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.

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, void *regexp, text *replace_text, bool glob)
 

Function Documentation

◆ replace_text_regexp()

text* replace_text_regexp ( text src_text,
void *  regexp,
text replace_text,
bool  glob 
)

Definition at line 4404 of file varlena.c.

References appendBinaryStringInfo(), appendStringInfoRegexpSubstr(), appendStringInfoText(), buf, charlen_to_bytelen(), CHECK_FOR_INTERRUPTS, check_replace_text_has_escape_char(), cstring_to_text_with_len(), StringInfoData::data, ereport, errcode(), errmsg(), ERROR, initStringInfo(), StringInfoData::len, palloc(), pfree(), pg_mb2wchar_with_len(), pg_regerror(), pg_regexec(), REG_NOMATCH, REG_OKAY, REGEXP_REPLACE_BACKREF_CNT, regmatch_t::rm_eo, regmatch_t::rm_so, VARDATA_ANY, VARSIZE_ANY, and VARSIZE_ANY_EXHDR.

Referenced by textregexreplace(), and textregexreplace_noopt().

4406 {
4407  text *ret_text;
4408  regex_t *re = (regex_t *) regexp;
4409  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
4412  pg_wchar *data;
4413  size_t data_len;
4414  int search_start;
4415  int data_pos;
4416  char *start_ptr;
4417  bool have_escape;
4418 
4419  initStringInfo(&buf);
4420 
4421  /* Convert data string to wide characters. */
4422  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
4423  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
4424 
4425  /* Check whether replace_text has escape char. */
4426  have_escape = check_replace_text_has_escape_char(replace_text);
4427 
4428  /* start_ptr points to the data_pos'th character of src_text */
4429  start_ptr = (char *) VARDATA_ANY(src_text);
4430  data_pos = 0;
4431 
4432  search_start = 0;
4433  while (search_start <= data_len)
4434  {
4435  int regexec_result;
4436 
4438 
4439  regexec_result = pg_regexec(re,
4440  data,
4441  data_len,
4442  search_start,
4443  NULL, /* no details */
4445  pmatch,
4446  0);
4447 
4448  if (regexec_result == REG_NOMATCH)
4449  break;
4450 
4451  if (regexec_result != REG_OKAY)
4452  {
4453  char errMsg[100];
4454 
4456  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
4457  ereport(ERROR,
4458  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4459  errmsg("regular expression failed: %s", errMsg)));
4460  }
4461 
4462  /*
4463  * Copy the text to the left of the match position. Note we are given
4464  * character not byte indexes.
4465  */
4466  if (pmatch[0].rm_so - data_pos > 0)
4467  {
4468  int chunk_len;
4469 
4470  chunk_len = charlen_to_bytelen(start_ptr,
4471  pmatch[0].rm_so - data_pos);
4472  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4473 
4474  /*
4475  * Advance start_ptr over that text, to avoid multiple rescans of
4476  * it if the replace_text contains multiple back-references.
4477  */
4478  start_ptr += chunk_len;
4479  data_pos = pmatch[0].rm_so;
4480  }
4481 
4482  /*
4483  * Copy the replace_text. Process back references when the
4484  * replace_text has escape characters.
4485  */
4486  if (have_escape)
4487  appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
4488  start_ptr, data_pos);
4489  else
4490  appendStringInfoText(&buf, replace_text);
4491 
4492  /* Advance start_ptr and data_pos over the matched text. */
4493  start_ptr += charlen_to_bytelen(start_ptr,
4494  pmatch[0].rm_eo - data_pos);
4495  data_pos = pmatch[0].rm_eo;
4496 
4497  /*
4498  * When global option is off, replace the first instance only.
4499  */
4500  if (!glob)
4501  break;
4502 
4503  /*
4504  * Advance search position. Normally we start the next search at the
4505  * end of the previous match; but if the match was of zero length, we
4506  * have to advance by one character, or we'd just find the same match
4507  * again.
4508  */
4509  search_start = data_pos;
4510  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4511  search_start++;
4512  }
4513 
4514  /*
4515  * Copy the text to the right of the last match.
4516  */
4517  if (data_pos < data_len)
4518  {
4519  int chunk_len;
4520 
4521  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4522  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4523  }
4524 
4525  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4526  pfree(buf.data);
4527  pfree(data);
4528 
4529  return ret_text;
4530 }
#define VARDATA_ANY(PTR)
Definition: postgres.h:348
regoff_t rm_so
Definition: regex.h:85
int errcode(int sqlerrcode)
Definition: elog.c:608
regoff_t rm_eo
Definition: regex.h:86
void pfree(void *pointer)
Definition: mcxt.c:1056
#define REG_OKAY
Definition: regex.h:137
#define ERROR
Definition: elog.h:43
static bool check_replace_text_has_escape_char(const text *replace_text)
Definition: varlena.c:4266
static int charlen_to_bytelen(const char *p, int n)
Definition: varlena.c:757
static void appendStringInfoText(StringInfo str, const text *t)
Definition: varlena.c:4177
static char * buf
Definition: pg_test_fsync.c:67
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:183
size_t pg_regerror(int errcode, const regex_t *preg, char *errbuf, size_t errbuf_size)
Definition: regerror.c:60
#define ereport(elevel, rest)
Definition: elog.h:141
unsigned int pg_wchar
Definition: mbprint.c:31
void initStringInfo(StringInfo str)
Definition: stringinfo.c:59
#define REGEXP_REPLACE_BACKREF_CNT
Definition: varlena.c:4393
#define VARSIZE_ANY(PTR)
Definition: postgres.h:335
int pg_mb2wchar_with_len(const char *from, pg_wchar *to, int len)
Definition: mbutils.c:765
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:4299
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
#define VARSIZE_ANY_EXHDR(PTR)
Definition: postgres.h:341
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:822
#define REG_NOMATCH
Definition: regex.h:138
Definition: c.h:550
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:99
void appendBinaryStringInfo(StringInfo str, const char *data, int datalen)
Definition: stringinfo.c:227
Definition: regex.h:55

◆ SplitDirectoriesString()

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

Definition at line 3779 of file varlena.c.

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

Referenced by load_libraries(), and PostmasterMain().

3781 {
3782  char *nextp = rawstring;
3783  bool done = false;
3784 
3785  *namelist = NIL;
3786 
3787  while (scanner_isspace(*nextp))
3788  nextp++; /* skip leading whitespace */
3789 
3790  if (*nextp == '\0')
3791  return true; /* allow empty string */
3792 
3793  /* At the top of the loop, we are at start of a new directory. */
3794  do
3795  {
3796  char *curname;
3797  char *endp;
3798 
3799  if (*nextp == '"')
3800  {
3801  /* Quoted name --- collapse quote-quote pairs */
3802  curname = nextp + 1;
3803  for (;;)
3804  {
3805  endp = strchr(nextp + 1, '"');
3806  if (endp == NULL)
3807  return false; /* mismatched quotes */
3808  if (endp[1] != '"')
3809  break; /* found end of quoted name */
3810  /* Collapse adjacent quotes into one quote, and look again */
3811  memmove(endp, endp + 1, strlen(endp));
3812  nextp = endp;
3813  }
3814  /* endp now points at the terminating quote */
3815  nextp = endp + 1;
3816  }
3817  else
3818  {
3819  /* Unquoted name --- extends to separator or end of string */
3820  curname = endp = nextp;
3821  while (*nextp && *nextp != separator)
3822  {
3823  /* trailing whitespace should not be included in name */
3824  if (!scanner_isspace(*nextp))
3825  endp = nextp + 1;
3826  nextp++;
3827  }
3828  if (curname == endp)
3829  return false; /* empty unquoted name not allowed */
3830  }
3831 
3832  while (scanner_isspace(*nextp))
3833  nextp++; /* skip trailing whitespace */
3834 
3835  if (*nextp == separator)
3836  {
3837  nextp++;
3838  while (scanner_isspace(*nextp))
3839  nextp++; /* skip leading whitespace for next */
3840  /* we expect another name, so done remains false */
3841  }
3842  else if (*nextp == '\0')
3843  done = true;
3844  else
3845  return false; /* invalid syntax */
3846 
3847  /* Now safe to overwrite separator with a null */
3848  *endp = '\0';
3849 
3850  /* Truncate path if it's overlength */
3851  if (strlen(curname) >= MAXPGPATH)
3852  curname[MAXPGPATH - 1] = '\0';
3853 
3854  /*
3855  * Finished isolating current name --- add it to list
3856  */
3857  curname = pstrdup(curname);
3858  canonicalize_path(curname);
3859  *namelist = lappend(*namelist, curname);
3860 
3861  /* Loop back if we didn't reach end of string */
3862  } while (!done);
3863 
3864  return true;
3865 }
#define NIL
Definition: pg_list.h:65
char * pstrdup(const char *in)
Definition: mcxt.c:1186
void canonicalize_path(char *path)
Definition: path.c:254
#define MAXPGPATH
#define memmove(d, s, c)
Definition: c.h:1261
List * lappend(List *list, void *datum)
Definition: list.c:322
bool scanner_isspace(char ch)
Definition: scansup.c:220

◆ SplitGUCList()

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

Definition at line 3900 of file varlena.c.

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

Referenced by dumpFunc(), and pg_get_functiondef().

3902 {
3903  char *nextp = rawstring;
3904  bool done = false;
3905 
3906  *namelist = NIL;
3907 
3908  while (scanner_isspace(*nextp))
3909  nextp++; /* skip leading whitespace */
3910 
3911  if (*nextp == '\0')
3912  return true; /* allow empty string */
3913 
3914  /* At the top of the loop, we are at start of a new identifier. */
3915  do
3916  {
3917  char *curname;
3918  char *endp;
3919 
3920  if (*nextp == '"')
3921  {
3922  /* Quoted name --- collapse quote-quote pairs */
3923  curname = nextp + 1;
3924  for (;;)
3925  {
3926  endp = strchr(nextp + 1, '"');
3927  if (endp == NULL)
3928  return false; /* mismatched quotes */
3929  if (endp[1] != '"')
3930  break; /* found end of quoted name */
3931  /* Collapse adjacent quotes into one quote, and look again */
3932  memmove(endp, endp + 1, strlen(endp));
3933  nextp = endp;
3934  }
3935  /* endp now points at the terminating quote */
3936  nextp = endp + 1;
3937  }
3938  else
3939  {
3940  /* Unquoted name --- extends to separator or whitespace */
3941  curname = nextp;
3942  while (*nextp && *nextp != separator &&
3943  !scanner_isspace(*nextp))
3944  nextp++;
3945  endp = nextp;
3946  if (curname == nextp)
3947  return false; /* empty unquoted name not allowed */
3948  }
3949 
3950  while (scanner_isspace(*nextp))
3951  nextp++; /* skip trailing whitespace */
3952 
3953  if (*nextp == separator)
3954  {
3955  nextp++;
3956  while (scanner_isspace(*nextp))
3957  nextp++; /* skip leading whitespace for next */
3958  /* we expect another name, so done remains false */
3959  }
3960  else if (*nextp == '\0')
3961  done = true;
3962  else
3963  return false; /* invalid syntax */
3964 
3965  /* Now safe to overwrite separator with a null */
3966  *endp = '\0';
3967 
3968  /*
3969  * Finished isolating current name --- add it to list
3970  */
3971  *namelist = lappend(*namelist, curname);
3972 
3973  /* Loop back if we didn't reach end of string */
3974  } while (!done);
3975 
3976  return true;
3977 }
#define NIL
Definition: pg_list.h:65
#define memmove(d, s, c)
Definition: c.h:1261
List * lappend(List *list, void *datum)
Definition: list.c:322
bool scanner_isspace(char ch)
Definition: scansup.c:220

◆ SplitIdentifierString()

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

Definition at line 3652 of file varlena.c.

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

Referenced by check_datestyle(), check_log_destination(), check_search_path(), check_temp_tablespaces(), check_wal_consistency_checking(), ExtractExtensionList(), parse_extension_control_file(), parse_hba_auth_opt(), parse_output_parameters(), parse_publication_options(), plpgsql_extra_checks_check_hook(), PostmasterMain(), PrepareTempTablespaces(), recomputeNamespacePath(), stringToQualifiedNameList(), and textToQualifiedNameList().

3654 {
3655  char *nextp = rawstring;
3656  bool done = false;
3657 
3658  *namelist = NIL;
3659 
3660  while (scanner_isspace(*nextp))
3661  nextp++; /* skip leading whitespace */
3662 
3663  if (*nextp == '\0')
3664  return true; /* allow empty string */
3665 
3666  /* At the top of the loop, we are at start of a new identifier. */
3667  do
3668  {
3669  char *curname;
3670  char *endp;
3671 
3672  if (*nextp == '"')
3673  {
3674  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3675  curname = nextp + 1;
3676  for (;;)
3677  {
3678  endp = strchr(nextp + 1, '"');
3679  if (endp == NULL)
3680  return false; /* mismatched quotes */
3681  if (endp[1] != '"')
3682  break; /* found end of quoted name */
3683  /* Collapse adjacent quotes into one quote, and look again */
3684  memmove(endp, endp + 1, strlen(endp));
3685  nextp = endp;
3686  }
3687  /* endp now points at the terminating quote */
3688  nextp = endp + 1;
3689  }
3690  else
3691  {
3692  /* Unquoted name --- extends to separator or whitespace */
3693  char *downname;
3694  int len;
3695 
3696  curname = nextp;
3697  while (*nextp && *nextp != separator &&
3698  !scanner_isspace(*nextp))
3699  nextp++;
3700  endp = nextp;
3701  if (curname == nextp)
3702  return false; /* empty unquoted name not allowed */
3703 
3704  /*
3705  * Downcase the identifier, using same code as main lexer does.
3706  *
3707  * XXX because we want to overwrite the input in-place, we cannot
3708  * support a downcasing transformation that increases the string
3709  * length. This is not a problem given the current implementation
3710  * of downcase_truncate_identifier, but we'll probably have to do
3711  * something about this someday.
3712  */
3713  len = endp - curname;
3714  downname = downcase_truncate_identifier(curname, len, false);
3715  Assert(strlen(downname) <= len);
3716  strncpy(curname, downname, len); /* strncpy is required here */
3717  pfree(downname);
3718  }
3719 
3720  while (scanner_isspace(*nextp))
3721  nextp++; /* skip trailing whitespace */
3722 
3723  if (*nextp == separator)
3724  {
3725  nextp++;
3726  while (scanner_isspace(*nextp))
3727  nextp++; /* skip leading whitespace for next */
3728  /* we expect another name, so done remains false */
3729  }
3730  else if (*nextp == '\0')
3731  done = true;
3732  else
3733  return false; /* invalid syntax */
3734 
3735  /* Now safe to overwrite separator with a null */
3736  *endp = '\0';
3737 
3738  /* Truncate name if it's overlength */
3739  truncate_identifier(curname, strlen(curname), false);
3740 
3741  /*
3742  * Finished isolating current name --- add it to list
3743  */
3744  *namelist = lappend(*namelist, curname);
3745 
3746  /* Loop back if we didn't reach end of string */
3747  } while (!done);
3748 
3749  return true;
3750 }
#define NIL
Definition: pg_list.h:65
char * downcase_truncate_identifier(const char *ident, int len, bool warn)
Definition: scansup.c:130
void truncate_identifier(char *ident, int len, bool warn)
Definition: scansup.c:186
void pfree(void *pointer)
Definition: mcxt.c:1056
#define memmove(d, s, c)
Definition: c.h:1261
List * lappend(List *list, void *datum)
Definition: list.c:322
bool scanner_isspace(char ch)
Definition: scansup.c:220
#define Assert(condition)
Definition: c.h:733

◆ textToQualifiedNameList()

List* textToQualifiedNameList ( text textval)

Definition at line 3594 of file varlena.c.

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

Referenced by bt_metap(), bt_page_items(), bt_page_stats(), 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().

3595 {
3596  char *rawname;
3597  List *result = NIL;
3598  List *namelist;
3599  ListCell *l;
3600 
3601  /* Convert to C string (handles possible detoasting). */
3602  /* Note we rely on being able to modify rawname below. */
3603  rawname = text_to_cstring(textval);
3604 
3605  if (!SplitIdentifierString(rawname, '.', &namelist))
3606  ereport(ERROR,
3607  (errcode(ERRCODE_INVALID_NAME),
3608  errmsg("invalid name syntax")));
3609 
3610  if (namelist == NIL)
3611  ereport(ERROR,
3612  (errcode(ERRCODE_INVALID_NAME),
3613  errmsg("invalid name syntax")));
3614 
3615  foreach(l, namelist)
3616  {
3617  char *curname = (char *) lfirst(l);
3618 
3619  result = lappend(result, makeString(pstrdup(curname)));
3620  }
3621 
3622  pfree(rawname);
3623  list_free(namelist);
3624 
3625  return result;
3626 }
Value * makeString(char *str)
Definition: value.c:53
#define NIL
Definition: pg_list.h:65
char * pstrdup(const char *in)
Definition: mcxt.c:1186
int errcode(int sqlerrcode)
Definition: elog.c:608
void pfree(void *pointer)
Definition: mcxt.c:1056
#define ERROR
Definition: elog.h:43
bool SplitIdentifierString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3652
#define ereport(elevel, rest)
Definition: elog.h:141
List * lappend(List *list, void *datum)
Definition: list.c:322
#define lfirst(lc)
Definition: pg_list.h:190
char * text_to_cstring(const text *t)
Definition: varlena.c:204
int errmsg(const char *fmt,...)
Definition: elog.c:822
void list_free(List *list)
Definition: list.c:1377
Definition: pg_list.h:50

◆ varstr_cmp()

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

Definition at line 1482 of file varlena.c.

References check_collation_set(), pg_locale_struct::deterministic, elog, ereport, errmsg(), ERROR, GetDatabaseEncoding(), pg_locale_struct::info, lc_collate_is_c(), Min, palloc(), pfree(), pg_newlocale_from_collation(), PG_UTF8, pg_locale_struct::provider, status(), strcoll_l, TEXTBUFLEN, and wcscoll_l.

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().

1483 {
1484  int result;
1485 
1486  check_collation_set(collid);
1487 
1488  /*
1489  * Unfortunately, there is no strncoll(), so in the non-C locale case we
1490  * have to do some memory copying. This turns out to be significantly
1491  * slower, so we optimize the case where LC_COLLATE is C. We also try to
1492  * optimize relatively-short strings by avoiding palloc/pfree overhead.
1493  */
1494  if (lc_collate_is_c(collid))
1495  {
1496  result = memcmp(arg1, arg2, Min(len1, len2));
1497  if ((result == 0) && (len1 != len2))
1498  result = (len1 < len2) ? -1 : 1;
1499  }
1500  else
1501  {
1502  char a1buf[TEXTBUFLEN];
1503  char a2buf[TEXTBUFLEN];
1504  char *a1p,
1505  *a2p;
1506  pg_locale_t mylocale = 0;
1507 
1508  if (collid != DEFAULT_COLLATION_OID)
1509  mylocale = pg_newlocale_from_collation(collid);
1510 
1511  /*
1512  * memcmp() can't tell us which of two unequal strings sorts first,
1513  * but it's a cheap way to tell if they're equal. Testing shows that
1514  * memcmp() followed by strcoll() is only trivially slower than
1515  * strcoll() by itself, so we don't lose much if this doesn't work out
1516  * very often, and if it does - for example, because there are many
1517  * equal strings in the input - then we win big by avoiding expensive
1518  * collation-aware comparisons.
1519  */
1520  if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1521  return 0;
1522 
1523 #ifdef WIN32
1524  /* Win32 does not have UTF-8, so we need to map to UTF-16 */
1525  if (GetDatabaseEncoding() == PG_UTF8
1526  && (!mylocale || mylocale->provider == COLLPROVIDER_LIBC))
1527  {
1528  int a1len;
1529  int a2len;
1530  int r;
1531 
1532  if (len1 >= TEXTBUFLEN / 2)
1533  {
1534  a1len = len1 * 2 + 2;
1535  a1p = palloc(a1len);
1536  }
1537  else
1538  {
1539  a1len = TEXTBUFLEN;
1540  a1p = a1buf;
1541  }
1542  if (len2 >= TEXTBUFLEN / 2)
1543  {
1544  a2len = len2 * 2 + 2;
1545  a2p = palloc(a2len);
1546  }
1547  else
1548  {
1549  a2len = TEXTBUFLEN;
1550  a2p = a2buf;
1551  }
1552 
1553  /* stupid Microsloth API does not work for zero-length input */
1554  if (len1 == 0)
1555  r = 0;
1556  else
1557  {
1558  r = MultiByteToWideChar(CP_UTF8, 0, arg1, len1,
1559  (LPWSTR) a1p, a1len / 2);
1560  if (!r)
1561  ereport(ERROR,
1562  (errmsg("could not convert string to UTF-16: error code %lu",
1563  GetLastError())));
1564  }
1565  ((LPWSTR) a1p)[r] = 0;
1566 
1567  if (len2 == 0)
1568  r = 0;
1569  else
1570  {
1571  r = MultiByteToWideChar(CP_UTF8, 0, arg2, len2,
1572  (LPWSTR) a2p, a2len / 2);
1573  if (!r)
1574  ereport(ERROR,
1575  (errmsg("could not convert string to UTF-16: error code %lu",
1576  GetLastError())));
1577  }
1578  ((LPWSTR) a2p)[r] = 0;
1579 
1580  errno = 0;
1581 #ifdef HAVE_LOCALE_T
1582  if (mylocale)
1583  result = wcscoll_l((LPWSTR) a1p, (LPWSTR) a2p, mylocale->info.lt);
1584  else
1585 #endif
1586  result = wcscoll((LPWSTR) a1p, (LPWSTR) a2p);
1587  if (result == 2147483647) /* _NLSCMPERROR; missing from mingw
1588  * headers */
1589  ereport(ERROR,
1590  (errmsg("could not compare Unicode strings: %m")));
1591 
1592  /* Break tie if necessary. */
1593  if (result == 0 &&
1594  (!mylocale || mylocale->deterministic))
1595  {
1596  result = memcmp(arg1, arg2, Min(len1, len2));
1597  if ((result == 0) && (len1 != len2))
1598  result = (len1 < len2) ? -1 : 1;
1599  }
1600 
1601  if (a1p != a1buf)
1602  pfree(a1p);
1603  if (a2p != a2buf)
1604  pfree(a2p);
1605 
1606  return result;
1607  }
1608 #endif /* WIN32 */
1609 
1610  if (len1 >= TEXTBUFLEN)
1611  a1p = (char *) palloc(len1 + 1);
1612  else
1613  a1p = a1buf;
1614  if (len2 >= TEXTBUFLEN)
1615  a2p = (char *) palloc(len2 + 1);
1616  else
1617  a2p = a2buf;
1618 
1619  memcpy(a1p, arg1, len1);
1620  a1p[len1] = '\0';
1621  memcpy(a2p, arg2, len2);
1622  a2p[len2] = '\0';
1623 
1624  if (mylocale)
1625  {
1626  if (mylocale->provider == COLLPROVIDER_ICU)
1627  {
1628 #ifdef USE_ICU
1629 #ifdef HAVE_UCOL_STRCOLLUTF8
1630  if (GetDatabaseEncoding() == PG_UTF8)
1631  {
1632  UErrorCode status;
1633 
1634  status = U_ZERO_ERROR;
1635  result = ucol_strcollUTF8(mylocale->info.icu.ucol,
1636  arg1, len1,
1637  arg2, len2,
1638  &status);
1639  if (U_FAILURE(status))
1640  ereport(ERROR,
1641  (errmsg("collation failed: %s", u_errorName(status))));
1642  }
1643  else
1644 #endif
1645  {
1646  int32_t ulen1,
1647  ulen2;
1648  UChar *uchar1,
1649  *uchar2;
1650 
1651  ulen1 = icu_to_uchar(&uchar1, arg1, len1);
1652  ulen2 = icu_to_uchar(&uchar2, arg2, len2);
1653 
1654  result = ucol_strcoll(mylocale->info.icu.ucol,
1655  uchar1, ulen1,
1656  uchar2, ulen2);
1657 
1658  pfree(uchar1);
1659  pfree(uchar2);
1660  }
1661 #else /* not USE_ICU */
1662  /* shouldn't happen */
1663  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1664 #endif /* not USE_ICU */
1665  }
1666  else
1667  {
1668 #ifdef HAVE_LOCALE_T
1669  result = strcoll_l(a1p, a2p, mylocale->info.lt);
1670 #else
1671  /* shouldn't happen */
1672  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1673 #endif
1674  }
1675  }
1676  else
1677  result = strcoll(a1p, a2p);
1678 
1679  /* Break tie if necessary. */
1680  if (result == 0 &&
1681  (!mylocale || mylocale->deterministic))
1682  result = strcmp(a1p, a2p);
1683 
1684  if (a1p != a1buf)
1685  pfree(a1p);
1686  if (a2p != a2buf)
1687  pfree(a2p);
1688  }
1689 
1690  return result;
1691 }
union pg_locale_struct::@144 info
#define Min(x, y)
Definition: c.h:905
void pfree(void *pointer)
Definition: mcxt.c:1056
#define ERROR
Definition: elog.h:43
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1176
static void check_collation_set(Oid collid)
Definition: varlena.c:1453
#define strcoll_l
Definition: win32_port.h:386
#define ereport(elevel, rest)
Definition: elog.h:141
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1323
int GetDatabaseEncoding(void)
Definition: mbutils.c:1046
bool deterministic
Definition: pg_locale.h:85
#define wcscoll_l
Definition: win32_port.h:388
#define TEXTBUFLEN
Definition: varlena.c:98
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:822
#define elog(elevel,...)
Definition: elog.h:228
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:226

◆ 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.

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

Referenced by levenshtein(), and levenshtein_with_costs().

78 {
79  int m,
80  n;
81  int *prev;
82  int *curr;
83  int *s_char_len = NULL;
84  int i,
85  j;
86  const char *y;
87 
88  /*
89  * For varstr_levenshtein_less_equal, we have real variables called
90  * start_column and stop_column; otherwise it's just short-hand for 0 and
91  * m.
92  */
93 #ifdef LEVENSHTEIN_LESS_EQUAL
94  int start_column,
95  stop_column;
96 
97 #undef START_COLUMN
98 #undef STOP_COLUMN
99 #define START_COLUMN start_column
100 #define STOP_COLUMN stop_column
101 #else
102 #undef START_COLUMN
103 #undef STOP_COLUMN
104 #define START_COLUMN 0
105 #define STOP_COLUMN m
106 #endif
107 
108  /* Convert string lengths (in bytes) to lengths in characters */
109  m = pg_mbstrlen_with_len(source, slen);
110  n = pg_mbstrlen_with_len(target, tlen);
111 
112  /*
113  * We can transform an empty s into t with n insertions, or a non-empty t
114  * into an empty s with m deletions.
115  */
116  if (!m)
117  return n * ins_c;
118  if (!n)
119  return m * del_c;
120 
121  /*
122  * For security concerns, restrict excessive CPU+RAM usage. (This
123  * implementation uses O(m) memory and has O(mn) complexity.) If
124  * "trusted" is true, caller is responsible for not making excessive
125  * requests, typically by using a small max_d along with strings that are
126  * bounded, though not necessarily to MAX_LEVENSHTEIN_STRLEN exactly.
127  */
128  if (!trusted &&
129  (m > MAX_LEVENSHTEIN_STRLEN ||
131  ereport(ERROR,
132  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
133  errmsg("levenshtein argument exceeds maximum length of %d characters",
135 
136 #ifdef LEVENSHTEIN_LESS_EQUAL
137  /* Initialize start and stop columns. */
138  start_column = 0;
139  stop_column = m + 1;
140 
141  /*
142  * If max_d >= 0, determine whether the bound is impossibly tight. If so,
143  * return max_d + 1 immediately. Otherwise, determine whether it's tight
144  * enough to limit the computation we must perform. If so, figure out
145  * initial stop column.
146  */
147  if (max_d >= 0)
148  {
149  int min_theo_d; /* Theoretical minimum distance. */
150  int max_theo_d; /* Theoretical maximum distance. */
151  int net_inserts = n - m;
152 
153  min_theo_d = net_inserts < 0 ?
154  -net_inserts * del_c : net_inserts * ins_c;
155  if (min_theo_d > max_d)
156  return max_d + 1;
157  if (ins_c + del_c < sub_c)
158  sub_c = ins_c + del_c;
159  max_theo_d = min_theo_d + sub_c * Min(m, n);
160  if (max_d >= max_theo_d)
161  max_d = -1;
162  else if (ins_c + del_c > 0)
163  {
164  /*
165  * Figure out how much of the first row of the notional matrix we
166  * need to fill in. If the string is growing, the theoretical
167  * minimum distance already incorporates the cost of deleting the
168  * number of characters necessary to make the two strings equal in
169  * length. Each additional deletion forces another insertion, so
170  * the best-case total cost increases by ins_c + del_c. If the
171  * string is shrinking, the minimum theoretical cost assumes no
172  * excess deletions; that is, we're starting no further right than
173  * column n - m. If we do start further right, the best-case
174  * total cost increases by ins_c + del_c for each move right.
175  */
176  int slack_d = max_d - min_theo_d;
177  int best_column = net_inserts < 0 ? -net_inserts : 0;
178 
179  stop_column = best_column + (slack_d / (ins_c + del_c)) + 1;
180  if (stop_column > m)
181  stop_column = m + 1;
182  }
183  }
184 #endif
185 
186  /*
187  * In order to avoid calling pg_mblen() repeatedly on each character in s,
188  * we cache all the lengths before starting the main loop -- but if all
189  * the characters in both strings are single byte, then we skip this and
190  * use a fast-path in the main loop. If only one string contains
191  * multi-byte characters, we still build the array, so that the fast-path
192  * needn't deal with the case where the array hasn't been initialized.
193  */
194  if (m != slen || n != tlen)
195  {
196  int i;
197  const char *cp = source;
198 
199  s_char_len = (int *) palloc((m + 1) * sizeof(int));
200  for (i = 0; i < m; ++i)
201  {
202  s_char_len[i] = pg_mblen(cp);
203  cp += s_char_len[i];
204  }
205  s_char_len[i] = 0;
206  }
207 
208  /* One more cell for initialization column and row. */
209  ++m;
210  ++n;
211 
212  /* Previous and current rows of notional array. */
213  prev = (int *) palloc(2 * m * sizeof(int));
214  curr = prev + m;
215 
216  /*
217  * To transform the first i characters of s into the first 0 characters of
218  * t, we must perform i deletions.
219  */
220  for (i = START_COLUMN; i < STOP_COLUMN; i++)
221  prev[i] = i * del_c;
222 
223  /* Loop through rows of the notional array */
224  for (y = target, j = 1; j < n; j++)
225  {
226  int *temp;
227  const char *x = source;
228  int y_char_len = n != tlen + 1 ? pg_mblen(y) : 1;
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 }
#define Min(x, y)
Definition: c.h:905
int errcode(int sqlerrcode)
Definition: elog.c:608
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:836
#define STOP_COLUMN
#define ERROR
Definition: elog.h:43
#define START_COLUMN
#define ereport(elevel, rest)
Definition: elog.h:141
#define MAX_LEVENSHTEIN_STRLEN
Definition: levenshtein.c:26
static bool rest_of_char_same(const char *s1, const char *s2, int len)
Definition: varlena.c:5944
int pg_mblen(const char *mbstr)
Definition: mbutils.c:802
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:822
int i

◆ 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 1967 of file varlena.c.

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, SortSupportData::comparator, VarStringSortSupport::full_card, GetDatabaseEncoding(), 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_UTF8, VarStringSortSupport::prop_card, pg_locale_struct::provider, SortSupportData::ssup_extra, TEXTBUFLEN, VarStringSortSupport::typid, varlenafastcmp_locale(), varstr_abbrev_abort(), varstr_abbrev_convert(), varstrcmp_abbrev(), and varstrfastcmp_c().

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

1968 {
1969  bool abbreviate = ssup->abbreviate;
1970  bool collate_c = false;
1971  VarStringSortSupport *sss;
1972  pg_locale_t locale = 0;
1973 
1974  check_collation_set(collid);
1975 
1976  /*
1977  * If possible, set ssup->comparator to a function which can be used to
1978  * directly compare two datums. If we can do this, we'll avoid the
1979  * overhead of a trip through the fmgr layer for every comparison, which
1980  * can be substantial.
1981  *
1982  * Most typically, we'll set the comparator to varlenafastcmp_locale,
1983  * which uses strcoll() to perform comparisons. We use that for the
1984  * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1985  * LC_COLLATE = C, we can make things quite a bit faster with
1986  * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1987  * memcmp() rather than strcoll().
1988  */
1989  if (lc_collate_is_c(collid))
1990  {
1991  if (typid == BPCHAROID)
1992  ssup->comparator = bpcharfastcmp_c;
1993  else if (typid == NAMEOID)
1994  {
1995  ssup->comparator = namefastcmp_c;
1996  /* Not supporting abbreviation with type NAME, for now */
1997  abbreviate = false;
1998  }
1999  else
2000  ssup->comparator = varstrfastcmp_c;
2001 
2002  collate_c = true;
2003  }
2004  else
2005  {
2006  /*
2007  * We need a collation-sensitive comparison. To make things faster,
2008  * we'll figure out the collation based on the locale id and cache the
2009  * result.
2010  */
2011  if (collid != DEFAULT_COLLATION_OID)
2012  locale = pg_newlocale_from_collation(collid);
2013 
2014  /*
2015  * There is a further exception on Windows. When the database
2016  * encoding is UTF-8 and we are not using the C collation, complex
2017  * hacks are required. We don't currently have a comparator that
2018  * handles that case, so we fall back on the slow method of having the
2019  * sort code invoke bttextcmp() (in the case of text) via the fmgr
2020  * trampoline. ICU locales work just the same on Windows, however.
2021  */
2022 #ifdef WIN32
2023  if (GetDatabaseEncoding() == PG_UTF8 &&
2024  !(locale && locale->provider == COLLPROVIDER_ICU))
2025  return;
2026 #endif
2027 
2028  /*
2029  * We use varlenafastcmp_locale except for type NAME.
2030  */
2031  if (typid == NAMEOID)
2032  {
2034  /* Not supporting abbreviation with type NAME, for now */
2035  abbreviate = false;
2036  }
2037  else
2039  }
2040 
2041  /*
2042  * Unfortunately, it seems that abbreviation for non-C collations is
2043  * broken on many common platforms; testing of multiple versions of glibc
2044  * reveals that, for many locales, strcoll() and strxfrm() do not return
2045  * consistent results, which is fatal to this optimization. While no
2046  * other libc other than Cygwin has so far been shown to have a problem,
2047  * we take the conservative course of action for right now and disable
2048  * this categorically. (Users who are certain this isn't a problem on
2049  * their system can define TRUST_STRXFRM.)
2050  *
2051  * Even apart from the risk of broken locales, it's possible that there
2052  * are platforms where the use of abbreviated keys should be disabled at
2053  * compile time. Having only 4 byte datums could make worst-case
2054  * performance drastically more likely, for example. Moreover, macOS's
2055  * strxfrm() implementation is known to not effectively concentrate a
2056  * significant amount of entropy from the original string in earlier
2057  * transformed blobs. It's possible that other supported platforms are
2058  * similarly encumbered. So, if we ever get past disabling this
2059  * categorically, we may still want or need to disable it for particular
2060  * platforms.
2061  */
2062 #ifndef TRUST_STRXFRM
2063  if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
2064  abbreviate = false;
2065 #endif
2066 
2067  /*
2068  * If we're using abbreviated keys, or if we're using a locale-aware
2069  * comparison, we need to initialize a VarStringSortSupport object. Both
2070  * cases will make use of the temporary buffers we initialize here for
2071  * scratch space (and to detect requirement for BpChar semantics from
2072  * caller), and the abbreviation case requires additional state.
2073  */
2074  if (abbreviate || !collate_c)
2075  {
2076  sss = palloc(sizeof(VarStringSortSupport));
2077  sss->buf1 = palloc(TEXTBUFLEN);
2078  sss->buflen1 = TEXTBUFLEN;
2079  sss->buf2 = palloc(TEXTBUFLEN);
2080  sss->buflen2 = TEXTBUFLEN;
2081  /* Start with invalid values */
2082  sss->last_len1 = -1;
2083  sss->last_len2 = -1;
2084  /* Initialize */
2085  sss->last_returned = 0;
2086  sss->locale = locale;
2087 
2088  /*
2089  * To avoid somehow confusing a strxfrm() blob and an original string,
2090  * constantly keep track of the variety of data that buf1 and buf2
2091  * currently contain.
2092  *
2093  * Comparisons may be interleaved with conversion calls. Frequently,
2094  * conversions and comparisons are batched into two distinct phases,
2095  * but the correctness of caching cannot hinge upon this. For
2096  * comparison caching, buffer state is only trusted if cache_blob is
2097  * found set to false, whereas strxfrm() caching only trusts the state
2098  * when cache_blob is found set to true.
2099  *
2100  * Arbitrarily initialize cache_blob to true.
2101  */
2102  sss->cache_blob = true;
2103  sss->collate_c = collate_c;
2104  sss->typid = typid;
2105  ssup->ssup_extra = sss;
2106 
2107  /*
2108  * If possible, plan to use the abbreviated keys optimization. The
2109  * core code may switch back to authoritative comparator should
2110  * abbreviation be aborted.
2111  */
2112  if (abbreviate)
2113  {
2114  sss->prop_card = 0.20;
2115  initHyperLogLog(&sss->abbr_card, 10);
2116  initHyperLogLog(&sss->full_card, 10);
2117  ssup->abbrev_full_comparator = ssup->comparator;
2118  ssup->comparator = varstrcmp_abbrev;
2121  }
2122  }
2123 }
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2166
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2433
static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2408
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2129
static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2211
void initHyperLogLog(hyperLogLogState *cState, uint8 bwidth)
Definition: hyperloglog.c:65
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1176
static void check_collation_set(Oid collid)
Definition: varlena.c:1453
static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2242
int(* comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:106
hyperLogLogState abbr_card
Definition: varlena.c:88
pg_locale_t locale
Definition: varlena.c:91
int(* abbrev_full_comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:191
Datum(* abbrev_converter)(Datum original, SortSupport ssup)
Definition: sortsupport.h:172
void * ssup_extra
Definition: sortsupport.h:87
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1323
int GetDatabaseEncoding(void)
Definition: mbutils.c:1046
hyperLogLogState full_card
Definition: varlena.c:89
bool(* abbrev_abort)(int memtupcount, SortSupport ssup)
Definition: sortsupport.h:182
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2673
#define TEXTBUFLEN
Definition: varlena.c:98
void * palloc(Size size)
Definition: mcxt.c:949
static char * locale
Definition: initdb.c:125
static int namefastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2199