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 4505 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().

4507 {
4508  text *ret_text;
4509  regex_t *re = (regex_t *) regexp;
4510  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
4513  pg_wchar *data;
4514  size_t data_len;
4515  int search_start;
4516  int data_pos;
4517  char *start_ptr;
4518  bool have_escape;
4519 
4520  initStringInfo(&buf);
4521 
4522  /* Convert data string to wide characters. */
4523  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
4524  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
4525 
4526  /* Check whether replace_text has escape char. */
4527  have_escape = check_replace_text_has_escape_char(replace_text);
4528 
4529  /* start_ptr points to the data_pos'th character of src_text */
4530  start_ptr = (char *) VARDATA_ANY(src_text);
4531  data_pos = 0;
4532 
4533  search_start = 0;
4534  while (search_start <= data_len)
4535  {
4536  int regexec_result;
4537 
4539 
4540  regexec_result = pg_regexec(re,
4541  data,
4542  data_len,
4543  search_start,
4544  NULL, /* no details */
4546  pmatch,
4547  0);
4548 
4549  if (regexec_result == REG_NOMATCH)
4550  break;
4551 
4552  if (regexec_result != REG_OKAY)
4553  {
4554  char errMsg[100];
4555 
4557  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
4558  ereport(ERROR,
4559  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4560  errmsg("regular expression failed: %s", errMsg)));
4561  }
4562 
4563  /*
4564  * Copy the text to the left of the match position. Note we are given
4565  * character not byte indexes.
4566  */
4567  if (pmatch[0].rm_so - data_pos > 0)
4568  {
4569  int chunk_len;
4570 
4571  chunk_len = charlen_to_bytelen(start_ptr,
4572  pmatch[0].rm_so - data_pos);
4573  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4574 
4575  /*
4576  * Advance start_ptr over that text, to avoid multiple rescans of
4577  * it if the replace_text contains multiple back-references.
4578  */
4579  start_ptr += chunk_len;
4580  data_pos = pmatch[0].rm_so;
4581  }
4582 
4583  /*
4584  * Copy the replace_text. Process back references when the
4585  * replace_text has escape characters.
4586  */
4587  if (have_escape)
4588  appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
4589  start_ptr, data_pos);
4590  else
4591  appendStringInfoText(&buf, replace_text);
4592 
4593  /* Advance start_ptr and data_pos over the matched text. */
4594  start_ptr += charlen_to_bytelen(start_ptr,
4595  pmatch[0].rm_eo - data_pos);
4596  data_pos = pmatch[0].rm_eo;
4597 
4598  /*
4599  * When global option is off, replace the first instance only.
4600  */
4601  if (!glob)
4602  break;
4603 
4604  /*
4605  * Advance search position. Normally we start the next search at the
4606  * end of the previous match; but if the match was of zero length, we
4607  * have to advance by one character, or we'd just find the same match
4608  * again.
4609  */
4610  search_start = data_pos;
4611  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4612  search_start++;
4613  }
4614 
4615  /*
4616  * Copy the text to the right of the last match.
4617  */
4618  if (data_pos < data_len)
4619  {
4620  int chunk_len;
4621 
4622  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4623  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4624  }
4625 
4626  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4627  pfree(buf.data);
4628  pfree(data);
4629 
4630  return ret_text;
4631 }
#define VARDATA_ANY(PTR)
Definition: postgres.h:361
regoff_t rm_so
Definition: regex.h:87
int errcode(int sqlerrcode)
Definition: elog.c:698
regoff_t rm_eo
Definition: regex.h:88
void pfree(void *pointer)
Definition: mcxt.c:1169
#define REG_OKAY
Definition: regex.h:139
#define ERROR
Definition: elog.h:46
static bool check_replace_text_has_escape_char(const text *replace_text)
Definition: varlena.c:4367
static int charlen_to_bytelen(const char *p, int n)
Definition: varlena.c:793
static void appendStringInfoText(StringInfo str, const text *t)
Definition: varlena.c:4278
static char * buf
Definition: pg_test_fsync.c:68
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:202
size_t pg_regerror(int errcode, const regex_t *preg, char *errbuf, size_t errbuf_size)
Definition: regerror.c:60
unsigned int pg_wchar
Definition: mbprint.c:31
void initStringInfo(StringInfo str)
Definition: stringinfo.c:59
#define REGEXP_REPLACE_BACKREF_CNT
Definition: varlena.c:4494
#define VARSIZE_ANY(PTR)
Definition: postgres.h:348
int pg_mb2wchar_with_len(const char *from, pg_wchar *to, int len)
Definition: mbutils.c:929
#define ereport(elevel,...)
Definition: elog.h:157
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:4400
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:176
#define VARSIZE_ANY_EXHDR(PTR)
Definition: postgres.h:354
void * palloc(Size size)
Definition: mcxt.c:1062
int errmsg(const char *fmt,...)
Definition: elog.c:909
#define REG_NOMATCH
Definition: regex.h:140
Definition: c.h:621
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:102
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 3880 of file varlena.c.

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

Referenced by load_libraries(), and PostmasterMain().

3882 {
3883  char *nextp = rawstring;
3884  bool done = false;
3885 
3886  *namelist = NIL;
3887 
3888  while (scanner_isspace(*nextp))
3889  nextp++; /* skip leading whitespace */
3890 
3891  if (*nextp == '\0')
3892  return true; /* allow empty string */
3893 
3894  /* At the top of the loop, we are at start of a new directory. */
3895  do
3896  {
3897  char *curname;
3898  char *endp;
3899 
3900  if (*nextp == '"')
3901  {
3902  /* Quoted name --- collapse quote-quote pairs */
3903  curname = nextp + 1;
3904  for (;;)
3905  {
3906  endp = strchr(nextp + 1, '"');
3907  if (endp == NULL)
3908  return false; /* mismatched quotes */
3909  if (endp[1] != '"')
3910  break; /* found end of quoted name */
3911  /* Collapse adjacent quotes into one quote, and look again */
3912  memmove(endp, endp + 1, strlen(endp));
3913  nextp = endp;
3914  }
3915  /* endp now points at the terminating quote */
3916  nextp = endp + 1;
3917  }
3918  else
3919  {
3920  /* Unquoted name --- extends to separator or end of string */
3921  curname = endp = nextp;
3922  while (*nextp && *nextp != separator)
3923  {
3924  /* trailing whitespace should not be included in name */
3925  if (!scanner_isspace(*nextp))
3926  endp = nextp + 1;
3927  nextp++;
3928  }
3929  if (curname == endp)
3930  return false; /* empty unquoted name not allowed */
3931  }
3932 
3933  while (scanner_isspace(*nextp))
3934  nextp++; /* skip trailing whitespace */
3935 
3936  if (*nextp == separator)
3937  {
3938  nextp++;
3939  while (scanner_isspace(*nextp))
3940  nextp++; /* skip leading whitespace for next */
3941  /* we expect another name, so done remains false */
3942  }
3943  else if (*nextp == '\0')
3944  done = true;
3945  else
3946  return false; /* invalid syntax */
3947 
3948  /* Now safe to overwrite separator with a null */
3949  *endp = '\0';
3950 
3951  /* Truncate path if it's overlength */
3952  if (strlen(curname) >= MAXPGPATH)
3953  curname[MAXPGPATH - 1] = '\0';
3954 
3955  /*
3956  * Finished isolating current name --- add it to list
3957  */
3958  curname = pstrdup(curname);
3959  canonicalize_path(curname);
3960  *namelist = lappend(*namelist, curname);
3961 
3962  /* Loop back if we didn't reach end of string */
3963  } while (!done);
3964 
3965  return true;
3966 }
#define NIL
Definition: pg_list.h:65
char * pstrdup(const char *in)
Definition: mcxt.c:1299
void canonicalize_path(char *path)
Definition: path.c:254
#define MAXPGPATH
List * lappend(List *list, void *datum)
Definition: list.c:336
bool scanner_isspace(char ch)
Definition: scansup.c:117

◆ SplitGUCList()

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

Definition at line 4001 of file varlena.c.

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

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

4003 {
4004  char *nextp = rawstring;
4005  bool done = false;
4006 
4007  *namelist = NIL;
4008 
4009  while (scanner_isspace(*nextp))
4010  nextp++; /* skip leading whitespace */
4011 
4012  if (*nextp == '\0')
4013  return true; /* allow empty string */
4014 
4015  /* At the top of the loop, we are at start of a new identifier. */
4016  do
4017  {
4018  char *curname;
4019  char *endp;
4020 
4021  if (*nextp == '"')
4022  {
4023  /* Quoted name --- collapse quote-quote pairs */
4024  curname = nextp + 1;
4025  for (;;)
4026  {
4027  endp = strchr(nextp + 1, '"');
4028  if (endp == NULL)
4029  return false; /* mismatched quotes */
4030  if (endp[1] != '"')
4031  break; /* found end of quoted name */
4032  /* Collapse adjacent quotes into one quote, and look again */
4033  memmove(endp, endp + 1, strlen(endp));
4034  nextp = endp;
4035  }
4036  /* endp now points at the terminating quote */
4037  nextp = endp + 1;
4038  }
4039  else
4040  {
4041  /* Unquoted name --- extends to separator or whitespace */
4042  curname = nextp;
4043  while (*nextp && *nextp != separator &&
4044  !scanner_isspace(*nextp))
4045  nextp++;
4046  endp = nextp;
4047  if (curname == nextp)
4048  return false; /* empty unquoted name not allowed */
4049  }
4050 
4051  while (scanner_isspace(*nextp))
4052  nextp++; /* skip trailing whitespace */
4053 
4054  if (*nextp == separator)
4055  {
4056  nextp++;
4057  while (scanner_isspace(*nextp))
4058  nextp++; /* skip leading whitespace for next */
4059  /* we expect another name, so done remains false */
4060  }
4061  else if (*nextp == '\0')
4062  done = true;
4063  else
4064  return false; /* invalid syntax */
4065 
4066  /* Now safe to overwrite separator with a null */
4067  *endp = '\0';
4068 
4069  /*
4070  * Finished isolating current name --- add it to list
4071  */
4072  *namelist = lappend(*namelist, curname);
4073 
4074  /* Loop back if we didn't reach end of string */
4075  } while (!done);
4076 
4077  return true;
4078 }
#define NIL
Definition: pg_list.h:65
List * lappend(List *list, void *datum)
Definition: list.c:336
bool scanner_isspace(char ch)
Definition: scansup.c:117

◆ SplitIdentifierString()

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

Definition at line 3753 of file varlena.c.

References Assert, downcase_truncate_identifier(), lappend(), 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_output_parameters(), parse_publication_options(), plpgsql_extra_checks_check_hook(), PrepareTempTablespaces(), recomputeNamespacePath(), stringToQualifiedNameList(), and textToQualifiedNameList().

3755 {
3756  char *nextp = rawstring;
3757  bool done = false;
3758 
3759  *namelist = NIL;
3760 
3761  while (scanner_isspace(*nextp))
3762  nextp++; /* skip leading whitespace */
3763 
3764  if (*nextp == '\0')
3765  return true; /* allow empty string */
3766 
3767  /* At the top of the loop, we are at start of a new identifier. */
3768  do
3769  {
3770  char *curname;
3771  char *endp;
3772 
3773  if (*nextp == '"')
3774  {
3775  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3776  curname = nextp + 1;
3777  for (;;)
3778  {
3779  endp = strchr(nextp + 1, '"');
3780  if (endp == NULL)
3781  return false; /* mismatched quotes */
3782  if (endp[1] != '"')
3783  break; /* found end of quoted name */
3784  /* Collapse adjacent quotes into one quote, and look again */
3785  memmove(endp, endp + 1, strlen(endp));
3786  nextp = endp;
3787  }
3788  /* endp now points at the terminating quote */
3789  nextp = endp + 1;
3790  }
3791  else
3792  {
3793  /* Unquoted name --- extends to separator or whitespace */
3794  char *downname;
3795  int len;
3796 
3797  curname = nextp;
3798  while (*nextp && *nextp != separator &&
3799  !scanner_isspace(*nextp))
3800  nextp++;
3801  endp = nextp;
3802  if (curname == nextp)
3803  return false; /* empty unquoted name not allowed */
3804 
3805  /*
3806  * Downcase the identifier, using same code as main lexer does.
3807  *
3808  * XXX because we want to overwrite the input in-place, we cannot
3809  * support a downcasing transformation that increases the string
3810  * length. This is not a problem given the current implementation
3811  * of downcase_truncate_identifier, but we'll probably have to do
3812  * something about this someday.
3813  */
3814  len = endp - curname;
3815  downname = downcase_truncate_identifier(curname, len, false);
3816  Assert(strlen(downname) <= len);
3817  strncpy(curname, downname, len); /* strncpy is required here */
3818  pfree(downname);
3819  }
3820 
3821  while (scanner_isspace(*nextp))
3822  nextp++; /* skip trailing whitespace */
3823 
3824  if (*nextp == separator)
3825  {
3826  nextp++;
3827  while (scanner_isspace(*nextp))
3828  nextp++; /* skip leading whitespace for next */
3829  /* we expect another name, so done remains false */
3830  }
3831  else if (*nextp == '\0')
3832  done = true;
3833  else
3834  return false; /* invalid syntax */
3835 
3836  /* Now safe to overwrite separator with a null */
3837  *endp = '\0';
3838 
3839  /* Truncate name if it's overlength */
3840  truncate_identifier(curname, strlen(curname), false);
3841 
3842  /*
3843  * Finished isolating current name --- add it to list
3844  */
3845  *namelist = lappend(*namelist, curname);
3846 
3847  /* Loop back if we didn't reach end of string */
3848  } while (!done);
3849 
3850  return true;
3851 }
#define NIL
Definition: pg_list.h:65
char * downcase_truncate_identifier(const char *ident, int len, bool warn)
Definition: scansup.c:37
void truncate_identifier(char *ident, int len, bool warn)
Definition: scansup.c:93
void pfree(void *pointer)
Definition: mcxt.c:1169
List * lappend(List *list, void *datum)
Definition: list.c:336
bool scanner_isspace(char ch)
Definition: scansup.c:117
#define Assert(condition)
Definition: c.h:804

◆ textToQualifiedNameList()

List* textToQualifiedNameList ( text textval)

Definition at line 3695 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_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().

3696 {
3697  char *rawname;
3698  List *result = NIL;
3699  List *namelist;
3700  ListCell *l;
3701 
3702  /* Convert to C string (handles possible detoasting). */
3703  /* Note we rely on being able to modify rawname below. */
3704  rawname = text_to_cstring(textval);
3705 
3706  if (!SplitIdentifierString(rawname, '.', &namelist))
3707  ereport(ERROR,
3708  (errcode(ERRCODE_INVALID_NAME),
3709  errmsg("invalid name syntax")));
3710 
3711  if (namelist == NIL)
3712  ereport(ERROR,
3713  (errcode(ERRCODE_INVALID_NAME),
3714  errmsg("invalid name syntax")));
3715 
3716  foreach(l, namelist)
3717  {
3718  char *curname = (char *) lfirst(l);
3719 
3720  result = lappend(result, makeString(pstrdup(curname)));
3721  }
3722 
3723  pfree(rawname);
3724  list_free(namelist);
3725 
3726  return result;
3727 }
Value * makeString(char *str)
Definition: value.c:53
#define NIL
Definition: pg_list.h:65
char * pstrdup(const char *in)
Definition: mcxt.c:1299
int errcode(int sqlerrcode)
Definition: elog.c:698
void pfree(void *pointer)
Definition: mcxt.c:1169
#define ERROR
Definition: elog.h:46
bool SplitIdentifierString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3753
List * lappend(List *list, void *datum)
Definition: list.c:336
#define ereport(elevel,...)
Definition: elog.h:157
#define lfirst(lc)
Definition: pg_list.h:169
char * text_to_cstring(const text *t)
Definition: varlena.c:223
int errmsg(const char *fmt,...)
Definition: elog.c:909
void list_free(List *list)
Definition: list.c:1391
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 1542 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().

1543 {
1544  int result;
1545 
1546  check_collation_set(collid);
1547 
1548  /*
1549  * Unfortunately, there is no strncoll(), so in the non-C locale case we
1550  * have to do some memory copying. This turns out to be significantly
1551  * slower, so we optimize the case where LC_COLLATE is C. We also try to
1552  * optimize relatively-short strings by avoiding palloc/pfree overhead.
1553  */
1554  if (lc_collate_is_c(collid))
1555  {
1556  result = memcmp(arg1, arg2, Min(len1, len2));
1557  if ((result == 0) && (len1 != len2))
1558  result = (len1 < len2) ? -1 : 1;
1559  }
1560  else
1561  {
1562  char a1buf[TEXTBUFLEN];
1563  char a2buf[TEXTBUFLEN];
1564  char *a1p,
1565  *a2p;
1566  pg_locale_t mylocale = 0;
1567 
1568  if (collid != DEFAULT_COLLATION_OID)
1569  mylocale = pg_newlocale_from_collation(collid);
1570 
1571  /*
1572  * memcmp() can't tell us which of two unequal strings sorts first,
1573  * but it's a cheap way to tell if they're equal. Testing shows that
1574  * memcmp() followed by strcoll() is only trivially slower than
1575  * strcoll() by itself, so we don't lose much if this doesn't work out
1576  * very often, and if it does - for example, because there are many
1577  * equal strings in the input - then we win big by avoiding expensive
1578  * collation-aware comparisons.
1579  */
1580  if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1581  return 0;
1582 
1583 #ifdef WIN32
1584  /* Win32 does not have UTF-8, so we need to map to UTF-16 */
1585  if (GetDatabaseEncoding() == PG_UTF8
1586  && (!mylocale || mylocale->provider == COLLPROVIDER_LIBC))
1587  {
1588  int a1len;
1589  int a2len;
1590  int r;
1591 
1592  if (len1 >= TEXTBUFLEN / 2)
1593  {
1594  a1len = len1 * 2 + 2;
1595  a1p = palloc(a1len);
1596  }
1597  else
1598  {
1599  a1len = TEXTBUFLEN;
1600  a1p = a1buf;
1601  }
1602  if (len2 >= TEXTBUFLEN / 2)
1603  {
1604  a2len = len2 * 2 + 2;
1605  a2p = palloc(a2len);
1606  }
1607  else
1608  {
1609  a2len = TEXTBUFLEN;
1610  a2p = a2buf;
1611  }
1612 
1613  /* stupid Microsloth API does not work for zero-length input */
1614  if (len1 == 0)
1615  r = 0;
1616  else
1617  {
1618  r = MultiByteToWideChar(CP_UTF8, 0, arg1, len1,
1619  (LPWSTR) a1p, a1len / 2);
1620  if (!r)
1621  ereport(ERROR,
1622  (errmsg("could not convert string to UTF-16: error code %lu",
1623  GetLastError())));
1624  }
1625  ((LPWSTR) a1p)[r] = 0;
1626 
1627  if (len2 == 0)
1628  r = 0;
1629  else
1630  {
1631  r = MultiByteToWideChar(CP_UTF8, 0, arg2, len2,
1632  (LPWSTR) a2p, a2len / 2);
1633  if (!r)
1634  ereport(ERROR,
1635  (errmsg("could not convert string to UTF-16: error code %lu",
1636  GetLastError())));
1637  }
1638  ((LPWSTR) a2p)[r] = 0;
1639 
1640  errno = 0;
1641 #ifdef HAVE_LOCALE_T
1642  if (mylocale)
1643  result = wcscoll_l((LPWSTR) a1p, (LPWSTR) a2p, mylocale->info.lt);
1644  else
1645 #endif
1646  result = wcscoll((LPWSTR) a1p, (LPWSTR) a2p);
1647  if (result == 2147483647) /* _NLSCMPERROR; missing from mingw
1648  * headers */
1649  ereport(ERROR,
1650  (errmsg("could not compare Unicode strings: %m")));
1651 
1652  /* Break tie if necessary. */
1653  if (result == 0 &&
1654  (!mylocale || mylocale->deterministic))
1655  {
1656  result = memcmp(arg1, arg2, Min(len1, len2));
1657  if ((result == 0) && (len1 != len2))
1658  result = (len1 < len2) ? -1 : 1;
1659  }
1660 
1661  if (a1p != a1buf)
1662  pfree(a1p);
1663  if (a2p != a2buf)
1664  pfree(a2p);
1665 
1666  return result;
1667  }
1668 #endif /* WIN32 */
1669 
1670  if (len1 >= TEXTBUFLEN)
1671  a1p = (char *) palloc(len1 + 1);
1672  else
1673  a1p = a1buf;
1674  if (len2 >= TEXTBUFLEN)
1675  a2p = (char *) palloc(len2 + 1);
1676  else
1677  a2p = a2buf;
1678 
1679  memcpy(a1p, arg1, len1);
1680  a1p[len1] = '\0';
1681  memcpy(a2p, arg2, len2);
1682  a2p[len2] = '\0';
1683 
1684  if (mylocale)
1685  {
1686  if (mylocale->provider == COLLPROVIDER_ICU)
1687  {
1688 #ifdef USE_ICU
1689 #ifdef HAVE_UCOL_STRCOLLUTF8
1690  if (GetDatabaseEncoding() == PG_UTF8)
1691  {
1692  UErrorCode status;
1693 
1694  status = U_ZERO_ERROR;
1695  result = ucol_strcollUTF8(mylocale->info.icu.ucol,
1696  arg1, len1,
1697  arg2, len2,
1698  &status);
1699  if (U_FAILURE(status))
1700  ereport(ERROR,
1701  (errmsg("collation failed: %s", u_errorName(status))));
1702  }
1703  else
1704 #endif
1705  {
1706  int32_t ulen1,
1707  ulen2;
1708  UChar *uchar1,
1709  *uchar2;
1710 
1711  ulen1 = icu_to_uchar(&uchar1, arg1, len1);
1712  ulen2 = icu_to_uchar(&uchar2, arg2, len2);
1713 
1714  result = ucol_strcoll(mylocale->info.icu.ucol,
1715  uchar1, ulen1,
1716  uchar2, ulen2);
1717 
1718  pfree(uchar1);
1719  pfree(uchar2);
1720  }
1721 #else /* not USE_ICU */
1722  /* shouldn't happen */
1723  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1724 #endif /* not USE_ICU */
1725  }
1726  else
1727  {
1728 #ifdef HAVE_LOCALE_T
1729  result = strcoll_l(a1p, a2p, mylocale->info.lt);
1730 #else
1731  /* shouldn't happen */
1732  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1733 #endif
1734  }
1735  }
1736  else
1737  result = strcoll(a1p, a2p);
1738 
1739  /* Break tie if necessary. */
1740  if (result == 0 &&
1741  (!mylocale || mylocale->deterministic))
1742  result = strcmp(a1p, a2p);
1743 
1744  if (a1p != a1buf)
1745  pfree(a1p);
1746  if (a2p != a2buf)
1747  pfree(a2p);
1748  }
1749 
1750  return result;
1751 }
union pg_locale_struct::@141 info
#define Min(x, y)
Definition: c.h:986
void pfree(void *pointer)
Definition: mcxt.c:1169
#define ERROR
Definition: elog.h:46
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1326
static void check_collation_set(Oid collid)
Definition: varlena.c:1513
#define strcoll_l
Definition: win32_port.h:414
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1473
int GetDatabaseEncoding(void)
Definition: mbutils.c:1210
bool deterministic
Definition: pg_locale.h:85
#define wcscoll_l
Definition: win32_port.h:416
#define ereport(elevel,...)
Definition: elog.h:157
#define TEXTBUFLEN
Definition: varlena.c:113
void * palloc(Size size)
Definition: mcxt.c:1062
int errmsg(const char *fmt,...)
Definition: elog.c:909
#define elog(elevel,...)
Definition: elog.h:232
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:227

◆ 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(), source, 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:986
int errcode(int sqlerrcode)
Definition: elog.c:698
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:1000
#define STOP_COLUMN
#define ERROR
Definition: elog.h:46
#define START_COLUMN
#define ereport(elevel,...)
Definition: elog.h:157
#define MAX_LEVENSHTEIN_STRLEN
Definition: levenshtein.c:26
static rewind_source * source
Definition: pg_rewind.c:79
static bool rest_of_char_same(const char *s1, const char *s2, int len)
Definition: varlena.c:6219
int pg_mblen(const char *mbstr)
Definition: mbutils.c:966
void * palloc(Size size)
Definition: mcxt.c:1062
int errmsg(const char *fmt,...)
Definition: elog.c:909
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 2027 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().

2028 {
2029  bool abbreviate = ssup->abbreviate;
2030  bool collate_c = false;
2031  VarStringSortSupport *sss;
2032  pg_locale_t locale = 0;
2033 
2034  check_collation_set(collid);
2035 
2036  /*
2037  * If possible, set ssup->comparator to a function which can be used to
2038  * directly compare two datums. If we can do this, we'll avoid the
2039  * overhead of a trip through the fmgr layer for every comparison, which
2040  * can be substantial.
2041  *
2042  * Most typically, we'll set the comparator to varlenafastcmp_locale,
2043  * which uses strcoll() to perform comparisons. We use that for the
2044  * BpChar case too, but type NAME uses namefastcmp_locale. However, if
2045  * LC_COLLATE = C, we can make things quite a bit faster with
2046  * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
2047  * memcmp() rather than strcoll().
2048  */
2049  if (lc_collate_is_c(collid))
2050  {
2051  if (typid == BPCHAROID)
2052  ssup->comparator = bpcharfastcmp_c;
2053  else if (typid == NAMEOID)
2054  {
2055  ssup->comparator = namefastcmp_c;
2056  /* Not supporting abbreviation with type NAME, for now */
2057  abbreviate = false;
2058  }
2059  else
2060  ssup->comparator = varstrfastcmp_c;
2061 
2062  collate_c = true;
2063  }
2064  else
2065  {
2066  /*
2067  * We need a collation-sensitive comparison. To make things faster,
2068  * we'll figure out the collation based on the locale id and cache the
2069  * result.
2070  */
2071  if (collid != DEFAULT_COLLATION_OID)
2072  locale = pg_newlocale_from_collation(collid);
2073 
2074  /*
2075  * There is a further exception on Windows. When the database
2076  * encoding is UTF-8 and we are not using the C collation, complex
2077  * hacks are required. We don't currently have a comparator that
2078  * handles that case, so we fall back on the slow method of having the
2079  * sort code invoke bttextcmp() (in the case of text) via the fmgr
2080  * trampoline. ICU locales work just the same on Windows, however.
2081  */
2082 #ifdef WIN32
2083  if (GetDatabaseEncoding() == PG_UTF8 &&
2084  !(locale && locale->provider == COLLPROVIDER_ICU))
2085  return;
2086 #endif
2087 
2088  /*
2089  * We use varlenafastcmp_locale except for type NAME.
2090  */
2091  if (typid == NAMEOID)
2092  {
2094  /* Not supporting abbreviation with type NAME, for now */
2095  abbreviate = false;
2096  }
2097  else
2099  }
2100 
2101  /*
2102  * Unfortunately, it seems that abbreviation for non-C collations is
2103  * broken on many common platforms; testing of multiple versions of glibc
2104  * reveals that, for many locales, strcoll() and strxfrm() do not return
2105  * consistent results, which is fatal to this optimization. While no
2106  * other libc other than Cygwin has so far been shown to have a problem,
2107  * we take the conservative course of action for right now and disable
2108  * this categorically. (Users who are certain this isn't a problem on
2109  * their system can define TRUST_STRXFRM.)
2110  *
2111  * Even apart from the risk of broken locales, it's possible that there
2112  * are platforms where the use of abbreviated keys should be disabled at
2113  * compile time. Having only 4 byte datums could make worst-case
2114  * performance drastically more likely, for example. Moreover, macOS's
2115  * strxfrm() implementation is known to not effectively concentrate a
2116  * significant amount of entropy from the original string in earlier
2117  * transformed blobs. It's possible that other supported platforms are
2118  * similarly encumbered. So, if we ever get past disabling this
2119  * categorically, we may still want or need to disable it for particular
2120  * platforms.
2121  */
2122 #ifndef TRUST_STRXFRM
2123  if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
2124  abbreviate = false;
2125 #endif
2126 
2127  /*
2128  * If we're using abbreviated keys, or if we're using a locale-aware
2129  * comparison, we need to initialize a VarStringSortSupport object. Both
2130  * cases will make use of the temporary buffers we initialize here for
2131  * scratch space (and to detect requirement for BpChar semantics from
2132  * caller), and the abbreviation case requires additional state.
2133  */
2134  if (abbreviate || !collate_c)
2135  {
2136  sss = palloc(sizeof(VarStringSortSupport));
2137  sss->buf1 = palloc(TEXTBUFLEN);
2138  sss->buflen1 = TEXTBUFLEN;
2139  sss->buf2 = palloc(TEXTBUFLEN);
2140  sss->buflen2 = TEXTBUFLEN;
2141  /* Start with invalid values */
2142  sss->last_len1 = -1;
2143  sss->last_len2 = -1;
2144  /* Initialize */
2145  sss->last_returned = 0;
2146  sss->locale = locale;
2147 
2148  /*
2149  * To avoid somehow confusing a strxfrm() blob and an original string,
2150  * constantly keep track of the variety of data that buf1 and buf2
2151  * currently contain.
2152  *
2153  * Comparisons may be interleaved with conversion calls. Frequently,
2154  * conversions and comparisons are batched into two distinct phases,
2155  * but the correctness of caching cannot hinge upon this. For
2156  * comparison caching, buffer state is only trusted if cache_blob is
2157  * found set to false, whereas strxfrm() caching only trusts the state
2158  * when cache_blob is found set to true.
2159  *
2160  * Arbitrarily initialize cache_blob to true.
2161  */
2162  sss->cache_blob = true;
2163  sss->collate_c = collate_c;
2164  sss->typid = typid;
2165  ssup->ssup_extra = sss;
2166 
2167  /*
2168  * If possible, plan to use the abbreviated keys optimization. The
2169  * core code may switch back to authoritative comparator should
2170  * abbreviation be aborted.
2171  */
2172  if (abbreviate)
2173  {
2174  sss->prop_card = 0.20;
2175  initHyperLogLog(&sss->abbr_card, 10);
2176  initHyperLogLog(&sss->full_card, 10);
2177  ssup->abbrev_full_comparator = ssup->comparator;
2178  ssup->comparator = varstrcmp_abbrev;
2181  }
2182  }
2183 }
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2226
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2493
static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2468
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2189
static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2271
void initHyperLogLog(hyperLogLogState *cState, uint8 bwidth)
Definition: hyperloglog.c:66
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1326
static void check_collation_set(Oid collid)
Definition: varlena.c:1513
static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2302
int(* comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:106
hyperLogLogState abbr_card
Definition: varlena.c:92
pg_locale_t locale
Definition: varlena.c:95
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:1473
int GetDatabaseEncoding(void)
Definition: mbutils.c:1210
hyperLogLogState full_card
Definition: varlena.c:93
bool(* abbrev_abort)(int memtupcount, SortSupport ssup)
Definition: sortsupport.h:182
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2733
#define TEXTBUFLEN
Definition: varlena.c:113
void * palloc(Size size)
Definition: mcxt.c:1062
static char * locale
Definition: initdb.c:126
static int namefastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2259