_IF_COPY(arg1, 0);
3607  PG_FREE_IF_COPY(arg2, 1);
3608 
3609  PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 > len2)));
3610 }
3611 
3612 Datum
3614 {
3615  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3616  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3617  int len1,
3618  len2;
3619  int cmp;
3620 
3621  len1 = VARSIZE_ANY_EXHDR(arg1);
3622  len2 = VARSIZE_ANY_EXHDR(arg2);
3623 
3624  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3625 
3626  PG_FREE_IF_COPY(arg1, 0);
3627  PG_FREE_IF_COPY(arg2, 1);
3628 
3629  PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 >= len2)));
3630 }
3631 
3632 Datum
3634 {
3635  bytea *arg1 = PG_GETARG_BYTEA_PP(0);
3636  bytea *arg2 = PG_GETARG_BYTEA_PP(1);
3637  int len1,
3638  len2;
3639  int cmp;
3640 
3641  len1 = VARSIZE_ANY_EXHDR(arg1);
3642  len2 = VARSIZE_ANY_EXHDR(arg2);
3643 
3644  cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
3645  if ((cmp == 0) && (len1 != len2))
3646  cmp = (len1 < len2) ? -1 : 1;
3647 
3648  PG_FREE_IF_COPY(arg1, 0);
3649  PG_FREE_IF_COPY(arg2, 1);
3650 
3651  PG_RETURN_INT32(cmp);
3652 }
3653 
3654 Datum
3656 {
3658  MemoryContext oldcontext;
3659 
3660  oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
3661 
3662  /* Use generic string SortSupport, forcing "C" collation */
3663  varstr_sortsupport(ssup, C_COLLATION_OID, false);
3664 
3665  MemoryContextSwitchTo(oldcontext);
3666 
3667  PG_RETURN_VOID();
3668 }
3669 
3670 /*
3671  * appendStringInfoText
3672  *
3673  * Append a text to str.
3674  * Like appendStringInfoString(str, text_to_cstring(t)) but faster.
3675  */
3676 static void
3678 {
3680 }
3681 
3682 /*
3683  * replace_text
3684  * replace all occurrences of 'old_sub_str' in 'orig_str'
3685  * with 'new_sub_str' to form 'new_str'
3686  *
3687  * returns 'orig_str' if 'old_sub_str' == '' or 'orig_str' == ''
3688  * otherwise returns 'new_str'
3689  */
3690 Datum
3692 {
3693  text *src_text = PG_GETARG_TEXT_PP(0);
3694  text *from_sub_text = PG_GETARG_TEXT_PP(1);
3695  text *to_sub_text = PG_GETARG_TEXT_PP(2);
3696  int src_text_len;
3697  int from_sub_text_len;
3699  text *ret_text;
3700  int start_posn;
3701  int curr_posn;
3702  int chunk_len;
3703  char *start_ptr;
3705 
3706  text_position_setup(src_text, from_sub_text, &state);
3707 
3708  /*
3709  * Note: we check the converted string length, not the original, because
3710  * they could be different if the input contained invalid encoding.
3711  */
3712  src_text_len = state.len1;
3713  from_sub_text_len = state.len2;
3714 
3715  /* Return unmodified source string if empty source or pattern */
3716  if (src_text_len < 1 || from_sub_text_len < 1)
3717  {
3718  text_position_cleanup(&state);
3719  PG_RETURN_TEXT_P(src_text);
3720  }
3721 
3722  start_posn = 1;
3723  curr_posn = text_position_next(1, &state);
3724 
3725  /* When the from_sub_text is not found, there is nothing to do. */
3726  if (curr_posn == 0)
3727  {
3728  text_position_cleanup(&state);
3729  PG_RETURN_TEXT_P(src_text);
3730  }
3731 
3732  /* start_ptr points to the start_posn'th character of src_text */
3733  start_ptr = VARDATA_ANY(src_text);
3734 
3735  initStringInfo(&str);
3736 
3737  do
3738  {
3740 
3741  /* copy the data skipped over by last text_position_next() */
3742  chunk_len = charlen_to_bytelen(start_ptr, curr_posn - start_posn);
3743  appendBinaryStringInfo(&str, start_ptr, chunk_len);
3744 
3745  appendStringInfoText(&str, to_sub_text);
3746 
3747  start_posn = curr_posn;
3748  start_ptr += chunk_len;
3749  start_posn += from_sub_text_len;
3750  start_ptr += charlen_to_bytelen(start_ptr, from_sub_text_len);
3751 
3752  curr_posn = text_position_next(start_posn, &state);
3753  }
3754  while (curr_posn > 0);
3755 
3756  /* copy trailing data */
3757  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
3758  appendBinaryStringInfo(&str, start_ptr, chunk_len);
3759 
3760  text_position_cleanup(&state);
3761 
3762  ret_text = cstring_to_text_with_len(str.data, str.len);
3763  pfree(str.data);
3764 
3765  PG_RETURN_TEXT_P(ret_text);
3766 }
3767 
3768 /*
3769  * check_replace_text_has_escape_char
3770  *
3771  * check whether replace_text contains escape char.
3772  */
3773 static bool
3775 {
3776  const char *p = VARDATA_ANY(replace_text);
3777  const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3778 
3780  {
3781  for (; p < p_end; p++)
3782  {
3783  if (*p == '\\')
3784  return true;
3785  }
3786  }
3787  else
3788  {
3789  for (; p < p_end; p += pg_mblen(p))
3790  {
3791  if (*p == '\\')
3792  return true;
3793  }
3794  }
3795 
3796  return false;
3797 }
3798 
3799 /*
3800  * appendStringInfoRegexpSubstr
3801  *
3802  * Append replace_text to str, substituting regexp back references for
3803  * \n escapes. start_ptr is the start of the match in the source string,
3804  * at logical character position data_pos.
3805  */
3806 static void
3808  regmatch_t *pmatch,
3809  char *start_ptr, int data_pos)
3810 {
3811  const char *p = VARDATA_ANY(replace_text);
3812  const char *p_end = p + VARSIZE_ANY_EXHDR(replace_text);
3813  int eml = pg_database_encoding_max_length();
3814 
3815  for (;;)
3816  {
3817  const char *chunk_start = p;
3818  int so;
3819  int eo;
3820 
3821  /* Find next escape char. */
3822  if (eml == 1)
3823  {
3824  for (; p < p_end && *p != '\\'; p++)
3825  /* nothing */ ;
3826  }
3827  else
3828  {
3829  for (; p < p_end && *p != '\\'; p += pg_mblen(p))
3830  /* nothing */ ;
3831  }
3832 
3833  /* Copy the text we just scanned over, if any. */
3834  if (p > chunk_start)
3835  appendBinaryStringInfo(str, chunk_start, p - chunk_start);
3836 
3837  /* Done if at end of string, else advance over escape char. */
3838  if (p >= p_end)
3839  break;
3840  p++;
3841 
3842  if (p >= p_end)
3843  {
3844  /* Escape at very end of input. Treat same as unexpected char */
3845  appendStringInfoChar(str, '\\');
3846  break;
3847  }
3848 
3849  if (*p >= '1' && *p <= '9')
3850  {
3851  /* Use the back reference of regexp. */
3852  int idx = *p - '0';
3853 
3854  so = pmatch[idx].rm_so;
3855  eo = pmatch[idx].rm_eo;
3856  p++;
3857  }
3858  else if (*p == '&')
3859  {
3860  /* Use the entire matched string. */
3861  so = pmatch[0].rm_so;
3862  eo = pmatch[0].rm_eo;
3863  p++;
3864  }
3865  else if (*p == '\\')
3866  {
3867  /* \\ means transfer one \ to output. */
3868  appendStringInfoChar(str, '\\');
3869  p++;
3870  continue;
3871  }
3872  else
3873  {
3874  /*
3875  * If escape char is not followed by any expected char, just treat
3876  * it as ordinary data to copy. (XXX would it be better to throw
3877  * an error?)
3878  */
3879  appendStringInfoChar(str, '\\');
3880  continue;
3881  }
3882 
3883  if (so != -1 && eo != -1)
3884  {
3885  /*
3886  * Copy the text that is back reference of regexp. Note so and eo
3887  * are counted in characters not bytes.
3888  */
3889  char *chunk_start;
3890  int chunk_len;
3891 
3892  Assert(so >= data_pos);
3893  chunk_start = start_ptr;
3894  chunk_start += charlen_to_bytelen(chunk_start, so - data_pos);
3895  chunk_len = charlen_to_bytelen(chunk_start, eo - so);
3896  appendBinaryStringInfo(str, chunk_start, chunk_len);
3897  }
3898  }
3899 }
3900 
3901 #define REGEXP_REPLACE_BACKREF_CNT 10
3902 
3903 /*
3904  * replace_text_regexp
3905  *
3906  * replace text that matches to regexp in src_text to replace_text.
3907  *
3908  * Note: to avoid having to include regex.h in builtins.h, we declare
3909  * the regexp argument as void *, but really it's regex_t *.
3910  */
3911 text *
3912 replace_text_regexp(text *src_text, void *regexp,
3913  text *replace_text, bool glob)
3914 {
3915  text *ret_text;
3916  regex_t *re = (regex_t *) regexp;
3917  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3920  pg_wchar *data;
3921  size_t data_len;
3922  int search_start;
3923  int data_pos;
3924  char *start_ptr;
3925  bool have_escape;
3926 
3927  initStringInfo(&buf);
3928 
3929  /* Convert data string to wide characters. */
3930  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3931  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3932 
3933  /* Check whether replace_text has escape char. */
3934  have_escape = check_replace_text_has_escape_char(replace_text);
3935 
3936  /* start_ptr points to the data_pos'th character of src_text */
3937  start_ptr = (char *) VARDATA_ANY(src_text);
3938  data_pos = 0;
3939 
3940  search_start = 0;
3941  while (search_start <= data_len)
3942  {
3943  int regexec_result;
3944 
3946 
3947  regexec_result = pg_regexec(re,
3948  data,
3949  data_len,
3950  search_start,
3951  NULL, /* no details */
3953  pmatch,
3954  0);
3955 
3956  if (regexec_result == REG_NOMATCH)
3957  break;
3958 
3959  if (regexec_result != REG_OKAY)
3960  {
3961  char errMsg[100];
3962 
3964  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3965  ereport(ERROR,
3966  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3967  errmsg("regular expression failed: %s", errMsg)));
3968  }
3969 
3970  /*
3971  * Copy the text to the left of the match position. Note we are given
3972  * character not byte indexes.
3973  */
3974  if (pmatch[0].rm_so - data_pos > 0)
3975  {
3976  int chunk_len;
3977 
3978  chunk_len = charlen_to_bytelen(start_ptr,
3979  pmatch[0].rm_so - data_pos);
3980  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3981 
3982  /*
3983  * Advance start_ptr over that text, to avoid multiple rescans of
3984  * it if the replace_text contains multiple back-references.
3985  */
3986  start_ptr += chunk_len;
3987  data_pos = pmatch[0].rm_so;
3988  }
3989 
3990  /*
3991  * Copy the replace_text. Process back references when the
3992  * replace_text has escape characters.
3993  */
3994  if (have_escape)
3995  appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3996  start_ptr, data_pos);
3997  else
3998  appendStringInfoText(&buf, replace_text);
3999 
4000  /* Advance start_ptr and data_pos over the matched text. */
4001  start_ptr += charlen_to_bytelen(start_ptr,
4002  pmatch[0].rm_eo - data_pos);
4003  data_pos = pmatch[0].rm_eo;
4004 
4005  /*
4006  * When global option is off, replace the first instance only.
4007  */
4008  if (!glob)
4009  break;
4010 
4011  /*
4012  * Advance search position. Normally we start the next search at the
4013  * end of the previous match; but if the match was of zero length, we
4014  * have to advance by one character, or we'd just find the same match
4015  * again.
4016  */
4017  search_start = data_pos;
4018  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4019  search_start++;
4020  }
4021 
4022  /*
4023  * Copy the text to the right of the last match.
4024  */
4025  if (data_pos < data_len)
4026  {
4027  int chunk_len;
4028 
4029  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4030  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4031  }
4032 
4033  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4034  pfree(buf.data);
4035  pfree(data);
4036 
4037  return ret_text;
4038 }
4039 
4040 /*
4041  * split_text
4042  * parse input string
4043  * return ord item (1 based)
4044  * based on provided field separator
4045  */
4046 Datum
4048 {
4049  text *inputstring = PG_GETARG_TEXT_PP(0);
4050  text *fldsep = PG_GETARG_TEXT_PP(1);
4051  int fldnum = PG_GETARG_INT32(2);
4052  int inputstring_len;
4053  int fldsep_len;
4055  int start_posn;
4056  int end_posn;
4057  text *result_text;
4058 
4059  /* field number is 1 based */
4060  if (fldnum < 1)
4061  ereport(ERROR,
4062  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4063  errmsg("field position must be greater than zero")));
4064 
4065  text_position_setup(inputstring, fldsep, &state);
4066 
4067  /*
4068  * Note: we check the converted string length, not the original, because
4069  * they could be different if the input contained invalid encoding.
4070  */
4071  inputstring_len = state.len1;
4072  fldsep_len = state.len2;
4073 
4074  /* return empty string for empty input string */
4075  if (inputstring_len < 1)
4076  {
4077  text_position_cleanup(&state);
4079  }
4080 
4081  /* empty field separator */
4082  if (fldsep_len < 1)
4083  {
4084  text_position_cleanup(&state);
4085  /* if first field, return input string, else empty string */
4086  if (fldnum == 1)
4087  PG_RETURN_TEXT_P(inputstring);
4088  else
4090  }
4091 
4092  /* identify bounds of first field */
4093  start_posn = 1;
4094  end_posn = text_position_next(1, &state);
4095 
4096  /* special case if fldsep not found at all */
4097  if (end_posn == 0)
4098  {
4099  text_position_cleanup(&state);
4100  /* if field 1 requested, return input string, else empty string */
4101  if (fldnum == 1)
4102  PG_RETURN_TEXT_P(inputstring);
4103  else
4105  }
4106 
4107  while (end_posn > 0 && --fldnum > 0)
4108  {
4109  /* identify bounds of next field */
4110  start_posn = end_posn + fldsep_len;
4111  end_posn = text_position_next(start_posn, &state);
4112  }
4113 
4114  text_position_cleanup(&state);
4115 
4116  if (fldnum > 0)
4117  {
4118  /* N'th field separator not found */
4119  /* if last field requested, return it, else empty string */
4120  if (fldnum == 1)
4121  result_text = text_substring(PointerGetDatum(inputstring),
4122  start_posn,
4123  -1,
4124  true);
4125  else
4126  result_text = cstring_to_text("");
4127  }
4128  else
4129  {
4130  /* non-last field requested */
4131  result_text = text_substring(PointerGetDatum(inputstring),
4132  start_posn,
4133  end_posn - start_posn,
4134  false);
4135  }
4136 
4137  PG_RETURN_TEXT_P(result_text);
4138 }
4139 
4140 /*
4141  * Convenience function to return true when two text params are equal.
4142  */
4143 static bool
4144 text_isequal(text *txt1, text *txt2)
4145 {
4147  PointerGetDatum(txt1),
4148  PointerGetDatum(txt2)));
4149 }
4150 
4151 /*
4152  * text_to_array
4153  * parse input string and return text array of elements,
4154  * based on provided field separator
4155  */
4156 Datum
4158 {
4159  return text_to_array_internal(fcinfo);
4160 }
4161 
4162 /*
4163  * text_to_array_null
4164  * parse input string and return text array of elements,
4165  * based on provided field separator and null string
4166  *
4167  * This is a separate entry point only to prevent the regression tests from
4168  * complaining about different argument sets for the same internal function.
4169  */
4170 Datum
4172 {
4173  return text_to_array_internal(fcinfo);
4174 }
4175 
4176 /*
4177  * common code for text_to_array and text_to_array_null functions
4178  *
4179  * These are not strict so we have to test for null inputs explicitly.
4180  */
4181 static Datum
4183 {
4184  text *inputstring;
4185  text *fldsep;
4186  text *null_string;
4187  int inputstring_len;
4188  int fldsep_len;
4189  char *start_ptr;
4190  text *result_text;
4191  bool is_null;
4192  ArrayBuildState *astate = NULL;
4193 
4194  /* when input string is NULL, then result is NULL too */
4195  if (PG_ARGISNULL(0))
4196  PG_RETURN_NULL();
4197 
4198  inputstring = PG_GETARG_TEXT_PP(0);
4199 
4200  /* fldsep can be NULL */
4201  if (!PG_ARGISNULL(1))
4202  fldsep = PG_GETARG_TEXT_PP(1);
4203  else
4204  fldsep = NULL;
4205 
4206  /* null_string can be NULL or omitted */
4207  if (PG_NARGS() > 2 && !PG_ARGISNULL(2))
4208  null_string = PG_GETARG_TEXT_PP(2);
4209  else
4210  null_string = NULL;
4211 
4212  if (fldsep != NULL)
4213  {
4214  /*
4215  * Normal case with non-null fldsep. Use the text_position machinery
4216  * to search for occurrences of fldsep.
4217  */
4219  int fldnum;
4220  int start_posn;
4221  int end_posn;
4222  int chunk_len;
4223 
4224  text_position_setup(inputstring, fldsep, &state);
4225 
4226  /*
4227  * Note: we check the converted string length, not the original,
4228  * because they could be different if the input contained invalid
4229  * encoding.
4230  */
4231  inputstring_len = state.len1;
4232  fldsep_len = state.len2;
4233 
4234  /* return empty array for empty input string */
4235  if (inputstring_len < 1)
4236  {
4237  text_position_cleanup(&state);
4239  }
4240 
4241  /*
4242  * empty field separator: return the input string as a one-element
4243  * array
4244  */
4245  if (fldsep_len < 1)
4246  {
4247  Datum elems[1];
4248  bool nulls[1];
4249  int dims[1];
4250  int lbs[1];
4251 
4252  text_position_cleanup(&state);
4253  /* single element can be a NULL too */
4254  is_null = null_string ? text_isequal(inputstring, null_string) : false;
4255 
4256  elems[0] = PointerGetDatum(inputstring);
4257  nulls[0] = is_null;
4258  dims[0] = 1;
4259  lbs[0] = 1;
4260  /* XXX: this hardcodes assumptions about the text type */
4262  1, dims, lbs,
4263  TEXTOID, -1, false, 'i'));
4264  }
4265 
4266  start_posn = 1;
4267  /* start_ptr points to the start_posn'th character of inputstring */
4268  start_ptr = VARDATA_ANY(inputstring);
4269 
4270  for (fldnum = 1;; fldnum++) /* field number is 1 based */
4271  {
4273 
4274  end_posn = text_position_next(start_posn, &state);
4275 
4276  if (end_posn == 0)
4277  {
4278  /* fetch last field */
4279  chunk_len = ((char *) inputstring + VARSIZE_ANY(inputstring)) - start_ptr;
4280  }
4281  else
4282  {
4283  /* fetch non-last field */
4284  chunk_len = charlen_to_bytelen(start_ptr, end_posn - start_posn);
4285  }
4286 
4287  /* must build a temp text datum to pass to accumArrayResult */
4288  result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4289  is_null = null_string ? text_isequal(result_text, null_string) : false;
4290 
4291  /* stash away this field */
4292  astate = accumArrayResult(astate,
4293  PointerGetDatum(result_text),
4294  is_null,
4295  TEXTOID,
4297 
4298  pfree(result_text);
4299 
4300  if (end_posn == 0)
4301  break;
4302 
4303  start_posn = end_posn;
4304  start_ptr += chunk_len;
4305  start_posn += fldsep_len;
4306  start_ptr += charlen_to_bytelen(start_ptr, fldsep_len);
4307  }
4308 
4309  text_position_cleanup(&state);
4310  }
4311  else
4312  {
4313  /*
4314  * When fldsep is NULL, each character in the inputstring becomes an
4315  * element in the result array. The separator is effectively the
4316  * space between characters.
4317  */
4318  inputstring_len = VARSIZE_ANY_EXHDR(inputstring);
4319 
4320  /* return empty array for empty input string */
4321  if (inputstring_len < 1)
4323 
4324  start_ptr = VARDATA_ANY(inputstring);
4325 
4326  while (inputstring_len > 0)
4327  {
4328  int chunk_len = pg_mblen(start_ptr);
4329 
4331 
4332  /* must build a temp text datum to pass to accumArrayResult */
4333  result_text = cstring_to_text_with_len(start_ptr, chunk_len);
4334  is_null = null_string ? text_isequal(result_text, null_string) : false;
4335 
4336  /* stash away this field */
4337  astate = accumArrayResult(astate,
4338  PointerGetDatum(result_text),
4339  is_null,
4340  TEXTOID,
4342 
4343  pfree(result_text);
4344 
4345  start_ptr += chunk_len;
4346  inputstring_len -= chunk_len;
4347  }
4348  }
4349 
4352 }
4353 
4354 /*
4355  * array_to_text
4356  * concatenate Cstring representation of input array elements
4357  * using provided field separator
4358  */
4359 Datum
4361 {
4363  char *fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4364 
4365  PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, NULL));
4366 }
4367 
4368 /*
4369  * array_to_text_null
4370  * concatenate Cstring representation of input array elements
4371  * using provided field separator and null string
4372  *
4373  * This version is not strict so we have to test for null inputs explicitly.
4374  */
4375 Datum
4377 {
4378  ArrayType *v;
4379  char *fldsep;
4380  char *null_string;
4381 
4382  /* returns NULL when first or second parameter is NULL */
4383  if (PG_ARGISNULL(0) || PG_ARGISNULL(1))
4384  PG_RETURN_NULL();
4385 
4386  v = PG_GETARG_ARRAYTYPE_P(0);
4387  fldsep = text_to_cstring(PG_GETARG_TEXT_PP(1));
4388 
4389  /* NULL null string is passed through as a null pointer */
4390  if (!PG_ARGISNULL(2))
4391  null_string = text_to_cstring(PG_GETARG_TEXT_PP(2));
4392  else
4393  null_string = NULL;
4394 
4395  PG_RETURN_TEXT_P(array_to_text_internal(fcinfo, v, fldsep, null_string));
4396 }
4397 
4398 /*
4399  * common code for array_to_text and array_to_text_null functions
4400  */
4401 static text *
4403  const char *fldsep, const char *null_string)
4404 {
4405  text *result;
4406  int nitems,
4407  *dims,
4408  ndims;
4409  Oid element_type;
4410  int typlen;
4411  bool typbyval;
4412  char typalign;
4414  bool printed = false;
4415  char *p;
4416  bits8 *bitmap;
4417  int bitmask;
4418  int i;
4419  ArrayMetaState *my_extra;
4420 
4421  ndims = ARR_NDIM(v);
4422  dims = ARR_DIMS(v);
4423  nitems = ArrayGetNItems(ndims, dims);
4424 
4425  /* if there are no elements, return an empty string */
4426  if (nitems == 0)
4427  return cstring_to_text_with_len("", 0);
4428 
4429  element_type = ARR_ELEMTYPE(v);
4430  initStringInfo(&buf);
4431 
4432  /*
4433  * We arrange to look up info about element type, including its output
4434  * conversion proc, only once per series of calls, assuming the element
4435  * type doesn't change underneath us.
4436  */
4437  my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4438  if (my_extra == NULL)
4439  {
4440  fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4441  sizeof(ArrayMetaState));
4442  my_extra = (ArrayMetaState *) fcinfo->flinfo->fn_extra;
4443  my_extra->element_type = ~element_type;
4444  }
4445 
4446  if (my_extra->element_type != element_type)
4447  {
4448  /*
4449  * Get info about element type, including its output conversion proc
4450  */
4451  get_type_io_data(element_type, IOFunc_output,
4452  &my_extra->typlen, &my_extra->typbyval,
4453  &my_extra->typalign, &my_extra->typdelim,
4454  &my_extra->typioparam, &my_extra->typiofunc);
4455  fmgr_info_cxt(my_extra->typiofunc, &my_extra->proc,
4456  fcinfo->flinfo->fn_mcxt);
4457  my_extra->element_type = element_type;
4458  }
4459  typlen = my_extra->typlen;
4460  typbyval = my_extra->typbyval;
4461  typalign = my_extra->typalign;
4462 
4463  p = ARR_DATA_PTR(v);
4464  bitmap = ARR_NULLBITMAP(v);
4465  bitmask = 1;
4466 
4467  for (i = 0; i < nitems; i++)
4468  {
4469  Datum itemvalue;
4470  char *value;
4471 
4472  /* Get source element, checking for NULL */
4473  if (bitmap && (*bitmap & bitmask) == 0)
4474  {
4475  /* if null_string is NULL, we just ignore null elements */
4476  if (null_string != NULL)
4477  {
4478  if (printed)
4479  appendStringInfo(&buf, "%s%s", fldsep, null_string);
4480  else
4481  appendStringInfoString(&buf, null_string);
4482  printed = true;
4483  }
4484  }
4485  else
4486  {
4487  itemvalue = fetch_att(p, typbyval, typlen);
4488 
4489  value = OutputFunctionCall(&my_extra->proc, itemvalue);
4490 
4491  if (printed)
4492  appendStringInfo(&buf, "%s%s", fldsep, value);
4493  else
4494  appendStringInfoString(&buf, value);
4495  printed = true;
4496 
4497  p = att_addlength_pointer(p, typlen, p);
4498  p = (char *) att_align_nominal(p, typalign);
4499  }
4500 
4501  /* advance bitmap pointer if any */
4502  if (bitmap)
4503  {
4504  bitmask <<= 1;
4505  if (bitmask == 0x100)
4506  {
4507  bitmap++;
4508  bitmask = 1;
4509  }
4510  }
4511  }
4512 
4513  result = cstring_to_text_with_len(buf.data, buf.len);
4514  pfree(buf.data);
4515 
4516  return result;
4517 }
4518 
4519 #define HEXBASE 16
4520 /*
4521  * Convert an int32 to a string containing a base 16 (hex) representation of
4522  * the number.
4523  */
4524 Datum
4526 {
4528  char *ptr;
4529  const char *digits = "0123456789abcdef";
4530  char buf[32]; /* bigger than needed, but reasonable */
4531 
4532  ptr = buf + sizeof(buf) - 1;
4533  *ptr = '\0';
4534 
4535  do
4536  {
4537  *--ptr = digits[value % HEXBASE];
4538  value /= HEXBASE;
4539  } while (ptr > buf && value);
4540 
4542 }
4543 
4544 /*
4545  * Convert an int64 to a string containing a base 16 (hex) representation of
4546  * the number.
4547  */
4548 Datum
4550 {
4551  uint64 value = (uint64) PG_GETARG_INT64(0);
4552  char *ptr;
4553  const char *digits = "0123456789abcdef";
4554  char buf[32]; /* bigger than needed, but reasonable */
4555 
4556  ptr = buf + sizeof(buf) - 1;
4557  *ptr = '\0';
4558 
4559  do
4560  {
4561  *--ptr = digits[value % HEXBASE];
4562  value /= HEXBASE;
4563  } while (ptr > buf && value);
4564 
4566 }
4567 
4568 /*
4569  * Create an md5 hash of a text string and return it as hex
4570  *
4571  * md5 produces a 16 byte (128 bit) hash; double it for hex
4572  */
4573 #define MD5_HASH_LEN 32
4574 
4575 Datum
4577 {
4578  text *in_text = PG_GETARG_TEXT_PP(0);
4579  size_t len;
4580  char hexsum[MD5_HASH_LEN + 1];
4581 
4582  /* Calculate the length of the buffer using varlena metadata */
4583  len = VARSIZE_ANY_EXHDR(in_text);
4584 
4585  /* get the hash result */
4586  if (pg_md5_hash(VARDATA_ANY(in_text), len, hexsum) == false)
4587  ereport(ERROR,
4588  (errcode(ERRCODE_OUT_OF_MEMORY),
4589  errmsg("out of memory")));
4590 
4591  /* convert to text and return it */
4593 }
4594 
4595 /*
4596  * Create an md5 hash of a bytea field and return it as a hex string:
4597  * 16-byte md5 digest is represented in 32 hex characters.
4598  */
4599 Datum
4601 {
4602  bytea *in = PG_GETARG_BYTEA_PP(0);
4603  size_t len;
4604  char hexsum[MD5_HASH_LEN + 1];
4605 
4606  len = VARSIZE_ANY_EXHDR(in);
4607  if (pg_md5_hash(VARDATA_ANY(in), len, hexsum) == false)
4608  ereport(ERROR,
4609  (errcode(ERRCODE_OUT_OF_MEMORY),
4610  errmsg("out of memory")));
4611 
4613 }
4614 
4615 /*
4616  * Return the size of a datum, possibly compressed
4617  *
4618  * Works on any data type
4619  */
4620 Datum
4622 {
4624  int32 result;
4625  int typlen;
4626 
4627  /* On first call, get the input type's typlen, and save at *fn_extra */
4628  if (fcinfo->flinfo->fn_extra == NULL)
4629  {
4630  /* Lookup the datatype of the supplied argument */
4631  Oid argtypeid = get_fn_expr_argtype(fcinfo->flinfo, 0);
4632 
4633  typlen = get_typlen(argtypeid);
4634  if (typlen == 0) /* should not happen */
4635  elog(ERROR, "cache lookup failed for type %u", argtypeid);
4636 
4637  fcinfo->flinfo->fn_extra = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4638  sizeof(int));
4639  *((int *) fcinfo->flinfo->fn_extra) = typlen;
4640  }
4641  else
4642  typlen = *((int *) fcinfo->flinfo->fn_extra);
4643 
4644  if (typlen == -1)
4645  {
4646  /* varlena type, possibly toasted */
4647  result = toast_datum_size(value);
4648  }
4649  else if (typlen == -2)
4650  {
4651  /* cstring */
4652  result = strlen(DatumGetCString(value)) + 1;
4653  }
4654  else
4655  {
4656  /* ordinary fixed-width type */
4657  result = typlen;
4658  }
4659 
4660  PG_RETURN_INT32(result);
4661 }
4662 
4663 /*
4664  * string_agg - Concatenates values and returns string.
4665  *
4666  * Syntax: string_agg(value text, delimiter text) RETURNS text
4667  *
4668  * Note: Any NULL values are ignored. The first-call delimiter isn't
4669  * actually used at all, and on subsequent calls the delimiter precedes
4670  * the associated value.
4671  */
4672 
4673 /* subroutine to initialize state */
4674 static StringInfo
4676 {
4677  StringInfo state;
4678  MemoryContext aggcontext;
4679  MemoryContext oldcontext;
4680 
4681  if (!AggCheckCallContext(fcinfo, &aggcontext))
4682  {
4683  /* cannot be called directly because of internal-type argument */
4684  elog(ERROR, "string_agg_transfn called in non-aggregate context");
4685  }
4686 
4687  /*
4688  * Create state in aggregate context. It'll stay there across subsequent
4689  * calls.
4690  */
4691  oldcontext = MemoryContextSwitchTo(aggcontext);
4692  state = makeStringInfo();
4693  MemoryContextSwitchTo(oldcontext);
4694 
4695  return state;
4696 }
4697 
4698 Datum
4700 {
4701  StringInfo state;
4702 
4703  state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4704 
4705  /* Append the value unless null. */
4706  if (!PG_ARGISNULL(1))
4707  {
4708  /* On the first time through, we ignore the delimiter. */
4709  if (state == NULL)
4710  state = makeStringAggState(fcinfo);
4711  else if (!PG_ARGISNULL(2))
4712  appendStringInfoText(state, PG_GETARG_TEXT_PP(2)); /* delimiter */
4713 
4714  appendStringInfoText(state, PG_GETARG_TEXT_PP(1)); /* value */
4715  }
4716 
4717  /*
4718  * The transition type for string_agg() is declared to be "internal",
4719  * which is a pass-by-value type the same size as a pointer.
4720  */
4721  PG_RETURN_POINTER(state);
4722 }
4723 
4724 Datum
4726 {
4727  StringInfo state;
4728 
4729  /* cannot be called directly because of internal-type argument */
4730  Assert(AggCheckCallContext(fcinfo, NULL));
4731 
4732  state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
4733 
4734  if (state != NULL)
4736  else
4737  PG_RETURN_NULL();
4738 }
4739 
4740 /*
4741  * Prepare cache with fmgr info for the output functions of the datatypes of
4742  * the arguments of a concat-like function, beginning with argument "argidx".
4743  * (Arguments before that will have corresponding slots in the resulting
4744  * FmgrInfo array, but we don't fill those slots.)
4745  */
4746 static FmgrInfo *
4748 {
4749  FmgrInfo *foutcache;
4750  int i;
4751 
4752  /* We keep the info in fn_mcxt so it survives across calls */
4753  foutcache = (FmgrInfo *) MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
4754  PG_NARGS() * sizeof(FmgrInfo));
4755 
4756  for (i = argidx; i < PG_NARGS(); i++)
4757  {
4758  Oid valtype;
4759  Oid typOutput;
4760  bool typIsVarlena;
4761 
4762  valtype = get_fn_expr_argtype(fcinfo->flinfo, i);
4763  if (!OidIsValid(valtype))
4764  elog(ERROR, "could not determine data type of concat() input");
4765 
4766  getTypeOutputInfo(valtype, &typOutput, &typIsVarlena);
4767  fmgr_info_cxt(typOutput, &foutcache[i], fcinfo->flinfo->fn_mcxt);
4768  }
4769 
4770  fcinfo->flinfo->fn_extra = foutcache;
4771 
4772  return foutcache;
4773 }
4774 
4775 /*
4776  * Implementation of both concat() and concat_ws().
4777  *
4778  * sepstr is the separator string to place between values.
4779  * argidx identifies the first argument to concatenate (counting from zero);
4780  * note that this must be constant across any one series of calls.
4781  *
4782  * Returns NULL if result should be NULL, else text value.
4783  */
4784 static text *
4785 concat_internal(const char *sepstr, int argidx,
4786  FunctionCallInfo fcinfo)
4787 {
4788  text *result;
4790  FmgrInfo *foutcache;
4791  bool first_arg = true;
4792  int i;
4793 
4794  /*
4795  * concat(VARIADIC some-array) is essentially equivalent to
4796  * array_to_text(), ie concat the array elements with the given separator.
4797  * So we just pass the case off to that code.
4798  */
4799  if (get_fn_expr_variadic(fcinfo->flinfo))
4800  {
4801  ArrayType *arr;
4802 
4803  /* Should have just the one argument */
4804  Assert(argidx == PG_NARGS() - 1);
4805 
4806  /* concat(VARIADIC NULL) is defined as NULL */
4807  if (PG_ARGISNULL(argidx))
4808  return NULL;
4809 
4810  /*
4811  * Non-null argument had better be an array. We assume that any call
4812  * context that could let get_fn_expr_variadic return true will have
4813  * checked that a VARIADIC-labeled parameter actually is an array. So
4814  * it should be okay to just Assert that it's an array rather than
4815  * doing a full-fledged error check.
4816  */
4818 
4819  /* OK, safe to fetch the array value */
4820  arr = PG_GETARG_ARRAYTYPE_P(argidx);
4821 
4822  /*
4823  * And serialize the array. We tell array_to_text to ignore null
4824  * elements, which matches the behavior of the loop below.
4825  */
4826  return array_to_text_internal(fcinfo, arr, sepstr, NULL);
4827  }
4828 
4829  /* Normal case without explicit VARIADIC marker */
4830  initStringInfo(&str);
4831 
4832  /* Get output function info, building it if first time through */
4833  foutcache = (FmgrInfo *) fcinfo->flinfo->fn_extra;
4834  if (foutcache == NULL)
4835  foutcache = build_concat_foutcache(fcinfo, argidx);
4836 
4837  for (i = argidx; i < PG_NARGS(); i++)
4838  {
4839  if (!PG_ARGISNULL(i))
4840  {
4842 
4843  /* add separator if appropriate */
4844  if (first_arg)
4845  first_arg = false;
4846  else
4847  appendStringInfoString(&str, sepstr);
4848 
4849  /* call the appropriate type output function, append the result */
4851  OutputFunctionCall(&foutcache[i], value));
4852  }
4853  }
4854 
4855  result = cstring_to_text_with_len(str.data, str.len);
4856  pfree(str.data);
4857 
4858  return result;
4859 }
4860 
4861 /*
4862  * Concatenate all arguments. NULL arguments are ignored.
4863  */
4864 Datum
4866 {
4867  text *result;
4868 
4869  result = concat_internal("", 0, fcinfo);
4870  if (result == NULL)
4871  PG_RETURN_NULL();
4872  PG_RETURN_TEXT_P(result);
4873 }
4874 
4875 /*
4876  * Concatenate all but first argument value with separators. The first
4877  * parameter is used as the separator. NULL arguments are ignored.
4878  */
4879 Datum
4881 {
4882  char *sep;
4883  text *result;
4884 
4885  /* return NULL when separator is NULL */
4886  if (PG_ARGISNULL(0))
4887  PG_RETURN_NULL();
4889 
4890  result = concat_internal(sep, 1, fcinfo);
4891  if (result == NULL)
4892  PG_RETURN_NULL();
4893  PG_RETURN_TEXT_P(result);
4894 }
4895 
4896 /*
4897  * Return first n characters in the string. When n is negative,
4898  * return all but last |n| characters.
4899  */
4900 Datum
4902 {
4903  text *str = PG_GETARG_TEXT_PP(0);
4904  const char *p = VARDATA_ANY(str);
4905  int len = VARSIZE_ANY_EXHDR(str);
4906  int n = PG_GETARG_INT32(1);
4907  int rlen;
4908 
4909  if (n < 0)
4910  n = pg_mbstrlen_with_len(p, len) + n;
4911  rlen = pg_mbcharcliplen(p, len, n);
4912 
4914 }
4915 
4916 /*
4917  * Return last n characters in the string. When n is negative,
4918  * return all but first |n| characters.
4919  */
4920 Datum
4922 {
4923  text *str = PG_GETARG_TEXT_PP(0);
4924  const char *p = VARDATA_ANY(str);
4925  int len = VARSIZE_ANY_EXHDR(str);
4926  int n = PG_GETARG_INT32(1);
4927  int off;
4928 
4929  if (n < 0)
4930  n = -n;
4931  else
4932  n = pg_mbstrlen_with_len(p, len) - n;
4933  off = pg_mbcharcliplen(p, len, n);
4934 
4935  PG_RETURN_TEXT_P(cstring_to_text_with_len(p + off, len - off));
4936 }
4937 
4938 /*
4939  * Return reversed string
4940  */
4941 Datum
4943 {
4944  text *str = PG_GETARG_TEXT_PP(0);
4945  const char *p = VARDATA_ANY(str);
4946  int len = VARSIZE_ANY_EXHDR(str);
4947  const char *endp = p + len;
4948  text *result;
4949  char *dst;
4950 
4951  result = palloc(len + VARHDRSZ);
4952  dst = (char *) VARDATA(result) + len;
4953  SET_VARSIZE(result, len + VARHDRSZ);
4954 
4956  {
4957  /* multibyte version */
4958  while (p < endp)
4959  {
4960  int sz;
4961 
4962  sz = pg_mblen(p);
4963  dst -= sz;
4964  memcpy(dst, p, sz);
4965  p += sz;
4966  }
4967  }
4968  else
4969  {
4970  /* single byte version */
4971  while (p < endp)
4972  *(--dst) = *p++;
4973  }
4974 
4975  PG_RETURN_TEXT_P(result);
4976 }
4977 
4978 
4979 /*
4980  * Support macros for text_format()
4981  */
4982 #define TEXT_FORMAT_FLAG_MINUS 0x0001 /* is minus flag present? */
4983 
4984 #define ADVANCE_PARSE_POINTER(ptr,end_ptr) \
4985  do { \
4986  if (++(ptr) >= (end_ptr)) \
4987  ereport(ERROR, \
4988  (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
4989  errmsg("unterminated format() type specifier"), \
4990  errhint("For a single \"%%\" use \"%%%%\"."))); \
4991  } while (0)
4992 
4993 /*
4994  * Returns a formatted string
4995  */
4996 Datum
4998 {
4999  text *fmt;
5001  const char *cp;
5002  const char *start_ptr;
5003  const char *end_ptr;
5004  text *result;
5005  int arg;
5006  bool funcvariadic;
5007  int nargs;
5008  Datum *elements = NULL;
5009  bool *nulls = NULL;
5010  Oid element_type = InvalidOid;
5011  Oid prev_type = InvalidOid;
5012  Oid prev_width_type = InvalidOid;
5013  FmgrInfo typoutputfinfo;
5014  FmgrInfo typoutputinfo_width;
5015 
5016  /* When format string is null, immediately return null */
5017  if (PG_ARGISNULL(0))
5018  PG_RETURN_NULL();
5019 
5020  /* If argument is marked VARIADIC, expand array into elements */
5021  if (get_fn_expr_variadic(fcinfo->flinfo))
5022  {
5023  ArrayType *arr;
5024  int16 elmlen;
5025  bool elmbyval;
5026  char elmalign;
5027  int nitems;
5028 
5029  /* Should have just the one argument */
5030  Assert(PG_NARGS() == 2);
5031 
5032  /* If argument is NULL, we treat it as zero-length array */
5033  if (PG_ARGISNULL(1))
5034  nitems = 0;
5035  else
5036  {
5037  /*
5038  * Non-null argument had better be an array. We assume that any
5039  * call context that could let get_fn_expr_variadic return true
5040  * will have checked that a VARIADIC-labeled parameter actually is
5041  * an array. So it should be okay to just Assert that it's an
5042  * array rather than doing a full-fledged error check.
5043  */
5045 
5046  /* OK, safe to fetch the array value */
5047  arr = PG_GETARG_ARRAYTYPE_P(1);
5048 
5049  /* Get info about array element type */
5050  element_type = ARR_ELEMTYPE(arr);
5051  get_typlenbyvalalign(element_type,
5052  &elmlen, &elmbyval, &elmalign);
5053 
5054  /* Extract all array elements */
5055  deconstruct_array(arr, element_type, elmlen, elmbyval, elmalign,
5056  &elements, &nulls, &nitems);
5057  }
5058 
5059  nargs = nitems + 1;
5060  funcvariadic = true;
5061  }
5062  else
5063  {
5064  /* Non-variadic case, we'll process the arguments individually */
5065  nargs = PG_NARGS();
5066  funcvariadic = false;
5067  }
5068 
5069  /* Setup for main loop. */
5070  fmt = PG_GETARG_TEXT_PP(0);
5071  start_ptr = VARDATA_ANY(fmt);
5072  end_ptr = start_ptr + VARSIZE_ANY_EXHDR(fmt);
5073  initStringInfo(&str);
5074  arg = 1; /* next argument position to print */
5075 
5076  /* Scan format string, looking for conversion specifiers. */
5077  for (cp = start_ptr; cp < end_ptr; cp++)
5078  {
5079  int argpos;
5080  int widthpos;
5081  int flags;
5082  int width;
5083  Datum value;
5084  bool isNull;
5085  Oid typid;
5086 
5087  /*
5088  * If it's not the start of a conversion specifier, just copy it to
5089  * the output buffer.
5090  */
5091  if (*cp != '%')
5092  {
5093  appendStringInfoCharMacro(&str, *cp);
5094  continue;
5095  }
5096 
5097  ADVANCE_PARSE_POINTER(cp, end_ptr);
5098 
5099  /* Easy case: %% outputs a single % */
5100  if (*cp == '%')
5101  {
5102  appendStringInfoCharMacro(&str, *cp);
5103  continue;
5104  }
5105 
5106  /* Parse the optional portions of the format specifier */
5107  cp = text_format_parse_format(cp, end_ptr,
5108  &argpos, &widthpos,
5109  &flags, &width);
5110 
5111  /*
5112  * Next we should see the main conversion specifier. Whether or not
5113  * an argument position was present, it's known that at least one
5114  * character remains in the string at this point. Experience suggests
5115  * that it's worth checking that that character is one of the expected
5116  * ones before we try to fetch arguments, so as to produce the least
5117  * confusing response to a mis-formatted specifier.
5118  */
5119  if (strchr("sIL", *cp) == NULL)
5120  ereport(ERROR,
5121  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5122  errmsg("unrecognized format() type specifier \"%c\"",
5123  *cp),
5124  errhint("For a single \"%%\" use \"%%%%\".")));
5125 
5126  /* If indirect width was specified, get its value */
5127  if (widthpos >= 0)
5128  {
5129  /* Collect the specified or next argument position */
5130  if (widthpos > 0)
5131  arg = widthpos;
5132  if (arg >= nargs)
5133  ereport(ERROR,
5134  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5135  errmsg("too few arguments for format()")));
5136 
5137  /* Get the value and type of the selected argument */
5138  if (!funcvariadic)
5139  {
5140  value = PG_GETARG_DATUM(arg);
5141  isNull = PG_ARGISNULL(arg);
5142  typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5143  }
5144  else
5145  {
5146  value = elements[arg - 1];
5147  isNull = nulls[arg - 1];
5148  typid = element_type;
5149  }
5150  if (!OidIsValid(typid))
5151  elog(ERROR, "could not determine data type of format() input");
5152 
5153  arg++;
5154 
5155  /* We can treat NULL width the same as zero */
5156  if (isNull)
5157  width = 0;
5158  else if (typid == INT4OID)
5159  width = DatumGetInt32(value);
5160  else if (typid == INT2OID)
5161  width = DatumGetInt16(value);
5162  else
5163  {
5164  /* For less-usual datatypes, convert to text then to int */
5165  char *str;
5166 
5167  if (typid != prev_width_type)
5168  {
5169  Oid typoutputfunc;
5170  bool typIsVarlena;
5171 
5172  getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5173  fmgr_info(typoutputfunc, &typoutputinfo_width);
5174  prev_width_type = typid;
5175  }
5176 
5177  str = OutputFunctionCall(&typoutputinfo_width, value);
5178 
5179  /* pg_atoi will complain about bad data or overflow */
5180  width = pg_atoi(str, sizeof(int), '\0');
5181 
5182  pfree(str);
5183  }
5184  }
5185 
5186  /* Collect the specified or next argument position */
5187  if (argpos > 0)
5188  arg = argpos;
5189  if (arg >= nargs)
5190  ereport(ERROR,
5191  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5192  errmsg("too few arguments for format()")));
5193 
5194  /* Get the value and type of the selected argument */
5195  if (!funcvariadic)
5196  {
5197  value = PG_GETARG_DATUM(arg);
5198  isNull = PG_ARGISNULL(arg);
5199  typid = get_fn_expr_argtype(fcinfo->flinfo, arg);
5200  }
5201  else
5202  {
5203  value = elements[arg - 1];
5204  isNull = nulls[arg - 1];
5205  typid = element_type;
5206  }
5207  if (!OidIsValid(typid))
5208  elog(ERROR, "could not determine data type of format() input");
5209 
5210  arg++;
5211 
5212  /*
5213  * Get the appropriate typOutput function, reusing previous one if
5214  * same type as previous argument. That's particularly useful in the
5215  * variadic-array case, but often saves work even for ordinary calls.
5216  */
5217  if (typid != prev_type)
5218  {
5219  Oid typoutputfunc;
5220  bool typIsVarlena;
5221 
5222  getTypeOutputInfo(typid, &typoutputfunc, &typIsVarlena);
5223  fmgr_info(typoutputfunc, &typoutputfinfo);
5224  prev_type = typid;
5225  }
5226 
5227  /*
5228  * And now we can format the value.
5229  */
5230  switch (*cp)
5231  {
5232  case 's':
5233  case 'I':
5234  case 'L':
5235  text_format_string_conversion(&str, *cp, &typoutputfinfo,
5236  value, isNull,
5237  flags, width);
5238  break;
5239  default:
5240  /* should not get here, because of previous check */
5241  ereport(ERROR,
5242  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5243  errmsg("unrecognized format() type specifier \"%c\"",
5244  *cp),
5245  errhint("For a single \"%%\" use \"%%%%\".")));
5246  break;
5247  }
5248  }
5249 
5250  /* Don't need deconstruct_array results anymore. */
5251  if (elements != NULL)
5252  pfree(elements);
5253  if (nulls != NULL)
5254  pfree(nulls);
5255 
5256  /* Generate results. */
5257  result = cstring_to_text_with_len(str.data, str.len);
5258  pfree(str.data);
5259 
5260  PG_RETURN_TEXT_P(result);
5261 }
5262 
5263 /*
5264  * Parse contiguous digits as a decimal number.
5265  *
5266  * Returns true if some digits could be parsed.
5267  * The value is returned into *value, and *ptr is advanced to the next
5268  * character to be parsed.
5269  *
5270  * Note parsing invariant: at least one character is known available before
5271  * string end (end_ptr) at entry, and this is still true at exit.
5272  */
5273 static bool
5274 text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
5275 {
5276  bool found = false;
5277  const char *cp = *ptr;
5278  int val = 0;
5279 
5280  while (*cp >= '0' && *cp <= '9')
5281  {
5282  int newval = val * 10 + (*cp - '0');
5283 
5284  if (newval / 10 != val) /* overflow? */
5285  ereport(ERROR,
5286  (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5287  errmsg("number is out of range")));
5288  val = newval;
5289  ADVANCE_PARSE_POINTER(cp, end_ptr);
5290  found = true;
5291  }
5292 
5293  *ptr = cp;
5294  *value = val;
5295 
5296  return found;
5297 }
5298 
5299 /*
5300  * Parse a format specifier (generally following the SUS printf spec).
5301  *
5302  * We have already advanced over the initial '%', and we are looking for
5303  * [argpos][flags][width]type (but the type character is not consumed here).
5304  *
5305  * Inputs are start_ptr (the position after '%') and end_ptr (string end + 1).
5306  * Output parameters:
5307  * argpos: argument position for value to be printed. -1 means unspecified.
5308  * widthpos: argument position for width. Zero means the argument position
5309  * was unspecified (ie, take the next arg) and -1 means no width
5310  * argument (width was omitted or specified as a constant).
5311  * flags: bitmask of flags.
5312  * width: directly-specified width value. Zero means the width was omitted
5313  * (note it's not necessary to distinguish this case from an explicit
5314  * zero width value).
5315  *
5316  * The function result is the next character position to be parsed, ie, the
5317  * location where the type character is/should be.
5318  *
5319  * Note parsing invariant: at least one character is known available before
5320  * string end (end_ptr) at entry, and this is still true at exit.
5321  */
5322 static const char *
5323 text_format_parse_format(const char *start_ptr, const char *end_ptr,
5324  int *argpos, int *widthpos,
5325  int *flags, int *width)
5326 {
5327  const char *cp = start_ptr;
5328  int n;
5329 
5330  /* set defaults for output parameters */
5331  *argpos = -1;
5332  *widthpos = -1;
5333  *flags = 0;
5334  *width = 0;
5335 
5336  /* try to identify first number */
5337  if (text_format_parse_digits(&cp, end_ptr, &n))
5338  {
5339  if (*cp != '$')
5340  {
5341  /* Must be just a width and a type, so we're done */
5342  *width = n;
5343  return cp;
5344  }
5345  /* The number was argument position */
5346  *argpos = n;
5347  /* Explicit 0 for argument index is immediately refused */
5348  if (n == 0)
5349  ereport(ERROR,
5350  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5351  errmsg("format specifies argument 0, but arguments are numbered from 1")));
5352  ADVANCE_PARSE_POINTER(cp, end_ptr);
5353  }
5354 
5355  /* Handle flags (only minus is supported now) */
5356  while (*cp == '-')
5357  {
5358  *flags |= TEXT_FORMAT_FLAG_MINUS;
5359  ADVANCE_PARSE_POINTER(cp, end_ptr);
5360  }
5361 
5362  if (*cp == '*')
5363  {
5364  /* Handle indirect width */
5365  ADVANCE_PARSE_POINTER(cp, end_ptr);
5366  if (text_format_parse_digits(&cp, end_ptr, &n))
5367  {
5368  /* number in this position must be closed by $ */
5369  if (*cp != '$')
5370  ereport(ERROR,
5371  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5372  errmsg("width argument position must be ended by \"$\"")));
5373  /* The number was width argument position */
5374  *widthpos = n;
5375  /* Explicit 0 for argument index is immediately refused */
5376  if (n == 0)
5377  ereport(ERROR,
5378  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5379  errmsg("format specifies argument 0, but arguments are numbered from 1")));
5380  ADVANCE_PARSE_POINTER(cp, end_ptr);
5381  }
5382  else
5383  *widthpos = 0; /* width's argument position is unspecified */
5384  }
5385  else
5386  {
5387  /* Check for direct width specification */
5388  if (text_format_parse_digits(&cp, end_ptr, &n))
5389  *width = n;
5390  }
5391 
5392  /* cp should now be pointing at type character */
5393  return cp;
5394 }
5395 
5396 /*
5397  * Format a %s, %I, or %L conversion
5398  */
5399 static void
5401  FmgrInfo *typOutputInfo,
5402  Datum value, bool isNull,
5403  int flags, int width)
5404 {
5405  char *str;
5406 
5407  /* Handle NULL arguments before trying to stringify the value. */
5408  if (isNull)
5409  {
5410  if (conversion == 's')
5411  text_format_append_string(buf, "", flags, width);
5412  else if (conversion == 'L')
5413  text_format_append_string(buf, "NULL", flags, width);
5414  else if (conversion == 'I')
5415  ereport(ERROR,
5416  (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
5417  errmsg("null values cannot be formatted as an SQL identifier")));
5418  return;
5419  }
5420 
5421  /* Stringify. */
5422  str = OutputFunctionCall(typOutputInfo, value);
5423 
5424  /* Escape. */
5425  if (conversion == 'I')
5426  {
5427  /* quote_identifier may or may not allocate a new string. */
5428  text_format_append_string(buf, quote_identifier(str), flags, width);
5429  }
5430  else if (conversion == 'L')
5431  {
5432  char *qstr = quote_literal_cstr(str);
5433 
5434  text_format_append_string(buf, qstr, flags, width);
5435  /* quote_literal_cstr() always allocates a new string */
5436  pfree(qstr);
5437  }
5438  else
5439  text_format_append_string(buf, str, flags, width);
5440 
5441  /* Cleanup. */
5442  pfree(str);
5443 }
5444 
5445 /*
5446  * Append str to buf, padding as directed by flags/width
5447  */
5448 static void
5450  int flags, int width)
5451 {
5452  bool align_to_left = false;
5453  int len;
5454 
5455  /* fast path for typical easy case */
5456  if (width == 0)
5457  {
5458  appendStringInfoString(buf, str);
5459  return;
5460  }
5461 
5462  if (width < 0)
5463  {
5464  /* Negative width: implicit '-' flag, then take absolute value */
5465  align_to_left = true;
5466  /* -INT_MIN is undefined */
5467  if (width <= INT_MIN)
5468  ereport(ERROR,
5469  (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
5470  errmsg("number is out of range")));
5471  width = -width;
5472  }
5473  else if (flags & TEXT_FORMAT_FLAG_MINUS)
5474  align_to_left = true;
5475 
5476  len = pg_mbstrlen(str);
5477  if (align_to_left)
5478  {
5479  /* left justify */
5480  appendStringInfoString(buf, str);
5481  if (len < width)
5482  appendStringInfoSpaces(buf, width - len);
5483  }
5484  else
5485  {
5486  /* right justify */
5487  if (len < width)
5488  appendStringInfoSpaces(buf, width - len);
5489  appendStringInfoString(buf, str);
5490  }
5491 }
5492 
5493 /*
5494  * text_format_nv - nonvariadic wrapper for text_format function.
5495  *
5496  * note: this wrapper is necessary to pass the sanity check in opr_sanity,
5497  * which checks that all built-in functions that share the implementing C
5498  * function take the same number of arguments.
5499  */
5500 Datum
5502 {
5503  return text_format(fcinfo);
5504 }
5505 
5506 /*
5507  * Helper function for Levenshtein distance functions. Faster than memcmp(),
5508  * for this use case.
5509  */
5510 static inline bool
5511 rest_of_char_same(const char *s1, const char *s2, int len)
5512 {
5513  while (len > 0)
5514  {
5515  len--;
5516  if (s1[len] != s2[len])
5517  return false;
5518  }
5519  return true;
5520 }
5521 
5522 /* Expand each Levenshtein distance variant */
5523 #include "levenshtein.c"
5524 #define LEVENSHTEIN_LESS_EQUAL
5525 #include "levenshtein.c"
Datum bttext_pattern_cmp(PG_FUNCTION_ARGS)
Definition: varlena.c:2710
#define PG_CACHE_LINE_SIZE
Datum text_to_array(PG_FUNCTION_ARGS)
Definition: varlena.c:4157
Datum bytea_substr_no_len(PG_FUNCTION_ARGS)
Definition: varlena.c:2846
struct SortSupportData * SortSupport
Definition: sortsupport.h:58
static int varstrfastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2034
Value * makeString(char *str)
Definition: value.c:53
#define COLLPROVIDER_ICU
Definition: pg_collation.h:85
signed short int16
Definition: c.h:283
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:4997
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2001
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:4984
Definition: fmgr.h:56
text * replace_text_regexp(text *src_text, void *regexp, text *replace_text, bool glob)
Definition: varlena.c:3912
#define VARATT_IS_COMPRESSED(PTR)
Definition: postgres.h:313
Datum byteaSetBit(PG_FUNCTION_ARGS)
Definition: varlena.c:3116
int pg_mbcharcliplen(const char *mbstr, int len, int limit)
Definition: mbutils.c:862
Datum split_text(PG_FUNCTION_ARGS)
Definition: varlena.c:4047
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:2646
#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:4573
const char * quote_identifier(const char *ident)
Definition: ruleutils.c:10406
Datum text_lt(PG_FUNCTION_ARGS)
Definition: varlena.c:1706
Datum text_pattern_le(PG_FUNCTION_ARGS)
Definition: varlena.c:2662
#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:2694
#define HEXBASE
Definition: varlena.c:4519
#define TEXTOID
Definition: pg_type.h:324
#define VARSIZE(PTR)
Definition: postgres.h:304
Datum replace_text(PG_FUNCTION_ARGS)
Definition: varlena.c:3691
Datum byteagt(PG_FUNCTION_ARGS)
Definition: varlena.c:3593
static void text_format_string_conversion(StringInfo buf, char conversion, FmgrInfo *typOutputInfo, Datum value, bool isNull, int flags, int width)
Definition: varlena.c:5400
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:328
#define VARHDRSZ
Definition: c.h:493
Datum md5_bytea(PG_FUNCTION_ARGS)
Definition: varlena.c:4600
char * pstrdup(const char *in)
Definition: mcxt.c:1076
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:28
StringInfoData * StringInfo
Definition: stringinfo.h:43
#define Min(x, y)
Definition: c.h:802
int ArrayGetNItems(int ndim, const int *dims)
Definition: arrayutils.c:75
union pg_locale_struct::@120 info
#define PG_GETARG_BYTEA_P_COPY(n)
Definition: fmgr.h:278
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2237
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2212
#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:2780
#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:2038
int errcode(int sqlerrcode)
Definition: elog.c:575
void pq_sendtext(StringInfo buf, const char *str, int slen)
Definition: pqformat.c:174
Datum idx(PG_FUNCTION_ARGS)
Definition: _int_op.c:264
#define DatumGetByteaPP(X)
Definition: fmgr.h:255
#define PG_GETARG_POINTER(n)
Definition: fmgr.h:241
Datum byteaSetByte(PG_FUNCTION_ARGS)
Definition: varlena.c:3084
pg_wchar * wstr2
Definition: varlena.c:52
Datum bytea_string_agg_finalfn(PG_FUNCTION_ARGS)
Definition: varlena.c:482
Datum string_agg_transfn(PG_FUNCTION_ARGS)
Definition: varlena.c:4699
Datum md5_text(PG_FUNCTION_ARGS)
Definition: varlena.c:4576
#define PG_RETURN_BYTEA_P(x)
Definition: fmgr.h:330
ArrayType * construct_empty_array(Oid elmtype)
Definition: arrayfuncs.c:3398
static bytea * bytea_substring(Datum str, int S, int L, bool length_not_specified)
Definition: varlena.c:2855
#define LOG
Definition: elog.h:26
unsigned int Oid
Definition: postgres_ext.h:31
Datum byteaeq(PG_FUNCTION_ARGS)
Definition: varlena.c:3489
Datum textlen(PG_FUNCTION_ARGS)
Definition: varlena.c:622
#define OidIsValid(objectId)
Definition: c.h:576
Datum bttextsortsupport(PG_FUNCTION_ARGS)
Definition: varlena.c:1781
bytea * pq_endtypsend(StringInfo buf)
Definition: pqformat.c:348
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:118
#define PG_GET_COLLATION()
Definition: fmgr.h:163
Datum byteaoverlay_no_len(PG_FUNCTION_ARGS)
Definition: varlena.c:2925
Datum text_concat(PG_FUNCTION_ARGS)
Definition: varlena.c:4865
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:5449
Datum array_to_text_null(PG_FUNCTION_ARGS)
Definition: varlena.c:4376
Datum text_concat_ws(PG_FUNCTION_ARGS)
Definition: varlena.c:4880
regoff_t rm_eo
Definition: regex.h:86
signed int int32
Definition: c.h:284
#define PG_STR_GET_BYTEA(str_)
Definition: varlena.c:2813
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:1964
#define PG_GETARG_TEXT_PP(n)
Definition: fmgr.h:273
char * OutputFunctionCall(FmgrInfo *flinfo, Datum val)
Definition: fmgr.c:1662
static int32 text_length(Datum str)
Definition: varlena.c:640
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:794
bool typbyval
Definition: array.h:225
#define NAMEDATALEN
void truncate_identifier(char *ident, int len, bool warn)
Definition: scansup.c:187
Datum to_hex64(PG_FUNCTION_ARGS)
Definition: varlena.c:4549
#define VARATT_IS_EXTERNAL(PTR)
Definition: postgres.h:314
static Datum text_to_array_internal(PG_FUNCTION_ARGS)
Definition: varlena.c:4182
Datum bytealt(PG_FUNCTION_ARGS)
Definition: varlena.c:3553
bool SplitDirectoriesString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3391
FmgrInfo * flinfo
Definition: fmgr.h:79
#define PG_GETARG_ARRAYTYPE_P(n)
Definition: array.h:248
#define appendStringInfoCharMacro(str, ch)
Definition: stringinfo.h:127
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:4360
void pfree(void *pointer)
Definition: mcxt.c:949
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:78
Datum string_agg_finalfn(PG_FUNCTION_ARGS)
Definition: varlena.c:4725
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:3774
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
int varstr_cmp(const char *arg1, int len1, const char *arg2, int len2, Oid collid)
Definition: varlena.c:1382
Oid get_fn_expr_argtype(FmgrInfo *flinfo, int argnum)
Definition: fmgr.c:1904
Datum byteage(PG_FUNCTION_ARGS)
Definition: varlena.c:3613
#define ARR_DIMS(a)
Definition: array.h:279
void fmgr_info(Oid functionId, FmgrInfo *finfo)
Definition: fmgr.c:122
MemoryContext ssup_cxt
Definition: sortsupport.h:66
struct varlena * pg_detoast_datum_packed(struct varlena *datum)
Definition: fmgr.c:1858
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:4171
#define MAXPGPATH
static struct @121 value
int pg_mbcliplen(const char *mbstr, int len, int limit)
Definition: mbutils.c:820
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:4402
Definition: c.h:541
static void appendStringInfoText(StringInfo str, const text *t)
Definition: varlena.c:3677
Datum text_larger(PG_FUNCTION_ARGS)
Definition: varlena.c:2591
#define INT2OID
Definition: pg_type.h:308
Datum texteq(PG_FUNCTION_ARGS)
Definition: varlena.c:1642
int(* comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:107
void appendStringInfoString(StringInfo str, const char *s)
Definition: stringinfo.c:157
#define ARR_DATA_PTR(a)
Definition: array.h:307
hyperLogLogState abbr_card
Definition: varlena.c:73
Datum text_smaller(PG_FUNCTION_ARGS)
Definition: varlena.c:2603
Datum textne(PG_FUNCTION_ARGS)
Definition: varlena.c:1677
int16 typlen
Definition: array.h:224
pg_locale_t locale
Definition: varlena.c:76
int(* abbrev_full_comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:192
static char * buf
Definition: pg_test_fsync.c:67
#define DatumBigEndianToNative(x)
Definition: pg_bswap.h:149
#define memmove(d, s, c)
Definition: c.h:1045
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:161
#define strcoll_l
Definition: win32_port.h:397
bool SplitIdentifierString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3264
#define DEFAULT_COLLATION_OID
Definition: pg_collation.h:75
char typdelim
Definition: array.h:227
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:2937
Datum text_name(PG_FUNCTION_ARGS)
Definition: varlena.c:3166
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
Datum(* abbrev_converter)(Datum original, SortSupport ssup)
Definition: sortsupport.h:173
unsigned int uint32
Definition: c.h:296
void * ssup_extra
Definition: sortsupport.h:87
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:132
static int text_cmp(text *arg1, text *arg2, Oid collid)
Definition: varlena.c:1617
Datum byteaGetByte(PG_FUNCTION_ARGS)
Definition: varlena.c:3018
#define S(n, x)
Definition: sha1.c:55
#define PG_RETURN_ARRAYTYPE_P(x)
Definition: array.h:250
Datum pg_column_size(PG_FUNCTION_ARGS)
Definition: varlena.c:4621
Datum text_gt(PG_FUNCTION_ARGS)
Definition: varlena.c:1736
#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:2624
Datum makeArrayResult(ArrayBuildState *astate, MemoryContext rcontext)
Definition: arrayfuncs.c:5106
static bool text_format_parse_digits(const char **ptr, const char *end_ptr, int *value)
Definition: varlena.c:5274
unsigned int pg_wchar
Definition: mbprint.c:31
#define DatumGetVarStringPP(X)
Definition: varlena.c:92
List * textToQualifiedNameList(text *textval)
Definition: varlena.c:3206
#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:3189
char * pq_getmsgtext(StringInfo msg, int rawbytes, int *nbytes)
Definition: pqformat.c:548
#define MaxAllocSize
Definition: memutils.h:40
int skiptable[256]
Definition: varlena.c:57
void appendStringInfoChar(StringInfo str, char ch)
Definition: stringinfo.c:169
void initStringInfo(StringInfo str)
Definition: stringinfo.c:46
Datum byteasend(PG_FUNCTION_ARGS)
Definition: varlena.c:442
void varstr_sortsupport(SortSupport ssup, Oid collid, bool bpchar)
Definition: varlena.c:1808
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1275
Datum text_le(PG_FUNCTION_ARGS)
Definition: varlena.c:1721
Datum hash_uint32(uint32 k)
Definition: hashfunc.c:853
uint8 bits8
Definition: c.h:303
Datum text_substr(PG_FUNCTION_ARGS)
Definition: varlena.c:781
#define TextDatumGetCString(d)
Definition: builtins.h:92
void * palloc0(Size size)
Definition: mcxt.c:877
Datum text_format_nv(PG_FUNCTION_ARGS)
Definition: varlena.c:5501
char * s2
#define PG_RETURN_BOOL(x)
Definition: fmgr.h:319
uintptr_t Datum
Definition: postgres.h:372
int GetDatabaseEncoding(void)
Definition: mbutils.c:1004
Datum text_reverse(PG_FUNCTION_ARGS)
Definition: varlena.c:4942
Datum bytea_substr(PG_FUNCTION_ARGS)
Definition: varlena.c:2832
int bpchartruelen(char *s, int len)
Definition: varchar.c:660
bool scanner_isspace(char ch)
Definition: scansup.c:221
#define wcscoll_l
Definition: win32_port.h:399
#define REGEXP_REPLACE_BACKREF_CNT
Definition: varlena.c:3901
void appendStringInfoSpaces(StringInfo str, int count)
Definition: stringinfo.c:187
int pg_mbstrlen(const char *mbstr)
Definition: mbutils.c:774
double estimateHyperLogLog(hyperLogLogState *cState)
Definition: hyperloglog.c:185
Datum text_ge(PG_FUNCTION_ARGS)
Definition: varlena.c:1751
#define VARSIZE_ANY(PTR)
Definition: postgres.h:334
static void text_position_cleanup(TextPositionState *state)
Definition: varlena.c:1365
Datum byteacmp(PG_FUNCTION_ARGS)
Definition: varlena.c:3633
#define InvalidOid
Definition: postgres_ext.h:36
int pg_mb2wchar_with_len(const char *from, pg_wchar *to, int len)
Definition: mbutils.c:723
void px(PlannerInfo *root, Gene *tour1, Gene *tour2, Gene *offspring, int num_gene, City *city_table)
Datum to_hex32(PG_FUNCTION_ARGS)
Definition: varlena.c:4525
hyperLogLogState full_card
Definition: varlena.c:74
#define PG_RETURN_VOID()
Definition: fmgr.h:309
#define PG_RETURN_TEXT_P(x)
Definition: fmgr.h:331
#define Max(x, y)
Definition: c.h:796
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
bool pg_md5_hash(const void *buff, size_t len, char *hexsum)
Definition: md5.c:293
#define Assert(condition)
Definition: c.h:670
#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:428
void pq_copymsgbytes(StringInfo msg, char *buf, int datalen)
Definition: pqformat.c:530
Datum text_right(PG_FUNCTION_ARGS)
Definition: varlena.c:4921
static text * concat_internal(const char *sepstr, int argidx, FunctionCallInfo fcinfo)
Definition: varlena.c:4785
bool(* abbrev_abort)(int memtupcount, SortSupport ssup)
Definition: sortsupport.h:183
Oid typioparam
Definition: array.h:228
#define PG_RETURN_CSTRING(x)
Definition: fmgr.h:322
Datum unknownin(PG_FUNCTION_ARGS)
Definition: varlena.c:563
size_t Size
Definition: c.h:404
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2477
static bool rest_of_char_same(const char *s1, const char *s2, int len)
Definition: varlena.c:5511
Datum text_pattern_lt(PG_FUNCTION_ARGS)
Definition: varlena.c:2646
struct FmgrInfo FmgrInfo
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:3521
void addHyperLogLog(hyperLogLogState *cState, uint32 hash)
Definition: hyperloglog.c:166
int AggCheckCallContext(FunctionCallInfo fcinfo, MemoryContext *aggcontext)
Definition: nodeAgg.c:4116
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:78
void * fn_extra
Definition: fmgr.h:64
int pg_mblen(const char *mbstr)
Definition: mbutils.c:760
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:3807
#define ARR_NDIM(a)
Definition: array.h:275
Datum byteapos(PG_FUNCTION_ARGS)
Definition: varlena.c:2974
#define TEXTBUFLEN
Definition: varlena.c:83
Oid typiofunc
Definition: array.h:229
#define DatumGetPointer(X)
Definition: postgres.h:555
char typalign
Definition: array.h:226
void deconstruct_array(ArrayType *array, Oid elmtype, int elmlen, bool elmbyval, char elmalign, Datum **elemsp, bool **nullsp, int *nelemsp)
Definition: arrayfuncs.c:3449
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:5042
#define DatumGetBpCharPP(X)
Definition: fmgr.h:257
Oid get_base_element_type(Oid typid)
Definition: lsyscache.c:2571
Datum bttextcmp(PG_FUNCTION_ARGS)
Definition: varlena.c:1766
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
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:848
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:4675
FmgrInfo proc
Definition: array.h:230
Datum bytea_string_agg_transfn(PG_FUNCTION_ARGS)
Definition: varlena.c:450
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:706
Datum byteaoverlay(PG_FUNCTION_ARGS)
Definition: varlena.c:2914
Datum bttext_pattern_sortsupport(PG_FUNCTION_ARGS)
Definition: varlena.c:2726
Datum byteaoctetlen(PG_FUNCTION_ARGS)
Definition: varlena.c:2749
void list_free(List *list)
Definition: list.c:1133
int i
static FmgrInfo * build_concat_foutcache(FunctionCallInfo fcinfo, int argidx)
Definition: varlena.c:4747
Oid element_type
Definition: array.h:223
#define REG_NOMATCH
Definition: regex.h:138
#define NameStr(name)
Definition: c.h:547
static char * locale
Definition: initdb.c:124
void * arg
static bool text_isequal(text *txt1, text *txt2)
Definition: varlena.c:4144
#define PG_GETARG_CSTRING(n)
Definition: fmgr.h:242
#define VAL(CH)
Definition: varlena.c:240