PostgreSQL Source Code  git master
All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Pages
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 6256 of file varlena.c.

6257 {
6258  Assert(state);
6259 
6260  return state->match;
6261 }
#define Assert(condition)
Definition: c.h:837
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 6201 of file varlena.c.

6202 {
6203  Assert(state);
6204  Assert(max_d >= 0);
6205 
6206  state->source = source;
6207  state->min_d = -1;
6208  state->max_d = max_d;
6209  state->match = NULL;
6210 }
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 4219 of file varlena.c.

4223 {
4224  text *ret_text;
4225  regex_t *re;
4226  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
4227  int nmatches = 0;
4229  regmatch_t pmatch[10]; /* main match, plus \1 to \9 */
4230  int nmatch = lengthof(pmatch);
4231  pg_wchar *data;
4232  size_t data_len;
4233  int data_pos;
4234  char *start_ptr;
4235  int escape_status;
4236 
4237  initStringInfo(&buf);
4238 
4239  /* Convert data string to wide characters. */
4240  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
4241  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
4242 
4243  /* Check whether replace_text has escapes, especially regexp submatches. */
4245 
4246  /* If no regexp submatches, we can use REG_NOSUB. */
4247  if (escape_status < 2)
4248  {
4249  cflags |= REG_NOSUB;
4250  /* Also tell pg_regexec we only want the whole-match location. */
4251  nmatch = 1;
4252  }
4253 
4254  /* Prepare the regexp. */
4255  re = RE_compile_and_cache(pattern_text, cflags, collation);
4256 
4257  /* start_ptr points to the data_pos'th character of src_text */
4258  start_ptr = (char *) VARDATA_ANY(src_text);
4259  data_pos = 0;
4260 
4261  while (search_start <= data_len)
4262  {
4263  int regexec_result;
4264 
4266 
4267  regexec_result = pg_regexec(re,
4268  data,
4269  data_len,
4270  search_start,
4271  NULL, /* no details */
4272  nmatch,
4273  pmatch,
4274  0);
4275 
4276  if (regexec_result == REG_NOMATCH)
4277  break;
4278 
4279  if (regexec_result != REG_OKAY)
4280  {
4281  char errMsg[100];
4282 
4283  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
4284  ereport(ERROR,
4285  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4286  errmsg("regular expression failed: %s", errMsg)));
4287  }
4288 
4289  /*
4290  * Count matches, and decide whether to replace this match.
4291  */
4292  nmatches++;
4293  if (n > 0 && nmatches != n)
4294  {
4295  /*
4296  * No, so advance search_start, but not start_ptr/data_pos. (Thus,
4297  * we treat the matched text as if it weren't matched, and copy it
4298  * to the output later.)
4299  */
4300  search_start = pmatch[0].rm_eo;
4301  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4302  search_start++;
4303  continue;
4304  }
4305 
4306  /*
4307  * Copy the text to the left of the match position. Note we are given
4308  * character not byte indexes.
4309  */
4310  if (pmatch[0].rm_so - data_pos > 0)
4311  {
4312  int chunk_len;
4313 
4314  chunk_len = charlen_to_bytelen(start_ptr,
4315  pmatch[0].rm_so - data_pos);
4316  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4317 
4318  /*
4319  * Advance start_ptr over that text, to avoid multiple rescans of
4320  * it if the replace_text contains multiple back-references.
4321  */
4322  start_ptr += chunk_len;
4323  data_pos = pmatch[0].rm_so;
4324  }
4325 
4326  /*
4327  * Copy the replace_text, processing escapes if any are present.
4328  */
4329  if (escape_status > 0)
4331  start_ptr, data_pos);
4332  else
4334 
4335  /* Advance start_ptr and data_pos over the matched text. */
4336  start_ptr += charlen_to_bytelen(start_ptr,
4337  pmatch[0].rm_eo - data_pos);
4338  data_pos = pmatch[0].rm_eo;
4339 
4340  /*
4341  * If we only want to replace one occurrence, we're done.
4342  */
4343  if (n > 0)
4344  break;
4345 
4346  /*
4347  * Advance search position. Normally we start the next search at the
4348  * end of the previous match; but if the match was of zero length, we
4349  * have to advance by one character, or we'd just find the same match
4350  * again.
4351  */
4352  search_start = data_pos;
4353  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4354  search_start++;
4355  }
4356 
4357  /*
4358  * Copy the text to the right of the last match.
4359  */
4360  if (data_pos < data_len)
4361  {
4362  int chunk_len;
4363 
4364  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4365  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4366  }
4367 
4368  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4369  pfree(buf.data);
4370  pfree(data);
4371 
4372  return ret_text;
4373 }
#define lengthof(array)
Definition: c.h:767
int errcode(int sqlerrcode)
Definition: elog.c:853
int errmsg(const char *fmt,...)
Definition: elog.c:1070
#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:72
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:230
void initStringInfo(StringInfo str)
Definition: stringinfo.c:56
Definition: c.h:666
#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:3995
static int check_replace_text_has_escape(const text *replace_text)
Definition: varlena.c:4086
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:4119
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:4009

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 3559 of file varlena.c.

3561 {
3562  char *nextp = rawstring;
3563  bool done = false;
3564 
3565  *namelist = NIL;
3566 
3567  while (scanner_isspace(*nextp))
3568  nextp++; /* skip leading whitespace */
3569 
3570  if (*nextp == '\0')
3571  return true; /* allow empty string */
3572 
3573  /* At the top of the loop, we are at start of a new directory. */
3574  do
3575  {
3576  char *curname;
3577  char *endp;
3578 
3579  if (*nextp == '"')
3580  {
3581  /* Quoted name --- collapse quote-quote pairs */
3582  curname = nextp + 1;
3583  for (;;)
3584  {
3585  endp = strchr(nextp + 1, '"');
3586  if (endp == NULL)
3587  return false; /* mismatched quotes */
3588  if (endp[1] != '"')
3589  break; /* found end of quoted name */
3590  /* Collapse adjacent quotes into one quote, and look again */
3591  memmove(endp, endp + 1, strlen(endp));
3592  nextp = endp;
3593  }
3594  /* endp now points at the terminating quote */
3595  nextp = endp + 1;
3596  }
3597  else
3598  {
3599  /* Unquoted name --- extends to separator or end of string */
3600  curname = endp = nextp;
3601  while (*nextp && *nextp != separator)
3602  {
3603  /* trailing whitespace should not be included in name */
3604  if (!scanner_isspace(*nextp))
3605  endp = nextp + 1;
3606  nextp++;
3607  }
3608  if (curname == endp)
3609  return false; /* empty unquoted name not allowed */
3610  }
3611 
3612  while (scanner_isspace(*nextp))
3613  nextp++; /* skip trailing whitespace */
3614 
3615  if (*nextp == separator)
3616  {
3617  nextp++;
3618  while (scanner_isspace(*nextp))
3619  nextp++; /* skip leading whitespace for next */
3620  /* we expect another name, so done remains false */
3621  }
3622  else if (*nextp == '\0')
3623  done = true;
3624  else
3625  return false; /* invalid syntax */
3626 
3627  /* Now safe to overwrite separator with a null */
3628  *endp = '\0';
3629 
3630  /* Truncate path if it's overlength */
3631  if (strlen(curname) >= MAXPGPATH)
3632  curname[MAXPGPATH - 1] = '\0';
3633 
3634  /*
3635  * Finished isolating current name --- add it to list
3636  */
3637  curname = pstrdup(curname);
3638  canonicalize_path(curname);
3639  *namelist = lappend(*namelist, curname);
3640 
3641  /* Loop back if we didn't reach end of string */
3642  } while (!done);
3643 
3644  return true;
3645 }
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:265
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 3680 of file varlena.c.

3682 {
3683  char *nextp = rawstring;
3684  bool done = false;
3685 
3686  *namelist = NIL;
3687 
3688  while (scanner_isspace(*nextp))
3689  nextp++; /* skip leading whitespace */
3690 
3691  if (*nextp == '\0')
3692  return true; /* allow empty string */
3693 
3694  /* At the top of the loop, we are at start of a new identifier. */
3695  do
3696  {
3697  char *curname;
3698  char *endp;
3699 
3700  if (*nextp == '"')
3701  {
3702  /* Quoted name --- collapse quote-quote pairs */
3703  curname = nextp + 1;
3704  for (;;)
3705  {
3706  endp = strchr(nextp + 1, '"');
3707  if (endp == NULL)
3708  return false; /* mismatched quotes */
3709  if (endp[1] != '"')
3710  break; /* found end of quoted name */
3711  /* Collapse adjacent quotes into one quote, and look again */
3712  memmove(endp, endp + 1, strlen(endp));
3713  nextp = endp;
3714  }
3715  /* endp now points at the terminating quote */
3716  nextp = endp + 1;
3717  }
3718  else
3719  {
3720  /* Unquoted name --- extends to separator or whitespace */
3721  curname = nextp;
3722  while (*nextp && *nextp != separator &&
3723  !scanner_isspace(*nextp))
3724  nextp++;
3725  endp = nextp;
3726  if (curname == nextp)
3727  return false; /* empty unquoted name not allowed */
3728  }
3729 
3730  while (scanner_isspace(*nextp))
3731  nextp++; /* skip trailing whitespace */
3732 
3733  if (*nextp == separator)
3734  {
3735  nextp++;
3736  while (scanner_isspace(*nextp))
3737  nextp++; /* skip leading whitespace for next */
3738  /* we expect another name, so done remains false */
3739  }
3740  else if (*nextp == '\0')
3741  done = true;
3742  else
3743  return false; /* invalid syntax */
3744 
3745  /* Now safe to overwrite separator with a null */
3746  *endp = '\0';
3747 
3748  /*
3749  * Finished isolating current name --- add it to list
3750  */
3751  *namelist = lappend(*namelist, curname);
3752 
3753  /* Loop back if we didn't reach end of string */
3754  } while (!done);
3755 
3756  return true;
3757 }

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 3432 of file varlena.c.

3434 {
3435  char *nextp = rawstring;
3436  bool done = false;
3437 
3438  *namelist = NIL;
3439 
3440  while (scanner_isspace(*nextp))
3441  nextp++; /* skip leading whitespace */
3442 
3443  if (*nextp == '\0')
3444  return true; /* allow empty string */
3445 
3446  /* At the top of the loop, we are at start of a new identifier. */
3447  do
3448  {
3449  char *curname;
3450  char *endp;
3451 
3452  if (*nextp == '"')
3453  {
3454  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3455  curname = nextp + 1;
3456  for (;;)
3457  {
3458  endp = strchr(nextp + 1, '"');
3459  if (endp == NULL)
3460  return false; /* mismatched quotes */
3461  if (endp[1] != '"')
3462  break; /* found end of quoted name */
3463  /* Collapse adjacent quotes into one quote, and look again */
3464  memmove(endp, endp + 1, strlen(endp));
3465  nextp = endp;
3466  }
3467  /* endp now points at the terminating quote */
3468  nextp = endp + 1;
3469  }
3470  else
3471  {
3472  /* Unquoted name --- extends to separator or whitespace */
3473  char *downname;
3474  int len;
3475 
3476  curname = nextp;
3477  while (*nextp && *nextp != separator &&
3478  !scanner_isspace(*nextp))
3479  nextp++;
3480  endp = nextp;
3481  if (curname == nextp)
3482  return false; /* empty unquoted name not allowed */
3483 
3484  /*
3485  * Downcase the identifier, using same code as main lexer does.
3486  *
3487  * XXX because we want to overwrite the input in-place, we cannot
3488  * support a downcasing transformation that increases the string
3489  * length. This is not a problem given the current implementation
3490  * of downcase_truncate_identifier, but we'll probably have to do
3491  * something about this someday.
3492  */
3493  len = endp - curname;
3494  downname = downcase_truncate_identifier(curname, len, false);
3495  Assert(strlen(downname) <= len);
3496  strncpy(curname, downname, len); /* strncpy is required here */
3497  pfree(downname);
3498  }
3499 
3500  while (scanner_isspace(*nextp))
3501  nextp++; /* skip trailing whitespace */
3502 
3503  if (*nextp == separator)
3504  {
3505  nextp++;
3506  while (scanner_isspace(*nextp))
3507  nextp++; /* skip leading whitespace for next */
3508  /* we expect another name, so done remains false */
3509  }
3510  else if (*nextp == '\0')
3511  done = true;
3512  else
3513  return false; /* invalid syntax */
3514 
3515  /* Now safe to overwrite separator with a null */
3516  *endp = '\0';
3517 
3518  /* Truncate name if it's overlength */
3519  truncate_identifier(curname, strlen(curname), false);
3520 
3521  /*
3522  * Finished isolating current name --- add it to list
3523  */
3524  *namelist = lappend(*namelist, curname);
3525 
3526  /* Loop back if we didn't reach end of string */
3527  } while (!done);
3528 
3529  return true;
3530 }
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_restrict_nonsystem_relation_kind(), 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 3374 of file varlena.c.

3375 {
3376  char *rawname;
3377  List *result = NIL;
3378  List *namelist;
3379  ListCell *l;
3380 
3381  /* Convert to C string (handles possible detoasting). */
3382  /* Note we rely on being able to modify rawname below. */
3383  rawname = text_to_cstring(textval);
3384 
3385  if (!SplitIdentifierString(rawname, '.', &namelist))
3386  ereport(ERROR,
3387  (errcode(ERRCODE_INVALID_NAME),
3388  errmsg("invalid name syntax")));
3389 
3390  if (namelist == NIL)
3391  ereport(ERROR,
3392  (errcode(ERRCODE_INVALID_NAME),
3393  errmsg("invalid name syntax")));
3394 
3395  foreach(l, namelist)
3396  {
3397  char *curname = (char *) lfirst(l);
3398 
3399  result = lappend(result, makeString(pstrdup(curname)));
3400  }
3401 
3402  pfree(rawname);
3403  list_free(namelist);
3404 
3405  return result;
3406 }
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:3432

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 6221 of file varlena.c.

6222 {
6223  int dist;
6224 
6225  Assert(state);
6226 
6227  if (state->source == NULL || state->source[0] == '\0' ||
6228  candidate == NULL || candidate[0] == '\0')
6229  return;
6230 
6231  /*
6232  * To avoid ERROR-ing, we check the lengths here instead of setting
6233  * 'trusted' to false in the call to varstr_levenshtein_less_equal().
6234  */
6235  if (strlen(state->source) > MAX_LEVENSHTEIN_STRLEN ||
6236  strlen(candidate) > MAX_LEVENSHTEIN_STRLEN)
6237  return;
6238 
6239  dist = varstr_levenshtein_less_equal(state->source, strlen(state->source),
6240  candidate, strlen(candidate), 1, 1, 1,
6241  state->max_d, true);
6242  if (dist <= state->max_d &&
6243  dist <= strlen(state->source) / 2 &&
6244  (state->min_d == -1 || dist < state->min_d))
6245  {
6246  state->min_d = dist;
6247  state->match = candidate;
6248  }
6249 }
#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 1538 of file varlena.c.

1539 {
1540  int result;
1541  pg_locale_t mylocale;
1542 
1544 
1545  mylocale = pg_newlocale_from_collation(collid);
1546 
1547  if (mylocale->collate_is_c)
1548  {
1549  result = memcmp(arg1, arg2, Min(len1, len2));
1550  if ((result == 0) && (len1 != len2))
1551  result = (len1 < len2) ? -1 : 1;
1552  }
1553  else
1554  {
1555  /*
1556  * memcmp() can't tell us which of two unequal strings sorts first,
1557  * but it's a cheap way to tell if they're equal. Testing shows that
1558  * memcmp() followed by strcoll() is only trivially slower than
1559  * strcoll() by itself, so we don't lose much if this doesn't work out
1560  * very often, and if it does - for example, because there are many
1561  * equal strings in the input - then we win big by avoiding expensive
1562  * collation-aware comparisons.
1563  */
1564  if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1565  return 0;
1566 
1567  result = pg_strncoll(arg1, len1, arg2, len2, mylocale);
1568 
1569  /* Break tie if necessary. */
1570  if (result == 0 && mylocale->deterministic)
1571  {
1572  result = memcmp(arg1, arg2, Min(len1, len2));
1573  if ((result == 0) && (len1 != len2))
1574  result = (len1 < len2) ? -1 : 1;
1575  }
1576  }
1577 
1578  return result;
1579 }
#define Min(x, y)
Definition: c.h:983
Oid collid
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1432
int pg_strncoll(const char *arg1, ssize_t len1, const char *arg2, ssize_t len2, pg_locale_t locale)
Definition: pg_locale.c:1621
bool deterministic
Definition: pg_locale.h:82
static void check_collation_set(Oid collid)
Definition: varlena.c:1509

References check_collation_set(), pg_locale_struct::collate_is_c, collid, pg_locale_struct::deterministic, Min, 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:71
int x
Definition: isn.c:70
int j
Definition: isn.c:73
int i
Definition: isn.c:72
#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:6165

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 1856 of file varlena.c.

1857 {
1858  bool abbreviate = ssup->abbreviate;
1859  bool collate_c = false;
1860  VarStringSortSupport *sss;
1862 
1864 
1866 
1867  /*
1868  * If possible, set ssup->comparator to a function which can be used to
1869  * directly compare two datums. If we can do this, we'll avoid the
1870  * overhead of a trip through the fmgr layer for every comparison, which
1871  * can be substantial.
1872  *
1873  * Most typically, we'll set the comparator to varlenafastcmp_locale,
1874  * which uses strcoll() to perform comparisons. We use that for the
1875  * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1876  * LC_COLLATE = C, we can make things quite a bit faster with
1877  * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1878  * memcmp() rather than strcoll().
1879  */
1880  if (locale->collate_is_c)
1881  {
1882  if (typid == BPCHAROID)
1883  ssup->comparator = bpcharfastcmp_c;
1884  else if (typid == NAMEOID)
1885  {
1886  ssup->comparator = namefastcmp_c;
1887  /* Not supporting abbreviation with type NAME, for now */
1888  abbreviate = false;
1889  }
1890  else
1891  ssup->comparator = varstrfastcmp_c;
1892 
1893  collate_c = true;
1894  }
1895  else
1896  {
1897  /*
1898  * We use varlenafastcmp_locale except for type NAME.
1899  */
1900  if (typid == NAMEOID)
1901  {
1903  /* Not supporting abbreviation with type NAME, for now */
1904  abbreviate = false;
1905  }
1906  else
1908 
1909  /*
1910  * Unfortunately, it seems that abbreviation for non-C collations is
1911  * broken on many common platforms; see pg_strxfrm_enabled().
1912  *
1913  * Even apart from the risk of broken locales, it's possible that
1914  * there are platforms where the use of abbreviated keys should be
1915  * disabled at compile time. Having only 4 byte datums could make
1916  * worst-case performance drastically more likely, for example.
1917  * Moreover, macOS's strxfrm() implementation is known to not
1918  * effectively concentrate a significant amount of entropy from the
1919  * original string in earlier transformed blobs. It's possible that
1920  * other supported platforms are similarly encumbered. So, if we ever
1921  * get past disabling this categorically, we may still want or need to
1922  * disable it for particular platforms.
1923  */
1924  if (!pg_strxfrm_enabled(locale))
1925  abbreviate = false;
1926  }
1927 
1928  /*
1929  * If we're using abbreviated keys, or if we're using a locale-aware
1930  * comparison, we need to initialize a VarStringSortSupport object. Both
1931  * cases will make use of the temporary buffers we initialize here for
1932  * scratch space (and to detect requirement for BpChar semantics from
1933  * caller), and the abbreviation case requires additional state.
1934  */
1935  if (abbreviate || !collate_c)
1936  {
1937  sss = palloc(sizeof(VarStringSortSupport));
1938  sss->buf1 = palloc(TEXTBUFLEN);
1939  sss->buflen1 = TEXTBUFLEN;
1940  sss->buf2 = palloc(TEXTBUFLEN);
1941  sss->buflen2 = TEXTBUFLEN;
1942  /* Start with invalid values */
1943  sss->last_len1 = -1;
1944  sss->last_len2 = -1;
1945  /* Initialize */
1946  sss->last_returned = 0;
1947  if (collate_c)
1948  sss->locale = NULL;
1949  else
1950  sss->locale = locale;
1951 
1952  /*
1953  * To avoid somehow confusing a strxfrm() blob and an original string,
1954  * constantly keep track of the variety of data that buf1 and buf2
1955  * currently contain.
1956  *
1957  * Comparisons may be interleaved with conversion calls. Frequently,
1958  * conversions and comparisons are batched into two distinct phases,
1959  * but the correctness of caching cannot hinge upon this. For
1960  * comparison caching, buffer state is only trusted if cache_blob is
1961  * found set to false, whereas strxfrm() caching only trusts the state
1962  * when cache_blob is found set to true.
1963  *
1964  * Arbitrarily initialize cache_blob to true.
1965  */
1966  sss->cache_blob = true;
1967  sss->collate_c = collate_c;
1968  sss->typid = typid;
1969  ssup->ssup_extra = sss;
1970 
1971  /*
1972  * If possible, plan to use the abbreviated keys optimization. The
1973  * core code may switch back to authoritative comparator should
1974  * abbreviation be aborted.
1975  */
1976  if (abbreviate)
1977  {
1978  sss->prop_card = 0.20;
1979  initHyperLogLog(&sss->abbr_card, 10);
1980  initHyperLogLog(&sss->full_card, 10);
1981  ssup->abbrev_full_comparator = ssup->comparator;
1985  }
1986  }
1987 }
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:1654
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:3139
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2418
static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2075
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2030
static int namefastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2063
static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2106
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2220
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:1993
#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, 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().