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 4438 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().

4440 {
4441  text *ret_text;
4442  regex_t *re = (regex_t *) regexp;
4443  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
4446  pg_wchar *data;
4447  size_t data_len;
4448  int search_start;
4449  int data_pos;
4450  char *start_ptr;
4451  bool have_escape;
4452 
4453  initStringInfo(&buf);
4454 
4455  /* Convert data string to wide characters. */
4456  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
4457  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
4458 
4459  /* Check whether replace_text has escape char. */
4460  have_escape = check_replace_text_has_escape_char(replace_text);
4461 
4462  /* start_ptr points to the data_pos'th character of src_text */
4463  start_ptr = (char *) VARDATA_ANY(src_text);
4464  data_pos = 0;
4465 
4466  search_start = 0;
4467  while (search_start <= data_len)
4468  {
4469  int regexec_result;
4470 
4472 
4473  regexec_result = pg_regexec(re,
4474  data,
4475  data_len,
4476  search_start,
4477  NULL, /* no details */
4479  pmatch,
4480  0);
4481 
4482  if (regexec_result == REG_NOMATCH)
4483  break;
4484 
4485  if (regexec_result != REG_OKAY)
4486  {
4487  char errMsg[100];
4488 
4490  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
4491  ereport(ERROR,
4492  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
4493  errmsg("regular expression failed: %s", errMsg)));
4494  }
4495 
4496  /*
4497  * Copy the text to the left of the match position. Note we are given
4498  * character not byte indexes.
4499  */
4500  if (pmatch[0].rm_so - data_pos > 0)
4501  {
4502  int chunk_len;
4503 
4504  chunk_len = charlen_to_bytelen(start_ptr,
4505  pmatch[0].rm_so - data_pos);
4506  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4507 
4508  /*
4509  * Advance start_ptr over that text, to avoid multiple rescans of
4510  * it if the replace_text contains multiple back-references.
4511  */
4512  start_ptr += chunk_len;
4513  data_pos = pmatch[0].rm_so;
4514  }
4515 
4516  /*
4517  * Copy the replace_text. Process back references when the
4518  * replace_text has escape characters.
4519  */
4520  if (have_escape)
4521  appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
4522  start_ptr, data_pos);
4523  else
4524  appendStringInfoText(&buf, replace_text);
4525 
4526  /* Advance start_ptr and data_pos over the matched text. */
4527  start_ptr += charlen_to_bytelen(start_ptr,
4528  pmatch[0].rm_eo - data_pos);
4529  data_pos = pmatch[0].rm_eo;
4530 
4531  /*
4532  * When global option is off, replace the first instance only.
4533  */
4534  if (!glob)
4535  break;
4536 
4537  /*
4538  * Advance search position. Normally we start the next search at the
4539  * end of the previous match; but if the match was of zero length, we
4540  * have to advance by one character, or we'd just find the same match
4541  * again.
4542  */
4543  search_start = data_pos;
4544  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4545  search_start++;
4546  }
4547 
4548  /*
4549  * Copy the text to the right of the last match.
4550  */
4551  if (data_pos < data_len)
4552  {
4553  int chunk_len;
4554 
4555  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4556  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4557  }
4558 
4559  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4560  pfree(buf.data);
4561  pfree(data);
4562 
4563  return ret_text;
4564 }
#define VARDATA_ANY(PTR)
Definition: postgres.h:348
regoff_t rm_so
Definition: regex.h:85
int errcode(int sqlerrcode)
Definition: elog.c:610
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:4300
static int charlen_to_bytelen(const char *p, int n)
Definition: varlena.c:769
static void appendStringInfoText(StringInfo str, const text *t)
Definition: varlena.c:4211
static char * buf
Definition: pg_test_fsync.c:67
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:184
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:4427
#define VARSIZE_ANY(PTR)
Definition: postgres.h:335
int pg_mb2wchar_with_len(const char *from, pg_wchar *to, int len)
Definition: mbutils.c:870
#define ereport(elevel,...)
Definition: elog.h:144
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:4333
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:824
#define REG_NOMATCH
Definition: regex.h:138
Definition: c.h:555
#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 3813 of file varlena.c.

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

Referenced by load_libraries(), and PostmasterMain().

3815 {
3816  char *nextp = rawstring;
3817  bool done = false;
3818 
3819  *namelist = NIL;
3820 
3821  while (scanner_isspace(*nextp))
3822  nextp++; /* skip leading whitespace */
3823 
3824  if (*nextp == '\0')
3825  return true; /* allow empty string */
3826 
3827  /* At the top of the loop, we are at start of a new directory. */
3828  do
3829  {
3830  char *curname;
3831  char *endp;
3832 
3833  if (*nextp == '"')
3834  {
3835  /* Quoted name --- collapse quote-quote pairs */
3836  curname = nextp + 1;
3837  for (;;)
3838  {
3839  endp = strchr(nextp + 1, '"');
3840  if (endp == NULL)
3841  return false; /* mismatched quotes */
3842  if (endp[1] != '"')
3843  break; /* found end of quoted name */
3844  /* Collapse adjacent quotes into one quote, and look again */
3845  memmove(endp, endp + 1, strlen(endp));
3846  nextp = endp;
3847  }
3848  /* endp now points at the terminating quote */
3849  nextp = endp + 1;
3850  }
3851  else
3852  {
3853  /* Unquoted name --- extends to separator or end of string */
3854  curname = endp = nextp;
3855  while (*nextp && *nextp != separator)
3856  {
3857  /* trailing whitespace should not be included in name */
3858  if (!scanner_isspace(*nextp))
3859  endp = nextp + 1;
3860  nextp++;
3861  }
3862  if (curname == endp)
3863  return false; /* empty unquoted name not allowed */
3864  }
3865 
3866  while (scanner_isspace(*nextp))
3867  nextp++; /* skip trailing whitespace */
3868 
3869  if (*nextp == separator)
3870  {
3871  nextp++;
3872  while (scanner_isspace(*nextp))
3873  nextp++; /* skip leading whitespace for next */
3874  /* we expect another name, so done remains false */
3875  }
3876  else if (*nextp == '\0')
3877  done = true;
3878  else
3879  return false; /* invalid syntax */
3880 
3881  /* Now safe to overwrite separator with a null */
3882  *endp = '\0';
3883 
3884  /* Truncate path if it's overlength */
3885  if (strlen(curname) >= MAXPGPATH)
3886  curname[MAXPGPATH - 1] = '\0';
3887 
3888  /*
3889  * Finished isolating current name --- add it to list
3890  */
3891  curname = pstrdup(curname);
3892  canonicalize_path(curname);
3893  *namelist = lappend(*namelist, curname);
3894 
3895  /* Loop back if we didn't reach end of string */
3896  } while (!done);
3897 
3898  return true;
3899 }
#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
List * lappend(List *list, void *datum)
Definition: list.c:321
bool scanner_isspace(char ch)
Definition: scansup.c:220

◆ SplitGUCList()

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

Definition at line 3934 of file varlena.c.

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

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

3936 {
3937  char *nextp = rawstring;
3938  bool done = false;
3939 
3940  *namelist = NIL;
3941 
3942  while (scanner_isspace(*nextp))
3943  nextp++; /* skip leading whitespace */
3944 
3945  if (*nextp == '\0')
3946  return true; /* allow empty string */
3947 
3948  /* At the top of the loop, we are at start of a new identifier. */
3949  do
3950  {
3951  char *curname;
3952  char *endp;
3953 
3954  if (*nextp == '"')
3955  {
3956  /* Quoted name --- collapse quote-quote pairs */
3957  curname = nextp + 1;
3958  for (;;)
3959  {
3960  endp = strchr(nextp + 1, '"');
3961  if (endp == NULL)
3962  return false; /* mismatched quotes */
3963  if (endp[1] != '"')
3964  break; /* found end of quoted name */
3965  /* Collapse adjacent quotes into one quote, and look again */
3966  memmove(endp, endp + 1, strlen(endp));
3967  nextp = endp;
3968  }
3969  /* endp now points at the terminating quote */
3970  nextp = endp + 1;
3971  }
3972  else
3973  {
3974  /* Unquoted name --- extends to separator or whitespace */
3975  curname = nextp;
3976  while (*nextp && *nextp != separator &&
3977  !scanner_isspace(*nextp))
3978  nextp++;
3979  endp = nextp;
3980  if (curname == nextp)
3981  return false; /* empty unquoted name not allowed */
3982  }
3983 
3984  while (scanner_isspace(*nextp))
3985  nextp++; /* skip trailing whitespace */
3986 
3987  if (*nextp == separator)
3988  {
3989  nextp++;
3990  while (scanner_isspace(*nextp))
3991  nextp++; /* skip leading whitespace for next */
3992  /* we expect another name, so done remains false */
3993  }
3994  else if (*nextp == '\0')
3995  done = true;
3996  else
3997  return false; /* invalid syntax */
3998 
3999  /* Now safe to overwrite separator with a null */
4000  *endp = '\0';
4001 
4002  /*
4003  * Finished isolating current name --- add it to list
4004  */
4005  *namelist = lappend(*namelist, curname);
4006 
4007  /* Loop back if we didn't reach end of string */
4008  } while (!done);
4009 
4010  return true;
4011 }
#define NIL
Definition: pg_list.h:65
List * lappend(List *list, void *datum)
Definition: list.c:321
bool scanner_isspace(char ch)
Definition: scansup.c:220

◆ SplitIdentifierString()

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

Definition at line 3686 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().

3688 {
3689  char *nextp = rawstring;
3690  bool done = false;
3691 
3692  *namelist = NIL;
3693 
3694  while (scanner_isspace(*nextp))
3695  nextp++; /* skip leading whitespace */
3696 
3697  if (*nextp == '\0')
3698  return true; /* allow empty string */
3699 
3700  /* At the top of the loop, we are at start of a new identifier. */
3701  do
3702  {
3703  char *curname;
3704  char *endp;
3705 
3706  if (*nextp == '"')
3707  {
3708  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3709  curname = nextp + 1;
3710  for (;;)
3711  {
3712  endp = strchr(nextp + 1, '"');
3713  if (endp == NULL)
3714  return false; /* mismatched quotes */
3715  if (endp[1] != '"')
3716  break; /* found end of quoted name */
3717  /* Collapse adjacent quotes into one quote, and look again */
3718  memmove(endp, endp + 1, strlen(endp));
3719  nextp = endp;
3720  }
3721  /* endp now points at the terminating quote */
3722  nextp = endp + 1;
3723  }
3724  else
3725  {
3726  /* Unquoted name --- extends to separator or whitespace */
3727  char *downname;
3728  int len;
3729 
3730  curname = nextp;
3731  while (*nextp && *nextp != separator &&
3732  !scanner_isspace(*nextp))
3733  nextp++;
3734  endp = nextp;
3735  if (curname == nextp)
3736  return false; /* empty unquoted name not allowed */
3737 
3738  /*
3739  * Downcase the identifier, using same code as main lexer does.
3740  *
3741  * XXX because we want to overwrite the input in-place, we cannot
3742  * support a downcasing transformation that increases the string
3743  * length. This is not a problem given the current implementation
3744  * of downcase_truncate_identifier, but we'll probably have to do
3745  * something about this someday.
3746  */
3747  len = endp - curname;
3748  downname = downcase_truncate_identifier(curname, len, false);
3749  Assert(strlen(downname) <= len);
3750  strncpy(curname, downname, len); /* strncpy is required here */
3751  pfree(downname);
3752  }
3753 
3754  while (scanner_isspace(*nextp))
3755  nextp++; /* skip trailing whitespace */
3756 
3757  if (*nextp == separator)
3758  {
3759  nextp++;
3760  while (scanner_isspace(*nextp))
3761  nextp++; /* skip leading whitespace for next */
3762  /* we expect another name, so done remains false */
3763  }
3764  else if (*nextp == '\0')
3765  done = true;
3766  else
3767  return false; /* invalid syntax */
3768 
3769  /* Now safe to overwrite separator with a null */
3770  *endp = '\0';
3771 
3772  /* Truncate name if it's overlength */
3773  truncate_identifier(curname, strlen(curname), false);
3774 
3775  /*
3776  * Finished isolating current name --- add it to list
3777  */
3778  *namelist = lappend(*namelist, curname);
3779 
3780  /* Loop back if we didn't reach end of string */
3781  } while (!done);
3782 
3783  return true;
3784 }
#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
List * lappend(List *list, void *datum)
Definition: list.c:321
bool scanner_isspace(char ch)
Definition: scansup.c:220
#define Assert(condition)
Definition: c.h:738

◆ textToQualifiedNameList()

List* textToQualifiedNameList ( text textval)

Definition at line 3628 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().

3629 {
3630  char *rawname;
3631  List *result = NIL;
3632  List *namelist;
3633  ListCell *l;
3634 
3635  /* Convert to C string (handles possible detoasting). */
3636  /* Note we rely on being able to modify rawname below. */
3637  rawname = text_to_cstring(textval);
3638 
3639  if (!SplitIdentifierString(rawname, '.', &namelist))
3640  ereport(ERROR,
3641  (errcode(ERRCODE_INVALID_NAME),
3642  errmsg("invalid name syntax")));
3643 
3644  if (namelist == NIL)
3645  ereport(ERROR,
3646  (errcode(ERRCODE_INVALID_NAME),
3647  errmsg("invalid name syntax")));
3648 
3649  foreach(l, namelist)
3650  {
3651  char *curname = (char *) lfirst(l);
3652 
3653  result = lappend(result, makeString(pstrdup(curname)));
3654  }
3655 
3656  pfree(rawname);
3657  list_free(namelist);
3658 
3659  return result;
3660 }
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:610
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:3686
List * lappend(List *list, void *datum)
Definition: list.c:321
#define ereport(elevel,...)
Definition: elog.h:144
#define lfirst(lc)
Definition: pg_list.h:190
char * text_to_cstring(const text *t)
Definition: varlena.c:205
int errmsg(const char *fmt,...)
Definition: elog.c:824
void list_free(List *list)
Definition: list.c:1376
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 1494 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().

1495 {
1496  int result;
1497 
1498  check_collation_set(collid);
1499 
1500  /*
1501  * Unfortunately, there is no strncoll(), so in the non-C locale case we
1502  * have to do some memory copying. This turns out to be significantly
1503  * slower, so we optimize the case where LC_COLLATE is C. We also try to
1504  * optimize relatively-short strings by avoiding palloc/pfree overhead.
1505  */
1506  if (lc_collate_is_c(collid))
1507  {
1508  result = memcmp(arg1, arg2, Min(len1, len2));
1509  if ((result == 0) && (len1 != len2))
1510  result = (len1 < len2) ? -1 : 1;
1511  }
1512  else
1513  {
1514  char a1buf[TEXTBUFLEN];
1515  char a2buf[TEXTBUFLEN];
1516  char *a1p,
1517  *a2p;
1518  pg_locale_t mylocale = 0;
1519 
1520  if (collid != DEFAULT_COLLATION_OID)
1521  mylocale = pg_newlocale_from_collation(collid);
1522 
1523  /*
1524  * memcmp() can't tell us which of two unequal strings sorts first,
1525  * but it's a cheap way to tell if they're equal. Testing shows that
1526  * memcmp() followed by strcoll() is only trivially slower than
1527  * strcoll() by itself, so we don't lose much if this doesn't work out
1528  * very often, and if it does - for example, because there are many
1529  * equal strings in the input - then we win big by avoiding expensive
1530  * collation-aware comparisons.
1531  */
1532  if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1533  return 0;
1534 
1535 #ifdef WIN32
1536  /* Win32 does not have UTF-8, so we need to map to UTF-16 */
1537  if (GetDatabaseEncoding() == PG_UTF8
1538  && (!mylocale || mylocale->provider == COLLPROVIDER_LIBC))
1539  {
1540  int a1len;
1541  int a2len;
1542  int r;
1543 
1544  if (len1 >= TEXTBUFLEN / 2)
1545  {
1546  a1len = len1 * 2 + 2;
1547  a1p = palloc(a1len);
1548  }
1549  else
1550  {
1551  a1len = TEXTBUFLEN;
1552  a1p = a1buf;
1553  }
1554  if (len2 >= TEXTBUFLEN / 2)
1555  {
1556  a2len = len2 * 2 + 2;
1557  a2p = palloc(a2len);
1558  }
1559  else
1560  {
1561  a2len = TEXTBUFLEN;
1562  a2p = a2buf;
1563  }
1564 
1565  /* stupid Microsloth API does not work for zero-length input */
1566  if (len1 == 0)
1567  r = 0;
1568  else
1569  {
1570  r = MultiByteToWideChar(CP_UTF8, 0, arg1, len1,
1571  (LPWSTR) a1p, a1len / 2);
1572  if (!r)
1573  ereport(ERROR,
1574  (errmsg("could not convert string to UTF-16: error code %lu",
1575  GetLastError())));
1576  }
1577  ((LPWSTR) a1p)[r] = 0;
1578 
1579  if (len2 == 0)
1580  r = 0;
1581  else
1582  {
1583  r = MultiByteToWideChar(CP_UTF8, 0, arg2, len2,
1584  (LPWSTR) a2p, a2len / 2);
1585  if (!r)
1586  ereport(ERROR,
1587  (errmsg("could not convert string to UTF-16: error code %lu",
1588  GetLastError())));
1589  }
1590  ((LPWSTR) a2p)[r] = 0;
1591 
1592  errno = 0;
1593 #ifdef HAVE_LOCALE_T
1594  if (mylocale)
1595  result = wcscoll_l((LPWSTR) a1p, (LPWSTR) a2p, mylocale->info.lt);
1596  else
1597 #endif
1598  result = wcscoll((LPWSTR) a1p, (LPWSTR) a2p);
1599  if (result == 2147483647) /* _NLSCMPERROR; missing from mingw
1600  * headers */
1601  ereport(ERROR,
1602  (errmsg("could not compare Unicode strings: %m")));
1603 
1604  /* Break tie if necessary. */
1605  if (result == 0 &&
1606  (!mylocale || mylocale->deterministic))
1607  {
1608  result = memcmp(arg1, arg2, Min(len1, len2));
1609  if ((result == 0) && (len1 != len2))
1610  result = (len1 < len2) ? -1 : 1;
1611  }
1612 
1613  if (a1p != a1buf)
1614  pfree(a1p);
1615  if (a2p != a2buf)
1616  pfree(a2p);
1617 
1618  return result;
1619  }
1620 #endif /* WIN32 */
1621 
1622  if (len1 >= TEXTBUFLEN)
1623  a1p = (char *) palloc(len1 + 1);
1624  else
1625  a1p = a1buf;
1626  if (len2 >= TEXTBUFLEN)
1627  a2p = (char *) palloc(len2 + 1);
1628  else
1629  a2p = a2buf;
1630 
1631  memcpy(a1p, arg1, len1);
1632  a1p[len1] = '\0';
1633  memcpy(a2p, arg2, len2);
1634  a2p[len2] = '\0';
1635 
1636  if (mylocale)
1637  {
1638  if (mylocale->provider == COLLPROVIDER_ICU)
1639  {
1640 #ifdef USE_ICU
1641 #ifdef HAVE_UCOL_STRCOLLUTF8
1642  if (GetDatabaseEncoding() == PG_UTF8)
1643  {
1644  UErrorCode status;
1645 
1646  status = U_ZERO_ERROR;
1647  result = ucol_strcollUTF8(mylocale->info.icu.ucol,
1648  arg1, len1,
1649  arg2, len2,
1650  &status);
1651  if (U_FAILURE(status))
1652  ereport(ERROR,
1653  (errmsg("collation failed: %s", u_errorName(status))));
1654  }
1655  else
1656 #endif
1657  {
1658  int32_t ulen1,
1659  ulen2;
1660  UChar *uchar1,
1661  *uchar2;
1662 
1663  ulen1 = icu_to_uchar(&uchar1, arg1, len1);
1664  ulen2 = icu_to_uchar(&uchar2, arg2, len2);
1665 
1666  result = ucol_strcoll(mylocale->info.icu.ucol,
1667  uchar1, ulen1,
1668  uchar2, ulen2);
1669 
1670  pfree(uchar1);
1671  pfree(uchar2);
1672  }
1673 #else /* not USE_ICU */
1674  /* shouldn't happen */
1675  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1676 #endif /* not USE_ICU */
1677  }
1678  else
1679  {
1680 #ifdef HAVE_LOCALE_T
1681  result = strcoll_l(a1p, a2p, mylocale->info.lt);
1682 #else
1683  /* shouldn't happen */
1684  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1685 #endif
1686  }
1687  }
1688  else
1689  result = strcoll(a1p, a2p);
1690 
1691  /* Break tie if necessary. */
1692  if (result == 0 &&
1693  (!mylocale || mylocale->deterministic))
1694  result = strcmp(a1p, a2p);
1695 
1696  if (a1p != a1buf)
1697  pfree(a1p);
1698  if (a2p != a2buf)
1699  pfree(a2p);
1700  }
1701 
1702  return result;
1703 }
#define Min(x, y)
Definition: c.h:920
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:1356
static void check_collation_set(Oid collid)
Definition: varlena.c:1465
#define strcoll_l
Definition: win32_port.h:386
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1503
int GetDatabaseEncoding(void)
Definition: mbutils.c:1151
bool deterministic
Definition: pg_locale.h:85
#define wcscoll_l
Definition: win32_port.h:388
#define ereport(elevel,...)
Definition: elog.h:144
#define TEXTBUFLEN
Definition: varlena.c:99
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:824
#define elog(elevel,...)
Definition: elog.h:214
union pg_locale_struct::@142 info
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:225

◆ 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:920
int errcode(int sqlerrcode)
Definition: elog.c:610
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:941
#define STOP_COLUMN
#define ERROR
Definition: elog.h:43
#define START_COLUMN
#define ereport(elevel,...)
Definition: elog.h:144
#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:5978
int pg_mblen(const char *mbstr)
Definition: mbutils.c:907
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:824
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 1979 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().

1980 {
1981  bool abbreviate = ssup->abbreviate;
1982  bool collate_c = false;
1983  VarStringSortSupport *sss;
1984  pg_locale_t locale = 0;
1985 
1986  check_collation_set(collid);
1987 
1988  /*
1989  * If possible, set ssup->comparator to a function which can be used to
1990  * directly compare two datums. If we can do this, we'll avoid the
1991  * overhead of a trip through the fmgr layer for every comparison, which
1992  * can be substantial.
1993  *
1994  * Most typically, we'll set the comparator to varlenafastcmp_locale,
1995  * which uses strcoll() to perform comparisons. We use that for the
1996  * BpChar case too, but type NAME uses namefastcmp_locale. However, if
1997  * LC_COLLATE = C, we can make things quite a bit faster with
1998  * varstrfastcmp_c, bpcharfastcmp_c, or namefastcmp_c, all of which use
1999  * memcmp() rather than strcoll().
2000  */
2001  if (lc_collate_is_c(collid))
2002  {
2003  if (typid == BPCHAROID)
2004  ssup->comparator = bpcharfastcmp_c;
2005  else if (typid == NAMEOID)
2006  {
2007  ssup->comparator = namefastcmp_c;
2008  /* Not supporting abbreviation with type NAME, for now */
2009  abbreviate = false;
2010  }
2011  else
2012  ssup->comparator = varstrfastcmp_c;
2013 
2014  collate_c = true;
2015  }
2016  else
2017  {
2018  /*
2019  * We need a collation-sensitive comparison. To make things faster,
2020  * we'll figure out the collation based on the locale id and cache the
2021  * result.
2022  */
2023  if (collid != DEFAULT_COLLATION_OID)
2024  locale = pg_newlocale_from_collation(collid);
2025 
2026  /*
2027  * There is a further exception on Windows. When the database
2028  * encoding is UTF-8 and we are not using the C collation, complex
2029  * hacks are required. We don't currently have a comparator that
2030  * handles that case, so we fall back on the slow method of having the
2031  * sort code invoke bttextcmp() (in the case of text) via the fmgr
2032  * trampoline. ICU locales work just the same on Windows, however.
2033  */
2034 #ifdef WIN32
2035  if (GetDatabaseEncoding() == PG_UTF8 &&
2036  !(locale && locale->provider == COLLPROVIDER_ICU))
2037  return;
2038 #endif
2039 
2040  /*
2041  * We use varlenafastcmp_locale except for type NAME.
2042  */
2043  if (typid == NAMEOID)
2044  {
2046  /* Not supporting abbreviation with type NAME, for now */
2047  abbreviate = false;
2048  }
2049  else
2051  }
2052 
2053  /*
2054  * Unfortunately, it seems that abbreviation for non-C collations is
2055  * broken on many common platforms; testing of multiple versions of glibc
2056  * reveals that, for many locales, strcoll() and strxfrm() do not return
2057  * consistent results, which is fatal to this optimization. While no
2058  * other libc other than Cygwin has so far been shown to have a problem,
2059  * we take the conservative course of action for right now and disable
2060  * this categorically. (Users who are certain this isn't a problem on
2061  * their system can define TRUST_STRXFRM.)
2062  *
2063  * Even apart from the risk of broken locales, it's possible that there
2064  * are platforms where the use of abbreviated keys should be disabled at
2065  * compile time. Having only 4 byte datums could make worst-case
2066  * performance drastically more likely, for example. Moreover, macOS's
2067  * strxfrm() implementation is known to not effectively concentrate a
2068  * significant amount of entropy from the original string in earlier
2069  * transformed blobs. It's possible that other supported platforms are
2070  * similarly encumbered. So, if we ever get past disabling this
2071  * categorically, we may still want or need to disable it for particular
2072  * platforms.
2073  */
2074 #ifndef TRUST_STRXFRM
2075  if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
2076  abbreviate = false;
2077 #endif
2078 
2079  /*
2080  * If we're using abbreviated keys, or if we're using a locale-aware
2081  * comparison, we need to initialize a VarStringSortSupport object. Both
2082  * cases will make use of the temporary buffers we initialize here for
2083  * scratch space (and to detect requirement for BpChar semantics from
2084  * caller), and the abbreviation case requires additional state.
2085  */
2086  if (abbreviate || !collate_c)
2087  {
2088  sss = palloc(sizeof(VarStringSortSupport));
2089  sss->buf1 = palloc(TEXTBUFLEN);
2090  sss->buflen1 = TEXTBUFLEN;
2091  sss->buf2 = palloc(TEXTBUFLEN);
2092  sss->buflen2 = TEXTBUFLEN;
2093  /* Start with invalid values */
2094  sss->last_len1 = -1;
2095  sss->last_len2 = -1;
2096  /* Initialize */
2097  sss->last_returned = 0;
2098  sss->locale = locale;
2099 
2100  /*
2101  * To avoid somehow confusing a strxfrm() blob and an original string,
2102  * constantly keep track of the variety of data that buf1 and buf2
2103  * currently contain.
2104  *
2105  * Comparisons may be interleaved with conversion calls. Frequently,
2106  * conversions and comparisons are batched into two distinct phases,
2107  * but the correctness of caching cannot hinge upon this. For
2108  * comparison caching, buffer state is only trusted if cache_blob is
2109  * found set to false, whereas strxfrm() caching only trusts the state
2110  * when cache_blob is found set to true.
2111  *
2112  * Arbitrarily initialize cache_blob to true.
2113  */
2114  sss->cache_blob = true;
2115  sss->collate_c = collate_c;
2116  sss->typid = typid;
2117  ssup->ssup_extra = sss;
2118 
2119  /*
2120  * If possible, plan to use the abbreviated keys optimization. The
2121  * core code may switch back to authoritative comparator should
2122  * abbreviation be aborted.
2123  */
2124  if (abbreviate)
2125  {
2126  sss->prop_card = 0.20;
2127  initHyperLogLog(&sss->abbr_card, 10);
2128  initHyperLogLog(&sss->full_card, 10);
2129  ssup->abbrev_full_comparator = ssup->comparator;
2130  ssup->comparator = varstrcmp_abbrev;
2133  }
2134  }
2135 }
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2178
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2445
static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2420
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2141
static int varlenafastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2223
void initHyperLogLog(hyperLogLogState *cState, uint8 bwidth)
Definition: hyperloglog.c:65
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1356
static void check_collation_set(Oid collid)
Definition: varlena.c:1465
static int namefastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2254
int(* comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:106
hyperLogLogState abbr_card
Definition: varlena.c:89
pg_locale_t locale
Definition: varlena.c:92
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:1503
int GetDatabaseEncoding(void)
Definition: mbutils.c:1151
hyperLogLogState full_card
Definition: varlena.c:90
bool(* abbrev_abort)(int memtupcount, SortSupport ssup)
Definition: sortsupport.h:182
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2685
#define TEXTBUFLEN
Definition: varlena.c:99
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:2211