PostgreSQL Source Code  git master
varlena.h File Reference
#include "nodes/pg_list.h"
#include "utils/sortsupport.h"
Include dependency graph for varlena.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Data Structures

struct  ClosestMatchState
 

Typedefs

typedef struct ClosestMatchState ClosestMatchState
 

Functions

int varstr_cmp (const char *arg1, int len1, const char *arg2, int len2, Oid collid)
 
void varstr_sortsupport (SortSupport ssup, Oid typid, Oid collid)
 
int varstr_levenshtein (const char *source, int slen, const char *target, int tlen, int ins_c, int del_c, int sub_c, bool trusted)
 
int varstr_levenshtein_less_equal (const char *source, int slen, const char *target, int tlen, int ins_c, int del_c, int sub_c, int max_d, bool trusted)
 
ListtextToQualifiedNameList (text *textval)
 
bool SplitIdentifierString (char *rawstring, char separator, List **namelist)
 
bool SplitDirectoriesString (char *rawstring, char separator, List **namelist)
 
bool SplitGUCList (char *rawstring, char separator, List **namelist)
 
textreplace_text_regexp (text *src_text, text *pattern_text, text *replace_text, int cflags, Oid collation, int search_start, int n)
 
void initClosestMatch (ClosestMatchState *state, const char *source, int max_d)
 
void updateClosestMatch (ClosestMatchState *state, const char *candidate)
 
const char * getClosestMatch (ClosestMatchState *state)
 

Typedef Documentation

◆ ClosestMatchState

Function Documentation

◆ getClosestMatch()

const char* getClosestMatch ( ClosestMatchState state)

Definition at line 6243 of file varlena.c.

6244 {
6245  Assert(state);
6246 
6247  return state->match;
6248 }
#define Assert(condition)
Definition: c.h:858
Definition: regguts.h:323

References Assert.

Referenced by dblink_fdw_validator(), file_fdw_validator(), postgres_fdw_validator(), and postgresql_fdw_validator().

◆ initClosestMatch()

void initClosestMatch ( ClosestMatchState state,
const char *  source,
int  max_d 
)

Definition at line 6188 of file varlena.c.

6189 {
6190  Assert(state);
6191  Assert(max_d >= 0);
6192 
6193  state->source = source;
6194  state->min_d = -1;
6195  state->max_d = max_d;
6196  state->match = NULL;
6197 }
static rewind_source * source
Definition: pg_rewind.c:89

References Assert, and source.

Referenced by dblink_fdw_validator(), file_fdw_validator(), postgres_fdw_validator(), and postgresql_fdw_validator().

◆ replace_text_regexp()

text* replace_text_regexp ( text src_text,
text pattern_text,
text replace_text,
int  cflags,
Oid  collation,
int  search_start,
int  n 
)

Definition at line 4206 of file varlena.c.

4210 {
4211  text *ret_text;
4212  regex_t *re;
4213  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
4214  int nmatches = 0;
4216  regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
4217  int nmatch = lengthof(pmatch);
4218  pg_wchar *data;
4219  size_t data_len;
4220  int data_pos;
4221  char *start_ptr;
4222  int escape_status;
4223 
4224  initStringInfo(&buf);
4225 
4226  /* Convert data string to wide characters. */
4227  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
4228  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
4229 
4230  /* Check whether replace_text has escapes, especially regexp submatches. */
4232 
4233  /* If no regexp submatches, we can use REG_NOSUB. */
4234  if (escape_status < 2)
4235  {
4236  cflags |= REG_NOSUB;
4237  /* Also tell pg_regexec we only want the whole-match location. */
4238  nmatch = 1;
4239  }
4240 
4241  /* Prepare the regexp. */
4242  re = RE_compile_and_cache(pattern_text, cflags, collation);
4243 
4244  /* start_ptr points to the data_pos'th character of src_text */
4245  start_ptr = (char *) VARDATA_ANY(src_text);
4246  data_pos = 0;
4247 
4248  while (search_start <= data_len)
4249  {
4250  int regexec_result;
4251 
4253 
4254  regexec_result = pg_regexec(re,
4255  data,
4256  data_len,
4257  search_start,
4258  NULL, /* no details */
4259  nmatch,
4260  pmatch,
4261  0);
4262 
4263  if (regexec_result == REG_NOMATCH)
4264  break;
4265 
4266  if (regexec_result != REG_OKAY)
4267  {
4268  char errMsg[100];
4269 
4270  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
4271  ereport(ERROR,
4272  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4273  errmsg("regular expression failed: %s", errMsg)));
4274  }
4275 
4276  /*
4277  * Count matches, and decide whether to replace this match.
4278  */
4279  nmatches++;
4280  if (n > 0 && nmatches != n)
4281  {
4282  /*
4283  * No, so advance search_start, but not start_ptr/data_pos. (Thus,
4284  * we treat the matched text as if it weren't matched, and copy it
4285  * to the output later.)
4286  */
4287  search_start = pmatch[0].rm_eo;
4288  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4289  search_start++;
4290  continue;
4291  }
4292 
4293  /*
4294  * Copy the text to the left of the match position. Note we are given
4295  * character not byte indexes.
4296  */
4297  if (pmatch[0].rm_so - data_pos > 0)
4298  {
4299  int chunk_len;
4300 
4301  chunk_len = charlen_to_bytelen(start_ptr,
4302  pmatch[0].rm_so - data_pos);
4303  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4304 
4305  /*
4306  * Advance start_ptr over that text, to avoid multiple rescans of
4307  * it if the replace_text contains multiple back-references.
4308  */
4309  start_ptr += chunk_len;
4310  data_pos = pmatch[0].rm_so;
4311  }
4312 
4313  /*
4314  * Copy the replace_text, processing escapes if any are present.
4315  */
4316  if (escape_status > 0)
4318  start_ptr, data_pos);
4319  else
4321 
4322  /* Advance start_ptr and data_pos over the matched text. */
4323  start_ptr += charlen_to_bytelen(start_ptr,
4324  pmatch[0].rm_eo - data_pos);
4325  data_pos = pmatch[0].rm_eo;
4326 
4327  /*
4328  * If we only want to replace one occurrence, we're done.
4329  */
4330  if (n > 0)
4331  break;
4332 
4333  /*
4334  * Advance search position. Normally we start the next search at the
4335  * end of the previous match; but if the match was of zero length, we
4336  * have to advance by one character, or we'd just find the same match
4337  * again.
4338  */
4339  search_start = data_pos;
4340  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4341  search_start++;
4342  }
4343 
4344  /*
4345  * Copy the text to the right of the last match.
4346  */
4347  if (data_pos < data_len)
4348  {
4349  int chunk_len;
4350 
4351  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4352  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4353  }
4354 
4355  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4356  pfree(buf.data);
4357  pfree(data);
4358 
4359  return ret_text;
4360 }
#define lengthof(array)
Definition: c.h:788
int errcode(int sqlerrcode)
Definition: elog.c:855
int errmsg(const char *fmt,...)
Definition: elog.c:1068
#define ERROR
Definition: elog.h:39
#define ereport(elevel,...)
Definition: elog.h:149
unsigned int pg_wchar
Definition: mbprint.c:31
int pg_mb2wchar_with_len(const char *from, pg_wchar *to, int len)
Definition: mbutils.c:986
void pfree(void *pointer)
Definition: mcxt.c:1521
void * palloc(Size size)
Definition: mcxt.c:1317
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:122
const void * data
static char * buf
Definition: pg_test_fsync.c:73
size_t pg_regerror(int errcode, const regex_t *preg, char *errbuf, size_t errbuf_size)
Definition: regerror.c:60
#define REG_NOMATCH
Definition: regex.h:216
#define regmatch_t
Definition: regex.h:246
#define REG_OKAY
Definition: regex.h:215
#define REG_NOSUB
Definition: regex.h:185
#define regex_t
Definition: regex.h:245
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:185
regex_t * RE_compile_and_cache(text *text_re, int cflags, Oid collation)
Definition: regexp.c:141
void appendBinaryStringInfo(StringInfo str, const void *data, int datalen)
Definition: stringinfo.c:233
void initStringInfo(StringInfo str)
Definition: stringinfo.c:59
Definition: c.h:687
#define VARSIZE_ANY(PTR)
Definition: varatt.h:311
#define VARDATA_ANY(PTR)
Definition: varatt.h:324
#define VARSIZE_ANY_EXHDR(PTR)
Definition: varatt.h:317
static void appendStringInfoText(StringInfo str, const text *t)
Definition: varlena.c:3982
static int check_replace_text_has_escape(const text *replace_text)
Definition: varlena.c:4073
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:4106
static int charlen_to_bytelen(const char *p, int n)
Definition: varlena.c:806
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:196
Datum replace_text(PG_FUNCTION_ARGS)
Definition: varlena.c:3996

References appendBinaryStringInfo(), appendStringInfoRegexpSubstr(), appendStringInfoText(), buf, charlen_to_bytelen(), CHECK_FOR_INTERRUPTS, check_replace_text_has_escape(), cstring_to_text_with_len(), data, ereport, errcode(), errmsg(), ERROR, initStringInfo(), lengthof, palloc(), pfree(), pg_mb2wchar_with_len(), pg_regerror(), pg_regexec(), RE_compile_and_cache(), REG_NOMATCH, REG_NOSUB, REG_OKAY, regex_t, regmatch_t, replace_text(), VARDATA_ANY, VARSIZE_ANY, and VARSIZE_ANY_EXHDR.

Referenced by textregexreplace(), textregexreplace_extended(), and textregexreplace_noopt().

◆ SplitDirectoriesString()

bool SplitDirectoriesString ( char *  rawstring,
char  separator,
List **  namelist 
)

Definition at line 3584 of file varlena.c.

3586 {
3587  char *nextp = rawstring;
3588  bool done = false;
3589 
3590  *namelist = NIL;
3591 
3592  while (scanner_isspace(*nextp))
3593  nextp++; /* skip leading whitespace */
3594 
3595  if (*nextp == '\0')
3596  return true; /* allow empty string */
3597 
3598  /* At the top of the loop, we are at start of a new directory. */
3599  do
3600  {
3601  char *curname;
3602  char *endp;
3603 
3604  if (*nextp == '"')
3605  {
3606  /* Quoted name --- collapse quote-quote pairs */
3607  curname = nextp + 1;
3608  for (;;)
3609  {
3610  endp = strchr(nextp + 1, '"');
3611  if (endp == NULL)
3612  return false; /* mismatched quotes */
3613  if (endp[1] != '"')
3614  break; /* found end of quoted name */
3615  /* Collapse adjacent quotes into one quote, and look again */
3616  memmove(endp, endp + 1, strlen(endp));
3617  nextp = endp;
3618  }
3619  /* endp now points at the terminating quote */
3620  nextp = endp + 1;
3621  }
3622  else
3623  {
3624  /* Unquoted name --- extends to separator or end of string */
3625  curname = endp = nextp;
3626  while (*nextp && *nextp != separator)
3627  {
3628  /* trailing whitespace should not be included in name */
3629  if (!scanner_isspace(*nextp))
3630  endp = nextp + 1;
3631  nextp++;
3632  }
3633  if (curname == endp)
3634  return false; /* empty unquoted name not allowed */
3635  }
3636 
3637  while (scanner_isspace(*nextp))
3638  nextp++; /* skip trailing whitespace */
3639 
3640  if (*nextp == separator)
3641  {
3642  nextp++;
3643  while (scanner_isspace(*nextp))
3644  nextp++; /* skip leading whitespace for next */
3645  /* we expect another name, so done remains false */
3646  }
3647  else if (*nextp == '\0')
3648  done = true;
3649  else
3650  return false; /* invalid syntax */
3651 
3652  /* Now safe to overwrite separator with a null */
3653  *endp = '\0';
3654 
3655  /* Truncate path if it's overlength */
3656  if (strlen(curname) >= MAXPGPATH)
3657  curname[MAXPGPATH - 1] = '\0';
3658 
3659  /*
3660  * Finished isolating current name --- add it to list
3661  */
3662  curname = pstrdup(curname);
3663  canonicalize_path(curname);
3664  *namelist = lappend(*namelist, curname);
3665 
3666  /* Loop back if we didn't reach end of string */
3667  } while (!done);
3668 
3669  return true;
3670 }
List * lappend(List *list, void *datum)
Definition: list.c:339
char * pstrdup(const char *in)
Definition: mcxt.c:1696
#define MAXPGPATH
#define NIL
Definition: pg_list.h:68
void canonicalize_path(char *path)
Definition: path.c:264
bool scanner_isspace(char ch)
Definition: scansup.c:117

References canonicalize_path(), lappend(), MAXPGPATH, NIL, pstrdup(), and scanner_isspace().

Referenced by load_libraries(), and PostmasterMain().

◆ SplitGUCList()

bool SplitGUCList ( char *  rawstring,
char  separator,
List **  namelist 
)

Definition at line 3705 of file varlena.c.

3707 {
3708  char *nextp = rawstring;
3709  bool done = false;
3710 
3711  *namelist = NIL;
3712 
3713  while (scanner_isspace(*nextp))
3714  nextp++; /* skip leading whitespace */
3715 
3716  if (*nextp == '\0')
3717  return true; /* allow empty string */
3718 
3719  /* At the top of the loop, we are at start of a new identifier. */
3720  do
3721  {
3722  char *curname;
3723  char *endp;
3724 
3725  if (*nextp == '"')
3726  {
3727  /* Quoted name --- collapse quote-quote pairs */
3728  curname = nextp + 1;
3729  for (;;)
3730  {
3731  endp = strchr(nextp + 1, '"');
3732  if (endp == NULL)
3733  return false; /* mismatched quotes */
3734  if (endp[1] != '"')
3735  break; /* found end of quoted name */
3736  /* Collapse adjacent quotes into one quote, and look again */
3737  memmove(endp, endp + 1, strlen(endp));
3738  nextp = endp;
3739  }
3740  /* endp now points at the terminating quote */
3741  nextp = endp + 1;
3742  }
3743  else
3744  {
3745  /* Unquoted name --- extends to separator or whitespace */
3746  curname = nextp;
3747  while (*nextp && *nextp != separator &&
3748  !scanner_isspace(*nextp))
3749  nextp++;
3750  endp = nextp;
3751  if (curname == nextp)
3752  return false; /* empty unquoted name not allowed */
3753  }
3754 
3755  while (scanner_isspace(*nextp))
3756  nextp++; /* skip trailing whitespace */
3757 
3758  if (*nextp == separator)
3759  {
3760  nextp++;
3761  while (scanner_isspace(*nextp))
3762  nextp++; /* skip leading whitespace for next */
3763  /* we expect another name, so done remains false */
3764  }
3765  else if (*nextp == '\0')
3766  done = true;
3767  else
3768  return false; /* invalid syntax */
3769 
3770  /* Now safe to overwrite separator with a null */
3771  *endp = '\0';
3772 
3773  /*
3774  * Finished isolating current name --- add it to list
3775  */
3776  *namelist = lappend(*namelist, curname);
3777 
3778  /* Loop back if we didn't reach end of string */
3779  } while (!done);
3780 
3781  return true;
3782 }

References lappend(), NIL, and scanner_isspace().

Referenced by check_debug_io_direct(), dumpFunc(), parse_hba_auth_opt(), pg_get_functiondef(), and PostmasterMain().

◆ SplitIdentifierString()

bool SplitIdentifierString ( char *  rawstring,
char  separator,
List **  namelist 
)

Definition at line 3457 of file varlena.c.

3459 {
3460  char *nextp = rawstring;
3461  bool done = false;
3462 
3463  *namelist = NIL;
3464 
3465  while (scanner_isspace(*nextp))
3466  nextp++; /* skip leading whitespace */
3467 
3468  if (*nextp == '\0')
3469  return true; /* allow empty string */
3470 
3471  /* At the top of the loop, we are at start of a new identifier. */
3472  do
3473  {
3474  char *curname;
3475  char *endp;
3476 
3477  if (*nextp == '"')
3478  {
3479  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3480  curname = nextp + 1;
3481  for (;;)
3482  {
3483  endp = strchr(nextp + 1, '"');
3484  if (endp == NULL)
3485  return false; /* mismatched quotes */
3486  if (endp[1] != '"')
3487  break; /* found end of quoted name */
3488  /* Collapse adjacent quotes into one quote, and look again */
3489  memmove(endp, endp + 1, strlen(endp));
3490  nextp = endp;
3491  }
3492  /* endp now points at the terminating quote */
3493  nextp = endp + 1;
3494  }
3495  else
3496  {
3497  /* Unquoted name --- extends to separator or whitespace */
3498  char *downname;
3499  int len;
3500 
3501  curname = nextp;
3502  while (*nextp && *nextp != separator &&
3503  !scanner_isspace(*nextp))
3504  nextp++;
3505  endp = nextp;
3506  if (curname == nextp)
3507  return false; /* empty unquoted name not allowed */
3508 
3509  /*
3510  * Downcase the identifier, using same code as main lexer does.
3511  *
3512  * XXX because we want to overwrite the input in-place, we cannot
3513  * support a downcasing transformation that increases the string
3514  * length. This is not a problem given the current implementation
3515  * of downcase_truncate_identifier, but we'll probably have to do
3516  * something about this someday.
3517  */
3518  len = endp - curname;
3519  downname = downcase_truncate_identifier(curname, len, false);
3520  Assert(strlen(downname) <= len);
3521  strncpy(curname, downname, len); /* strncpy is required here */
3522  pfree(downname);
3523  }
3524 
3525  while (scanner_isspace(*nextp))
3526  nextp++; /* skip trailing whitespace */
3527 
3528  if (*nextp == separator)
3529  {
3530  nextp++;
3531  while (scanner_isspace(*nextp))
3532  nextp++; /* skip leading whitespace for next */
3533  /* we expect another name, so done remains false */
3534  }
3535  else if (*nextp == '\0')
3536  done = true;
3537  else
3538  return false; /* invalid syntax */
3539 
3540  /* Now safe to overwrite separator with a null */
3541  *endp = '\0';
3542 
3543  /* Truncate name if it's overlength */
3544  truncate_identifier(curname, strlen(curname), false);
3545 
3546  /*
3547  * Finished isolating current name --- add it to list
3548  */
3549  *namelist = lappend(*namelist, curname);
3550 
3551  /* Loop back if we didn't reach end of string */
3552  } while (!done);
3553 
3554  return true;
3555 }
const void size_t len
void truncate_identifier(char *ident, int len, bool warn)
Definition: scansup.c:93
char * downcase_truncate_identifier(const char *ident, int len, bool warn)
Definition: scansup.c:37

References Assert, downcase_truncate_identifier(), lappend(), len, NIL, pfree(), scanner_isspace(), and truncate_identifier().

Referenced by check_createrole_self_grant(), check_datestyle(), check_log_destination(), check_search_path(), check_temp_tablespaces(), check_wal_consistency_checking(), ExtractExtensionList(), parse_extension_control_file(), parse_output_parameters(), parse_publication_options(), plpgsql_extra_checks_check_hook(), PrepareTempTablespaces(), preprocessNamespacePath(), stringToQualifiedNameList(), textToQualifiedNameList(), and validate_sync_standby_slots().

◆ textToQualifiedNameList()

List* textToQualifiedNameList ( text textval)

Definition at line 3399 of file varlena.c.

3400 {
3401  char *rawname;
3402  List *result = NIL;
3403  List *namelist;
3404  ListCell *l;
3405 
3406  /* Convert to C string (handles possible detoasting). */
3407  /* Note we rely on being able to modify rawname below. */
3408  rawname = text_to_cstring(textval);
3409 
3410  if (!SplitIdentifierString(rawname, '.', &namelist))
3411  ereport(ERROR,
3412  (errcode(ERRCODE_INVALID_NAME),
3413  errmsg("invalid name syntax")));
3414 
3415  if (namelist == NIL)
3416  ereport(ERROR,
3417  (errcode(ERRCODE_INVALID_NAME),
3418  errmsg("invalid name syntax")));
3419 
3420  foreach(l, namelist)
3421  {
3422  char *curname = (char *) lfirst(l);
3423 
3424  result = lappend(result, makeString(pstrdup(curname)));
3425  }
3426 
3427  pfree(rawname);
3428  list_free(namelist);
3429 
3430  return result;
3431 }
void list_free(List *list)
Definition: list.c:1546
#define lfirst(lc)
Definition: pg_list.h:172
Definition: pg_list.h:54
String * makeString(char *str)
Definition: value.c:63
char * text_to_cstring(const text *t)
Definition: varlena.c:217
bool SplitIdentifierString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3457

References ereport, errcode(), errmsg(), ERROR, lappend(), lfirst, list_free(), makeString(), NIL, pfree(), pstrdup(), SplitIdentifierString(), and text_to_cstring().

Referenced by bt_metap(), bt_multi_page_stats(), bt_page_items_internal(), bt_page_stats_internal(), convert_table_name(), currtid_byrelname(), get_raw_page_internal(), get_rel_from_relname(), nextval(), pg_get_serial_sequence(), pg_get_viewdef_name(), pg_get_viewdef_name_ext(), pg_relpages(), pg_relpages_v1_5(), pgrowlocks(), pgstatindex(), pgstatindex_v1_5(), pgstattuple(), pgstattuple_v1_5(), row_security_active_name(), text_regclass(), ts_parse_byname(), and ts_token_type_byname().

◆ updateClosestMatch()

void updateClosestMatch ( ClosestMatchState state,
const char *  candidate 
)

Definition at line 6208 of file varlena.c.

6209 {
6210  int dist;
6211 
6212  Assert(state);
6213 
6214  if (state->source == NULL || state->source[0] == '\0' ||
6215  candidate == NULL || candidate[0] == '\0')
6216  return;
6217 
6218  /*
6219  * To avoid ERROR-ing, we check the lengths here instead of setting
6220  * 'trusted' to false in the call to varstr_levenshtein_less_equal().
6221  */
6222  if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
6223  strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
6224  return;
6225 
6226  dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
6227  candidate, strlen(candidate), 1, 1, 1,
6228  state->max_d, true);
6229  if (dist <= state->max_d &&
6230  dist <= strlen(state->source) / 2 &&
6231  (state->min_d == -1 || dist < state->min_d))
6232  {
6233  state->min_d = dist;
6234  state->match = candidate;
6235  }
6236 }
#define MAX_LEVENSHTEIN_STRLEN
Definition: levenshtein.c:26
int varstr_levenshtein_less_equal(const char *source, int slen, const char *target, int tlen, int ins_c, int del_c, int sub_c, int max_d, bool trusted)

References Assert, MAX_LEVENSHTEIN_STRLEN, and varstr_levenshtein_less_equal().

Referenced by dblink_fdw_validator(), file_fdw_validator(), postgres_fdw_validator(), and postgresql_fdw_validator().

◆ varstr_cmp()

int varstr_cmp ( const char *  arg1,
int  len1,
const char *  arg2,
int  len2,
Oid  collid 
)

Definition at line 1539 of file varlena.c.

1540 {
1541  int result;
1542 
1544 
1545  /*
1546  * Unfortunately, there is no strncoll(), so in the non-C locale case we
1547  * have to do some memory copying. This turns out to be significantly
1548  * slower, so we optimize the case where LC_COLLATE is C. We also try to
1549  * optimize relatively-short strings by avoiding palloc/pfree overhead.
1550  */
1551  if (lc_collate_is_c(collid))
1552  {
1553  result = memcmp(arg1, arg2, Min(len1, len2));
1554  if ((result == 0) && (len1 != len2))
1555  result = (len1 < len2) ? -1 : 1;
1556  }
1557  else
1558  {
1559  pg_locale_t mylocale;
1560 
1561  mylocale = pg_newlocale_from_collation(collid);
1562 
1563  /*
1564  * memcmp() can't tell us which of two unequal strings sorts first,
1565  * but it's a cheap way to tell if they're equal. Testing shows that
1566  * memcmp() followed by strcoll() is only trivially slower than
1567  * strcoll() by itself, so we don't lose much if this doesn't work out
1568  * very often, and if it does - for example, because there are many
1569  * equal strings in the input - then we win big by avoiding expensive
1570  * collation-aware comparisons.
1571  */
1572  if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1573  return 0;
1574 
1575  result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1576 
1577  /* Break tie if necessary. */
1578  if (result == 0 && pg_locale_deterministic(mylocale))
1579  {
1580  result = memcmp(arg1, arg2, Min(len1, len2));
1581  if ((result == 0) && (len1 != len2))
1582  result = (len1 < len2) ? -1 : 1;
1583  }
1584  }
1585 
1586  return result;
1587 }
#define Min(x, y)
Definition: c.h:1004
Oid collid
int pg_strncoll(const char *arg1, size_t len1, const char *arg2, size_t len2, pg_locale_t locale)
Definition: pg_locale.c:2134
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1317
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1552
bool pg_locale_deterministic(pg_locale_t locale)
Definition: pg_locale.c:1532
static void check_collation_set(Oid collid)
Definition: varlena.c:1510

References check_collation_set(), collid, lc_collate_is_c(), Min, pg_locale_deterministic(), pg_newlocale_from_collation(), and pg_strncoll().

Referenced by bpchar_larger(), bpchar_smaller(), bpcharcmp(), bpchareq(), bpcharge(), bpchargt(), bpcharle(), bpcharlt(), bpcharne(), btnametextcmp(), bttextnamecmp(), citextcmp(), compareJsonbScalarValue(), gin_compare_jsonb(), make_greater_string(), namecmp(), nameeqtext(), namenetext(), spg_text_leaf_consistent(), text_cmp(), texteqname(), and textnename().

◆ varstr_levenshtein()

int varstr_levenshtein ( const char *  source,
int  slen,
const char *  target,
int  tlen,
int  ins_c,
int  del_c,
int  sub_c,
bool  trusted 
)

Definition at line 73 of file levenshtein.c.

78 {
79  int m,
80  n;
81  int *prev;
82  int *curr;
83  int *s_char_len = NULL;
84  int j;
85  const char *y;
86 
87  /*
88  * For varstr_levenshtein_less_equal, we have real variables called
89  * start_column and stop_column; otherwise it's just short-hand for 0 and
90  * m.
91  */
92 #ifdef LEVENSHTEIN_LESS_EQUAL
93  int start_column,
94  stop_column;
95 
96 #undef START_COLUMN
97 #undef STOP_COLUMN
98 #define START_COLUMN start_column
99 #define STOP_COLUMN stop_column
100 #else
101 #undef START_COLUMN
102 #undef STOP_COLUMN
103 #define START_COLUMN 0
104 #define STOP_COLUMN m
105 #endif
106 
107  /* Convert string lengths (in bytes) to lengths in characters */
108  m = pg_mbstrlen_with_len(source, slen);
109  n = pg_mbstrlen_with_len(target, tlen);
110 
111  /*
112  * We can transform an empty s into t with n insertions, or a non-empty t
113  * into an empty s with m deletions.
114  */
115  if (!m)
116  return n * ins_c;
117  if (!n)
118  return m * del_c;
119 
120  /*
121  * For security concerns, restrict excessive CPU+RAM usage. (This
122  * implementation uses O(m) memory and has O(mn) complexity.) If
123  * "trusted" is true, caller is responsible for not making excessive
124  * requests, typically by using a small max_d along with strings that are
125  * bounded, though not necessarily to MAX_LEVENSHTEIN_STRLEN exactly.
126  */
127  if (!trusted &&
128  (m > MAX_LEVENSHTEIN_STRLEN ||
130  ereport(ERROR,
131  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
132  errmsg("levenshtein argument exceeds maximum length of %d characters",
134 
135 #ifdef LEVENSHTEIN_LESS_EQUAL
136  /* Initialize start and stop columns. */
137  start_column = 0;
138  stop_column = m + 1;
139 
140  /*
141  * If max_d >= 0, determine whether the bound is impossibly tight. If so,
142  * return max_d + 1 immediately. Otherwise, determine whether it's tight
143  * enough to limit the computation we must perform. If so, figure out
144  * initial stop column.
145  */
146  if (max_d >= 0)
147  {
148  int min_theo_d; /* Theoretical minimum distance. */
149  int max_theo_d; /* Theoretical maximum distance. */
150  int net_inserts = n - m;
151 
152  min_theo_d = net_inserts < 0 ?
153  -net_inserts * del_c : net_inserts * ins_c;
154  if (min_theo_d > max_d)
155  return max_d + 1;
156  if (ins_c + del_c < sub_c)
157  sub_c = ins_c + del_c;
158  max_theo_d = min_theo_d + sub_c * Min(m, n);
159  if (max_d >= max_theo_d)
160  max_d = -1;
161  else if (ins_c + del_c > 0)
162  {
163  /*
164  * Figure out how much of the first row of the notional matrix we
165  * need to fill in. If the string is growing, the theoretical
166  * minimum distance already incorporates the cost of deleting the
167  * number of characters necessary to make the two strings equal in
168  * length. Each additional deletion forces another insertion, so
169  * the best-case total cost increases by ins_c + del_c. If the
170  * string is shrinking, the minimum theoretical cost assumes no
171  * excess deletions; that is, we're starting no further right than
172  * column n - m. If we do start further right, the best-case
173  * total cost increases by ins_c + del_c for each move right.
174  */
175  int slack_d = max_d - min_theo_d;
176  int best_column = net_inserts < 0 ? -net_inserts : 0;
177 
178  stop_column = best_column + (slack_d / (ins_c + del_c)) + 1;
179  if (stop_column > m)
180  stop_column = m + 1;
181  }
182  }
183 #endif
184 
185  /*
186  * In order to avoid calling pg_mblen() repeatedly on each character in s,
187  * we cache all the lengths before starting the main loop -- but if all
188  * the characters in both strings are single byte, then we skip this and
189  * use a fast-path in the main loop. If only one string contains
190  * multi-byte characters, we still build the array, so that the fast-path
191  * needn't deal with the case where the array hasn't been initialized.
192  */
193  if (m != slen || n != tlen)
194  {
195  int i;
196  const char *cp = source;
197 
198  s_char_len = (int *) palloc((m + 1) * sizeof(int));
199  for (i = 0; i < m; ++i)
200  {
201  s_char_len[i] = pg_mblen(cp);
202  cp += s_char_len[i];
203  }
204  s_char_len[i] = 0;
205  }
206 
207  /* One more cell for initialization column and row. */
208  ++m;
209  ++n;
210 
211  /* Previous and current rows of notional array. */
212  prev = (int *) palloc(2 * m * sizeof(int));
213  curr = prev + m;
214 
215  /*
216  * To transform the first i characters of s into the first 0 characters of
217  * t, we must perform i deletions.
218  */
219  for (int i = START_COLUMN; i < STOP_COLUMN; i++)
220  prev[i] = i * del_c;
221 
222  /* Loop through rows of the notional array */
223  for (y = target, j = 1; j < n; j++)
224  {
225  int *temp;
226  const char *x = source;
227  int y_char_len = n != tlen + 1 ? pg_mblen(y) : 1;
228  int i;
229 
230 #ifdef LEVENSHTEIN_LESS_EQUAL
231 
232  /*
233  * In the best case, values percolate down the diagonal unchanged, so
234  * we must increment stop_column unless it's already on the right end
235  * of the array. The inner loop will read prev[stop_column], so we
236  * have to initialize it even though it shouldn't affect the result.
237  */
238  if (stop_column < m)
239  {
240  prev[stop_column] = max_d + 1;
241  ++stop_column;
242  }
243 
244  /*
245  * The main loop fills in curr, but curr[0] needs a special case: to
246  * transform the first 0 characters of s into the first j characters
247  * of t, we must perform j insertions. However, if start_column > 0,
248  * this special case does not apply.
249  */
250  if (start_column == 0)
251  {
252  curr[0] = j * ins_c;
253  i = 1;
254  }
255  else
256  i = start_column;
257 #else
258  curr[0] = j * ins_c;
259  i = 1;
260 #endif
261 
262  /*
263  * This inner loop is critical to performance, so we include a
264  * fast-path to handle the (fairly common) case where no multibyte
265  * characters are in the mix. The fast-path is entitled to assume
266  * that if s_char_len is not initialized then BOTH strings contain
267  * only single-byte characters.
268  */
269  if (s_char_len != NULL)
270  {
271  for (; i < STOP_COLUMN; i++)
272  {
273  int ins;
274  int del;
275  int sub;
276  int x_char_len = s_char_len[i - 1];
277 
278  /*
279  * Calculate costs for insertion, deletion, and substitution.
280  *
281  * When calculating cost for substitution, we compare the last
282  * character of each possibly-multibyte character first,
283  * because that's enough to rule out most mis-matches. If we
284  * get past that test, then we compare the lengths and the
285  * remaining bytes.
286  */
287  ins = prev[i] + ins_c;
288  del = curr[i - 1] + del_c;
289  if (x[x_char_len - 1] == y[y_char_len - 1]
290  && x_char_len == y_char_len &&
291  (x_char_len == 1 || rest_of_char_same(x, y, x_char_len)))
292  sub = prev[i - 1];
293  else
294  sub = prev[i - 1] + sub_c;
295 
296  /* Take the one with minimum cost. */
297  curr[i] = Min(ins, del);
298  curr[i] = Min(curr[i], sub);
299 
300  /* Point to next character. */
301  x += x_char_len;
302  }
303  }
304  else
305  {
306  for (; i < STOP_COLUMN; i++)
307  {
308  int ins;
309  int del;
310  int sub;
311 
312  /* Calculate costs for insertion, deletion, and substitution. */
313  ins = prev[i] + ins_c;
314  del = curr[i - 1] + del_c;
315  sub = prev[i - 1] + ((*x == *y) ? 0 : sub_c);
316 
317  /* Take the one with minimum cost. */
318  curr[i] = Min(ins, del);
319  curr[i] = Min(curr[i], sub);
320 
321  /* Point to next character. */
322  x++;
323  }
324  }
325 
326  /* Swap current row with previous row. */
327  temp = curr;
328  curr = prev;
329  prev = temp;
330 
331  /* Point to next character. */
332  y += y_char_len;
333 
334 #ifdef LEVENSHTEIN_LESS_EQUAL
335 
336  /*
337  * This chunk of code represents a significant performance hit if used
338  * in the case where there is no max_d bound. This is probably not
339  * because the max_d >= 0 test itself is expensive, but rather because
340  * the possibility of needing to execute this code prevents tight
341  * optimization of the loop as a whole.
342  */
343  if (max_d >= 0)
344  {
345  /*
346  * The "zero point" is the column of the current row where the
347  * remaining portions of the strings are of equal length. There
348  * are (n - 1) characters in the target string, of which j have
349  * been transformed. There are (m - 1) characters in the source
350  * string, so we want to find the value for zp where (n - 1) - j =
351  * (m - 1) - zp.
352  */
353  int zp = j - (n - m);
354 
355  /* Check whether the stop column can slide left. */
356  while (stop_column > 0)
357  {
358  int ii = stop_column - 1;
359  int net_inserts = ii - zp;
360 
361  if (prev[ii] + (net_inserts > 0 ? net_inserts * ins_c :
362  -net_inserts * del_c) <= max_d)
363  break;
364  stop_column--;
365  }
366 
367  /* Check whether the start column can slide right. */
368  while (start_column < stop_column)
369  {
370  int net_inserts = start_column - zp;
371 
372  if (prev[start_column] +
373  (net_inserts > 0 ? net_inserts * ins_c :
374  -net_inserts * del_c) <= max_d)
375  break;
376 
377  /*
378  * We'll never again update these values, so we must make sure
379  * there's nothing here that could confuse any future
380  * iteration of the outer loop.
381  */
382  prev[start_column] = max_d + 1;
383  curr[start_column] = max_d + 1;
384  if (start_column != 0)
385  source += (s_char_len != NULL) ? s_char_len[start_column - 1] : 1;
386  start_column++;
387  }
388 
389  /* If they cross, we're going to exceed the bound. */
390  if (start_column >= stop_column)
391  return max_d + 1;
392  }
393 #endif
394  }
395 
396  /*
397  * Because the final value was swapped from the previous row to the
398  * current row, that's where we'll find it.
399  */
400  return prev[m - 1];
401 }
int y
Definition: isn.c:72
int x
Definition: isn.c:71
int j
Definition: isn.c:74
int i
Definition: isn.c:73
#define START_COLUMN
#define STOP_COLUMN
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:1057
int pg_mblen(const char *mbstr)
Definition: mbutils.c:1023
static bool rest_of_char_same(const char *s1, const char *s2, int len)
Definition: varlena.c:6152

References ereport, errcode(), errmsg(), ERROR, i, j, MAX_LEVENSHTEIN_STRLEN, Min, palloc(), pg_mblen(), pg_mbstrlen_with_len(), rest_of_char_same(), source, START_COLUMN, STOP_COLUMN, x, and y.

Referenced by levenshtein(), and levenshtein_with_costs().

◆ varstr_levenshtein_less_equal()

int varstr_levenshtein_less_equal ( const char *  source,
int  slen,
const char *  target,
int  tlen,
int  ins_c,
int  del_c,
int  sub_c,
int  max_d,
bool  trusted 
)

◆ varstr_sortsupport()

void varstr_sortsupport ( SortSupport  ssup,
Oid  typid,
Oid  collid 
)

Definition at line 1873 of file varlena.c.

1874 {
1875  bool abbreviate = ssup->abbreviate;
1876  bool collate_c = false;
1877  VarStringSortSupport *sss;
1878  pg_locale_t locale = 0;
1879 
1881 
1882  /*
1883  * If possible, set ssup->comparator to a function which can be used to
1884  * directly compare two datums. If we can do this, we'll avoid the
1885  * overhead of a trip through the fmgr layer for every comparison, which
1886  * can be substantial.
1887  *
1888  * Most typically, we'll set the comparator to varlenafastcmp_locale,
1889  * which uses strcoll() to perform comparisons. We use that for the
1890  * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1891  * LC_COLLATE = C, we can make things quite a bit faster with
1892  * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1893  * memcmp() rather than strcoll().
1894  */
1895  if (lc_collate_is_c(collid))
1896  {
1897  if (typid == BPCHAROID)
1898  ssup->comparator = bpcharfastcmp_c;
1899  else if (typid == NAMEOID)
1900  {
1901  ssup->comparator = namefastcmp_c;
1902  /* Not supporting abbreviation with type NAME, for now */
1903  abbreviate = false;
1904  }
1905  else
1906  ssup->comparator = varstrfastcmp_c;
1907 
1908  collate_c = true;
1909  }
1910  else
1911  {
1912  /*
1913  * We need a collation-sensitive comparison. To make things faster,
1914  * we'll figure out the collation based on the locale id and cache the
1915  * result.
1916  */
1918 
1919  /*
1920  * We use varlenafastcmp_locale except for type NAME.
1921  */
1922  if (typid == NAMEOID)
1923  {
1925  /* Not supporting abbreviation with type NAME, for now */
1926  abbreviate = false;
1927  }
1928  else
1930  }
1931 
1932  /*
1933  * Unfortunately, it seems that abbreviation for non-C collations is
1934  * broken on many common platforms; see pg_strxfrm_enabled().
1935  *
1936  * Even apart from the risk of broken locales, it's possible that there
1937  * are platforms where the use of abbreviated keys should be disabled at
1938  * compile time. Having only 4 byte datums could make worst-case
1939  * performance drastically more likely, for example. Moreover, macOS's
1940  * strxfrm() implementation is known to not effectively concentrate a
1941  * significant amount of entropy from the original string in earlier
1942  * transformed blobs. It's possible that other supported platforms are
1943  * similarly encumbered. So, if we ever get past disabling this
1944  * categorically, we may still want or need to disable it for particular
1945  * platforms.
1946  */
1947  if (!collate_c && !pg_strxfrm_enabled(locale))
1948  abbreviate = false;
1949 
1950  /*
1951  * If we're using abbreviated keys, or if we're using a locale-aware
1952  * comparison, we need to initialize a VarStringSortSupport object. Both
1953  * cases will make use of the temporary buffers we initialize here for
1954  * scratch space (and to detect requirement for BpChar semantics from
1955  * caller), and the abbreviation case requires additional state.
1956  */
1957  if (abbreviate || !collate_c)
1958  {
1959  sss = palloc(sizeof(VarStringSortSupport));
1960  sss->buf1 = palloc(TEXTBUFLEN);
1961  sss->buflen1 = TEXTBUFLEN;
1962  sss->buf2 = palloc(TEXTBUFLEN);
1963  sss->buflen2 = TEXTBUFLEN;
1964  /* Start with invalid values */
1965  sss->last_len1 = -1;
1966  sss->last_len2 = -1;
1967  /* Initialize */
1968  sss->last_returned = 0;
1969  sss->locale = locale;
1970 
1971  /*
1972  * To avoid somehow confusing a strxfrm() blob and an original string,
1973  * constantly keep track of the variety of data that buf1 and buf2
1974  * currently contain.
1975  *
1976  * Comparisons may be interleaved with conversion calls. Frequently,
1977  * conversions and comparisons are batched into two distinct phases,
1978  * but the correctness of caching cannot hinge upon this. For
1979  * comparison caching, buffer state is only trusted if cache_blob is
1980  * found set to false, whereas strxfrm() caching only trusts the state
1981  * when cache_blob is found set to true.
1982  *
1983  * Arbitrarily initialize cache_blob to true.
1984  */
1985  sss->cache_blob = true;
1986  sss->collate_c = collate_c;
1987  sss->typid = typid;
1988  ssup->ssup_extra = sss;
1989 
1990  /*
1991  * If possible, plan to use the abbreviated keys optimization. The
1992  * core code may switch back to authoritative comparator should
1993  * abbreviation be aborted.
1994  */
1995  if (abbreviate)
1996  {
1997  sss->prop_card = 0.20;
1998  initHyperLogLog(&sss->abbr_card, 10);
1999  initHyperLogLog(&sss->full_card, 10);
2000  ssup->abbrev_full_comparator = ssup->comparator;
2004  }
2005  }
2006 }
void initHyperLogLog(hyperLogLogState *cState, uint8 bwidth)
Definition: hyperloglog.c:66
static char * locale
Definition: initdb.c:140
bool pg_strxfrm_enabled(pg_locale_t locale)
Definition: pg_locale.c:2350
int(* comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:106
Datum(* abbrev_converter)(Datum original, SortSupport ssup)
Definition: sortsupport.h:172
void * ssup_extra
Definition: sortsupport.h:87
int(* abbrev_full_comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:191
bool(* abbrev_abort)(int memtupcount, SortSupport ssup)
Definition: sortsupport.h:182
pg_locale_t locale
Definition: varlena.c:96
hyperLogLogState full_card
Definition: varlena.c:94
hyperLogLogState abbr_card
Definition: varlena.c:93
int ssup_datum_unsigned_cmp(Datum x, Datum y, SortSupport ssup)
Definition: tuplesort.c:3178
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2437
static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2094
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2049
static int namefastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2082
static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2125
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2239
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2012
#define TEXTBUFLEN
Definition: varlena.c:114

References VarStringSortSupport::abbr_card, SortSupportData::abbrev_abort, SortSupportData::abbrev_converter, SortSupportData::abbrev_full_comparator, SortSupportData::abbreviate, bpcharfastcmp_c(), VarStringSortSupport::buf1, VarStringSortSupport::buf2, VarStringSortSupport::buflen1, VarStringSortSupport::buflen2, VarStringSortSupport::cache_blob, check_collation_set(), VarStringSortSupport::collate_c, collid, SortSupportData::comparator, VarStringSortSupport::full_card, initHyperLogLog(), VarStringSortSupport::last_len1, VarStringSortSupport::last_len2, VarStringSortSupport::last_returned, lc_collate_is_c(), VarStringSortSupport::locale, locale, namefastcmp_c(), namefastcmp_locale(), palloc(), pg_newlocale_from_collation(), pg_strxfrm_enabled(), VarStringSortSupport::prop_card, ssup_datum_unsigned_cmp(), SortSupportData::ssup_extra, TEXTBUFLEN, VarStringSortSupport::typid, varlenafastcmp_locale(), varstr_abbrev_abort(), varstr_abbrev_convert(), and varstrfastcmp_c().

Referenced by bpchar_sortsupport(), btbpchar_pattern_sortsupport(), btnamesortsupport(), bttext_pattern_sortsupport(), bttextsortsupport(), and bytea_sortsupport().