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 collid, bool bpchar)
 
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)
 
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 3912 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().

3914 {
3915  text *ret_text;
3916  regex_t *re = (regex_t *) regexp;
3917  int src_text_len = VARSIZE_ANY_EXHDR(src_text);
3920  pg_wchar *data;
3921  size_t data_len;
3922  int search_start;
3923  int data_pos;
3924  char *start_ptr;
3925  bool have_escape;
3926 
3927  initStringInfo(&buf);
3928 
3929  /* Convert data string to wide characters. */
3930  data = (pg_wchar *) palloc((src_text_len + 1) * sizeof(pg_wchar));
3931  data_len = pg_mb2wchar_with_len(VARDATA_ANY(src_text), data, src_text_len);
3932 
3933  /* Check whether replace_text has escape char. */
3934  have_escape = check_replace_text_has_escape_char(replace_text);
3935 
3936  /* start_ptr points to the data_pos'th character of src_text */
3937  start_ptr = (char *) VARDATA_ANY(src_text);
3938  data_pos = 0;
3939 
3940  search_start = 0;
3941  while (search_start <= data_len)
3942  {
3943  int regexec_result;
3944 
3946 
3947  regexec_result = pg_regexec(re,
3948  data,
3949  data_len,
3950  search_start,
3951  NULL, /* no details */
3953  pmatch,
3954  0);
3955 
3956  if (regexec_result == REG_NOMATCH)
3957  break;
3958 
3959  if (regexec_result != REG_OKAY)
3960  {
3961  char errMsg[100];
3962 
3964  pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
3965  ereport(ERROR,
3966  (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
3967  errmsg("regular expression failed: %s", errMsg)));
3968  }
3969 
3970  /*
3971  * Copy the text to the left of the match position. Note we are given
3972  * character not byte indexes.
3973  */
3974  if (pmatch[0].rm_so - data_pos > 0)
3975  {
3976  int chunk_len;
3977 
3978  chunk_len = charlen_to_bytelen(start_ptr,
3979  pmatch[0].rm_so - data_pos);
3980  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
3981 
3982  /*
3983  * Advance start_ptr over that text, to avoid multiple rescans of
3984  * it if the replace_text contains multiple back-references.
3985  */
3986  start_ptr += chunk_len;
3987  data_pos = pmatch[0].rm_so;
3988  }
3989 
3990  /*
3991  * Copy the replace_text. Process back references when the
3992  * replace_text has escape characters.
3993  */
3994  if (have_escape)
3995  appendStringInfoRegexpSubstr(&buf, replace_text, pmatch,
3996  start_ptr, data_pos);
3997  else
3998  appendStringInfoText(&buf, replace_text);
3999 
4000  /* Advance start_ptr and data_pos over the matched text. */
4001  start_ptr += charlen_to_bytelen(start_ptr,
4002  pmatch[0].rm_eo - data_pos);
4003  data_pos = pmatch[0].rm_eo;
4004 
4005  /*
4006  * When global option is off, replace the first instance only.
4007  */
4008  if (!glob)
4009  break;
4010 
4011  /*
4012  * Advance search position. Normally we start the next search at the
4013  * end of the previous match; but if the match was of zero length, we
4014  * have to advance by one character, or we'd just find the same match
4015  * again.
4016  */
4017  search_start = data_pos;
4018  if (pmatch[0].rm_so == pmatch[0].rm_eo)
4019  search_start++;
4020  }
4021 
4022  /*
4023  * Copy the text to the right of the last match.
4024  */
4025  if (data_pos < data_len)
4026  {
4027  int chunk_len;
4028 
4029  chunk_len = ((char *) src_text + VARSIZE_ANY(src_text)) - start_ptr;
4030  appendBinaryStringInfo(&buf, start_ptr, chunk_len);
4031  }
4032 
4033  ret_text = cstring_to_text_with_len(buf.data, buf.len);
4034  pfree(buf.data);
4035  pfree(data);
4036 
4037  return ret_text;
4038 }
#define VARDATA_ANY(PTR)
Definition: postgres.h:347
regoff_t rm_so
Definition: regex.h:85
int errcode(int sqlerrcode)
Definition: elog.c:575
regoff_t rm_eo
Definition: regex.h:86
void pfree(void *pointer)
Definition: mcxt.c:949
#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:3774
static int charlen_to_bytelen(const char *p, int n)
Definition: varlena.c:735
static void appendStringInfoText(StringInfo str, const text *t)
Definition: varlena.c:3677
static char * buf
Definition: pg_test_fsync.c:67
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:161
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:122
unsigned int pg_wchar
Definition: mbprint.c:31
void initStringInfo(StringInfo str)
Definition: stringinfo.c:46
#define REGEXP_REPLACE_BACKREF_CNT
Definition: varlena.c:3901
#define VARSIZE_ANY(PTR)
Definition: postgres.h:334
int pg_mb2wchar_with_len(const char *from, pg_wchar *to, int len)
Definition: mbutils.c:723
static void appendStringInfoRegexpSubstr(StringInfo str, text *replace_text, regmatch_t *pmatch, char *start_ptr, int data_pos)
Definition: varlena.c:3807
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:340
void * palloc(Size size)
Definition: mcxt.c:848
int errmsg(const char *fmt,...)
Definition: elog.c:797
#define REG_NOMATCH
Definition: regex.h:138
Definition: c.h:487
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:98
void appendBinaryStringInfo(StringInfo str, const char *data, int datalen)
Definition: stringinfo.c:208
Definition: regex.h:55

◆ SplitDirectoriesString()

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

Definition at line 3391 of file varlena.c.

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

Referenced by load_libraries(), and PostmasterMain().

3393 {
3394  char *nextp = rawstring;
3395  bool done = false;
3396 
3397  *namelist = NIL;
3398 
3399  while (scanner_isspace(*nextp))
3400  nextp++; /* skip leading whitespace */
3401 
3402  if (*nextp == '\0')
3403  return true; /* allow empty string */
3404 
3405  /* At the top of the loop, we are at start of a new directory. */
3406  do
3407  {
3408  char *curname;
3409  char *endp;
3410 
3411  if (*nextp == '"')
3412  {
3413  /* Quoted name --- collapse quote-quote pairs */
3414  curname = nextp + 1;
3415  for (;;)
3416  {
3417  endp = strchr(nextp + 1, '"');
3418  if (endp == NULL)
3419  return false; /* mismatched quotes */
3420  if (endp[1] != '"')
3421  break; /* found end of quoted name */
3422  /* Collapse adjacent quotes into one quote, and look again */
3423  memmove(endp, endp + 1, strlen(endp));
3424  nextp = endp;
3425  }
3426  /* endp now points at the terminating quote */
3427  nextp = endp + 1;
3428  }
3429  else
3430  {
3431  /* Unquoted name --- extends to separator or end of string */
3432  curname = endp = nextp;
3433  while (*nextp && *nextp != separator)
3434  {
3435  /* trailing whitespace should not be included in name */
3436  if (!scanner_isspace(*nextp))
3437  endp = nextp + 1;
3438  nextp++;
3439  }
3440  if (curname == endp)
3441  return false; /* empty unquoted name not allowed */
3442  }
3443 
3444  while (scanner_isspace(*nextp))
3445  nextp++; /* skip trailing whitespace */
3446 
3447  if (*nextp == separator)
3448  {
3449  nextp++;
3450  while (scanner_isspace(*nextp))
3451  nextp++; /* skip leading whitespace for next */
3452  /* we expect another name, so done remains false */
3453  }
3454  else if (*nextp == '\0')
3455  done = true;
3456  else
3457  return false; /* invalid syntax */
3458 
3459  /* Now safe to overwrite separator with a null */
3460  *endp = '\0';
3461 
3462  /* Truncate path if it's overlength */
3463  if (strlen(curname) >= MAXPGPATH)
3464  curname[MAXPGPATH - 1] = '\0';
3465 
3466  /*
3467  * Finished isolating current name --- add it to list
3468  */
3469  curname = pstrdup(curname);
3470  canonicalize_path(curname);
3471  *namelist = lappend(*namelist, curname);
3472 
3473  /* Loop back if we didn't reach end of string */
3474  } while (!done);
3475 
3476  return true;
3477 }
#define NIL
Definition: pg_list.h:69
char * pstrdup(const char *in)
Definition: mcxt.c:1076
void canonicalize_path(char *path)
Definition: path.c:254
#define MAXPGPATH
#define memmove(d, s, c)
Definition: c.h:1045
List * lappend(List *list, void *datum)
Definition: list.c:128
bool scanner_isspace(char ch)
Definition: scansup.c:221

◆ SplitIdentifierString()

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

Definition at line 3264 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().

3266 {
3267  char *nextp = rawstring;
3268  bool done = false;
3269 
3270  *namelist = NIL;
3271 
3272  while (scanner_isspace(*nextp))
3273  nextp++; /* skip leading whitespace */
3274 
3275  if (*nextp == '\0')
3276  return true; /* allow empty string */
3277 
3278  /* At the top of the loop, we are at start of a new identifier. */
3279  do
3280  {
3281  char *curname;
3282  char *endp;
3283 
3284  if (*nextp == '"')
3285  {
3286  /* Quoted name --- collapse quote-quote pairs, no downcasing */
3287  curname = nextp + 1;
3288  for (;;)
3289  {
3290  endp = strchr(nextp + 1, '"');
3291  if (endp == NULL)
3292  return false; /* mismatched quotes */
3293  if (endp[1] != '"')
3294  break; /* found end of quoted name */
3295  /* Collapse adjacent quotes into one quote, and look again */
3296  memmove(endp, endp + 1, strlen(endp));
3297  nextp = endp;
3298  }
3299  /* endp now points at the terminating quote */
3300  nextp = endp + 1;
3301  }
3302  else
3303  {
3304  /* Unquoted name --- extends to separator or whitespace */
3305  char *downname;
3306  int len;
3307 
3308  curname = nextp;
3309  while (*nextp && *nextp != separator &&
3310  !scanner_isspace(*nextp))
3311  nextp++;
3312  endp = nextp;
3313  if (curname == nextp)
3314  return false; /* empty unquoted name not allowed */
3315 
3316  /*
3317  * Downcase the identifier, using same code as main lexer does.
3318  *
3319  * XXX because we want to overwrite the input in-place, we cannot
3320  * support a downcasing transformation that increases the string
3321  * length. This is not a problem given the current implementation
3322  * of downcase_truncate_identifier, but we'll probably have to do
3323  * something about this someday.
3324  */
3325  len = endp - curname;
3326  downname = downcase_truncate_identifier(curname, len, false);
3327  Assert(strlen(downname) <= len);
3328  strncpy(curname, downname, len); /* strncpy is required here */
3329  pfree(downname);
3330  }
3331 
3332  while (scanner_isspace(*nextp))
3333  nextp++; /* skip trailing whitespace */
3334 
3335  if (*nextp == separator)
3336  {
3337  nextp++;
3338  while (scanner_isspace(*nextp))
3339  nextp++; /* skip leading whitespace for next */
3340  /* we expect another name, so done remains false */
3341  }
3342  else if (*nextp == '\0')
3343  done = true;
3344  else
3345  return false; /* invalid syntax */
3346 
3347  /* Now safe to overwrite separator with a null */
3348  *endp = '\0';
3349 
3350  /* Truncate name if it's overlength */
3351  truncate_identifier(curname, strlen(curname), false);
3352 
3353  /*
3354  * Finished isolating current name --- add it to list
3355  */
3356  *namelist = lappend(*namelist, curname);
3357 
3358  /* Loop back if we didn't reach end of string */
3359  } while (!done);
3360 
3361  return true;
3362 }
#define NIL
Definition: pg_list.h:69
char * downcase_truncate_identifier(const char *ident, int len, bool warn)
Definition: scansup.c:131
void truncate_identifier(char *ident, int len, bool warn)
Definition: scansup.c:187
void pfree(void *pointer)
Definition: mcxt.c:949
#define memmove(d, s, c)
Definition: c.h:1045
List * lappend(List *list, void *datum)
Definition: list.c:128
bool scanner_isspace(char ch)
Definition: scansup.c:221
#define Assert(condition)
Definition: c.h:670

◆ textToQualifiedNameList()

List* textToQualifiedNameList ( text textval)

Definition at line 3206 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().

3207 {
3208  char *rawname;
3209  List *result = NIL;
3210  List *namelist;
3211  ListCell *l;
3212 
3213  /* Convert to C string (handles possible detoasting). */
3214  /* Note we rely on being able to modify rawname below. */
3215  rawname = text_to_cstring(textval);
3216 
3217  if (!SplitIdentifierString(rawname, '.', &namelist))
3218  ereport(ERROR,
3219  (errcode(ERRCODE_INVALID_NAME),
3220  errmsg("invalid name syntax")));
3221 
3222  if (namelist == NIL)
3223  ereport(ERROR,
3224  (errcode(ERRCODE_INVALID_NAME),
3225  errmsg("invalid name syntax")));
3226 
3227  foreach(l, namelist)
3228  {
3229  char *curname = (char *) lfirst(l);
3230 
3231  result = lappend(result, makeString(pstrdup(curname)));
3232  }
3233 
3234  pfree(rawname);
3235  list_free(namelist);
3236 
3237  return result;
3238 }
Value * makeString(char *str)
Definition: value.c:53
#define NIL
Definition: pg_list.h:69
char * pstrdup(const char *in)
Definition: mcxt.c:1076
int errcode(int sqlerrcode)
Definition: elog.c:575
void pfree(void *pointer)
Definition: mcxt.c:949
#define ERROR
Definition: elog.h:43
bool SplitIdentifierString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3264
#define ereport(elevel, rest)
Definition: elog.h:122
List * lappend(List *list, void *datum)
Definition: list.c:128
#define lfirst(lc)
Definition: pg_list.h:106
char * text_to_cstring(const text *t)
Definition: varlena.c:182
int errmsg(const char *fmt,...)
Definition: elog.c:797
void list_free(List *list)
Definition: list.c:1133
Definition: pg_list.h:45

◆ varstr_cmp()

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

Definition at line 1382 of file varlena.c.

References COLLPROVIDER_ICU, COLLPROVIDER_LIBC, DEFAULT_COLLATION_OID, elog, ereport, errcode(), errhint(), errmsg(), ERROR, GetDatabaseEncoding(), pg_locale_struct::info, lc_collate_is_c(), Min, OidIsValid, 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(), bpcharge(), bpchargt(), bpcharle(), bpcharlt(), citextcmp(), compareJsonbScalarValue(), gin_compare_jsonb(), make_greater_string(), spg_text_leaf_consistent(), and text_cmp().

1383 {
1384  int result;
1385 
1386  /*
1387  * Unfortunately, there is no strncoll(), so in the non-C locale case we
1388  * have to do some memory copying. This turns out to be significantly
1389  * slower, so we optimize the case where LC_COLLATE is C. We also try to
1390  * optimize relatively-short strings by avoiding palloc/pfree overhead.
1391  */
1392  if (lc_collate_is_c(collid))
1393  {
1394  result = memcmp(arg1, arg2, Min(len1, len2));
1395  if ((result == 0) && (len1 != len2))
1396  result = (len1 < len2) ? -1 : 1;
1397  }
1398  else
1399  {
1400  char a1buf[TEXTBUFLEN];
1401  char a2buf[TEXTBUFLEN];
1402  char *a1p,
1403  *a2p;
1404  pg_locale_t mylocale = 0;
1405 
1406  if (collid != DEFAULT_COLLATION_OID)
1407  {
1408  if (!OidIsValid(collid))
1409  {
1410  /*
1411  * This typically means that the parser could not resolve a
1412  * conflict of implicit collations, so report it that way.
1413  */
1414  ereport(ERROR,
1415  (errcode(ERRCODE_INDETERMINATE_COLLATION),
1416  errmsg("could not determine which collation to use for string comparison"),
1417  errhint("Use the COLLATE clause to set the collation explicitly.")));
1418  }
1419  mylocale = pg_newlocale_from_collation(collid);
1420  }
1421 
1422  /*
1423  * memcmp() can't tell us which of two unequal strings sorts first,
1424  * but it's a cheap way to tell if they're equal. Testing shows that
1425  * memcmp() followed by strcoll() is only trivially slower than
1426  * strcoll() by itself, so we don't lose much if this doesn't work out
1427  * very often, and if it does - for example, because there are many
1428  * equal strings in the input - then we win big by avoiding expensive
1429  * collation-aware comparisons.
1430  */
1431  if (len1 == len2 && memcmp(arg1, arg2, len1) == 0)
1432  return 0;
1433 
1434 #ifdef WIN32
1435  /* Win32 does not have UTF-8, so we need to map to UTF-16 */
1436  if (GetDatabaseEncoding() == PG_UTF8
1437  && (!mylocale || mylocale->provider == COLLPROVIDER_LIBC))
1438  {
1439  int a1len;
1440  int a2len;
1441  int r;
1442 
1443  if (len1 >= TEXTBUFLEN / 2)
1444  {
1445  a1len = len1 * 2 + 2;
1446  a1p = palloc(a1len);
1447  }
1448  else
1449  {
1450  a1len = TEXTBUFLEN;
1451  a1p = a1buf;
1452  }
1453  if (len2 >= TEXTBUFLEN / 2)
1454  {
1455  a2len = len2 * 2 + 2;
1456  a2p = palloc(a2len);
1457  }
1458  else
1459  {
1460  a2len = TEXTBUFLEN;
1461  a2p = a2buf;
1462  }
1463 
1464  /* stupid Microsloth API does not work for zero-length input */
1465  if (len1 == 0)
1466  r = 0;
1467  else
1468  {
1469  r = MultiByteToWideChar(CP_UTF8, 0, arg1, len1,
1470  (LPWSTR) a1p, a1len / 2);
1471  if (!r)
1472  ereport(ERROR,
1473  (errmsg("could not convert string to UTF-16: error code %lu",
1474  GetLastError())));
1475  }
1476  ((LPWSTR) a1p)[r] = 0;
1477 
1478  if (len2 == 0)
1479  r = 0;
1480  else
1481  {
1482  r = MultiByteToWideChar(CP_UTF8, 0, arg2, len2,
1483  (LPWSTR) a2p, a2len / 2);
1484  if (!r)
1485  ereport(ERROR,
1486  (errmsg("could not convert string to UTF-16: error code %lu",
1487  GetLastError())));
1488  }
1489  ((LPWSTR) a2p)[r] = 0;
1490 
1491  errno = 0;
1492 #ifdef HAVE_LOCALE_T
1493  if (mylocale)
1494  result = wcscoll_l((LPWSTR) a1p, (LPWSTR) a2p, mylocale->info.lt);
1495  else
1496 #endif
1497  result = wcscoll((LPWSTR) a1p, (LPWSTR) a2p);
1498  if (result == 2147483647) /* _NLSCMPERROR; missing from mingw
1499  * headers */
1500  ereport(ERROR,
1501  (errmsg("could not compare Unicode strings: %m")));
1502 
1503  /*
1504  * In some locales wcscoll() can claim that nonidentical strings
1505  * are equal. Believing that would be bad news for a number of
1506  * reasons, so we follow Perl's lead and sort "equal" strings
1507  * according to strcmp (on the UTF-8 representation).
1508  */
1509  if (result == 0)
1510  {
1511  result = memcmp(arg1, arg2, Min(len1, len2));
1512  if ((result == 0) && (len1 != len2))
1513  result = (len1 < len2) ? -1 : 1;
1514  }
1515 
1516  if (a1p != a1buf)
1517  pfree(a1p);
1518  if (a2p != a2buf)
1519  pfree(a2p);
1520 
1521  return result;
1522  }
1523 #endif /* WIN32 */
1524 
1525  if (len1 >= TEXTBUFLEN)
1526  a1p = (char *) palloc(len1 + 1);
1527  else
1528  a1p = a1buf;
1529  if (len2 >= TEXTBUFLEN)
1530  a2p = (char *) palloc(len2 + 1);
1531  else
1532  a2p = a2buf;
1533 
1534  memcpy(a1p, arg1, len1);
1535  a1p[len1] = '\0';
1536  memcpy(a2p, arg2, len2);
1537  a2p[len2] = '\0';
1538 
1539  if (mylocale)
1540  {
1541  if (mylocale->provider == COLLPROVIDER_ICU)
1542  {
1543 #ifdef USE_ICU
1544 #ifdef HAVE_UCOL_STRCOLLUTF8
1545  if (GetDatabaseEncoding() == PG_UTF8)
1546  {
1547  UErrorCode status;
1548 
1549  status = U_ZERO_ERROR;
1550  result = ucol_strcollUTF8(mylocale->info.icu.ucol,
1551  arg1, len1,
1552  arg2, len2,
1553  &status);
1554  if (U_FAILURE(status))
1555  ereport(ERROR,
1556  (errmsg("collation failed: %s", u_errorName(status))));
1557  }
1558  else
1559 #endif
1560  {
1561  int32_t ulen1,
1562  ulen2;
1563  UChar *uchar1,
1564  *uchar2;
1565 
1566  ulen1 = icu_to_uchar(&uchar1, arg1, len1);
1567  ulen2 = icu_to_uchar(&uchar2, arg2, len2);
1568 
1569  result = ucol_strcoll(mylocale->info.icu.ucol,
1570  uchar1, ulen1,
1571  uchar2, ulen2);
1572 
1573  pfree(uchar1);
1574  pfree(uchar2);
1575  }
1576 #else /* not USE_ICU */
1577  /* shouldn't happen */
1578  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1579 #endif /* not USE_ICU */
1580  }
1581  else
1582  {
1583 #ifdef HAVE_LOCALE_T
1584  result = strcoll_l(a1p, a2p, mylocale->info.lt);
1585 #else
1586  /* shouldn't happen */
1587  elog(ERROR, "unsupported collprovider: %c", mylocale->provider);
1588 #endif
1589  }
1590  }
1591  else
1592  result = strcoll(a1p, a2p);
1593 
1594  /*
1595  * In some locales strcoll() can claim that nonidentical strings are
1596  * equal. Believing that would be bad news for a number of reasons,
1597  * so we follow Perl's lead and sort "equal" strings according to
1598  * strcmp().
1599  */
1600  if (result == 0)
1601  result = strcmp(a1p, a2p);
1602 
1603  if (a1p != a1buf)
1604  pfree(a1p);
1605  if (a2p != a2buf)
1606  pfree(a2p);
1607  }
1608 
1609  return result;
1610 }
#define COLLPROVIDER_ICU
Definition: pg_collation.h:85
int errhint(const char *fmt,...)
Definition: elog.c:987
#define Min(x, y)
Definition: c.h:802
union pg_locale_struct::@120 info
int errcode(int sqlerrcode)
Definition: elog.c:575
#define OidIsValid(objectId)
Definition: c.h:576
void pfree(void *pointer)
Definition: mcxt.c:949
#define ERROR
Definition: elog.h:43
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1128
#define strcoll_l
Definition: win32_port.h:397
#define DEFAULT_COLLATION_OID
Definition: pg_collation.h:75
#define ereport(elevel, rest)
Definition: elog.h:122
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1275
int GetDatabaseEncoding(void)
Definition: mbutils.c:1004
#define wcscoll_l
Definition: win32_port.h:399
#define TEXTBUFLEN
Definition: varlena.c:83
void * palloc(Size size)
Definition: mcxt.c:848
int errmsg(const char *fmt,...)
Definition: elog.c:797
int int32_t
Definition: crypt.c:108
#define elog
Definition: elog.h:219
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:225
#define COLLPROVIDER_LIBC
Definition: pg_collation.h:86

◆ 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:802
int errcode(int sqlerrcode)
Definition: elog.c:575
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:794
#define STOP_COLUMN
#define ERROR
Definition: elog.h:43
#define START_COLUMN
#define ereport(elevel, rest)
Definition: elog.h:122
#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:5511
int pg_mblen(const char *mbstr)
Definition: mbutils.c:760
void * palloc(Size size)
Definition: mcxt.c:848
int errmsg(const char *fmt,...)
Definition: elog.c:797
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  collid,
bool  bpchar 
)

Definition at line 1808 of file varlena.c.

References VarStringSortSupport::abbr_card, SortSupportData::abbrev_abort, SortSupportData::abbrev_converter, SortSupportData::abbrev_full_comparator, SortSupportData::abbreviate, VarStringSortSupport::bpchar, bpchar(), bpcharfastcmp_c(), VarStringSortSupport::buf1, VarStringSortSupport::buf2, VarStringSortSupport::buflen1, VarStringSortSupport::buflen2, VarStringSortSupport::cache_blob, VarStringSortSupport::collate_c, COLLPROVIDER_ICU, SortSupportData::comparator, DEFAULT_COLLATION_OID, ereport, errcode(), errhint(), errmsg(), ERROR, VarStringSortSupport::full_card, GetDatabaseEncoding(), initHyperLogLog(), VarStringSortSupport::last_len1, VarStringSortSupport::last_len2, VarStringSortSupport::last_returned, lc_collate_is_c(), VarStringSortSupport::locale, locale, OidIsValid, palloc(), pg_newlocale_from_collation(), PG_UTF8, VarStringSortSupport::prop_card, pg_locale_struct::provider, SortSupportData::ssup_extra, TEXTBUFLEN, varstr_abbrev_abort(), varstr_abbrev_convert(), varstrcmp_abbrev(), varstrfastcmp_c(), and varstrfastcmp_locale().

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

1809 {
1810  bool abbreviate = ssup->abbreviate;
1811  bool collate_c = false;
1812  VarStringSortSupport *sss;
1813  pg_locale_t locale = 0;
1814 
1815  /*
1816  * If possible, set ssup->comparator to a function which can be used to
1817  * directly compare two datums. If we can do this, we'll avoid the
1818  * overhead of a trip through the fmgr layer for every comparison, which
1819  * can be substantial.
1820  *
1821  * Most typically, we'll set the comparator to varstrfastcmp_locale, which
1822  * uses strcoll() to perform comparisons and knows about the special
1823  * requirements of BpChar callers. However, if LC_COLLATE = C, we can
1824  * make things quite a bit faster with varstrfastcmp_c or bpcharfastcmp_c,
1825  * both of which use memcmp() rather than strcoll().
1826  */
1827  if (lc_collate_is_c(collid))
1828  {
1829  if (!bpchar)
1830  ssup->comparator = varstrfastcmp_c;
1831  else
1832  ssup->comparator = bpcharfastcmp_c;
1833 
1834  collate_c = true;
1835  }
1836  else
1837  {
1838  /*
1839  * We need a collation-sensitive comparison. To make things faster,
1840  * we'll figure out the collation based on the locale id and cache the
1841  * result.
1842  */
1843  if (collid != DEFAULT_COLLATION_OID)
1844  {
1845  if (!OidIsValid(collid))
1846  {
1847  /*
1848  * This typically means that the parser could not resolve a
1849  * conflict of implicit collations, so report it that way.
1850  */
1851  ereport(ERROR,
1852  (errcode(ERRCODE_INDETERMINATE_COLLATION),
1853  errmsg("could not determine which collation to use for string comparison"),
1854  errhint("Use the COLLATE clause to set the collation explicitly.")));
1855  }
1856  locale = pg_newlocale_from_collation(collid);
1857  }
1858 
1859  /*
1860  * There is a further exception on Windows. When the database
1861  * encoding is UTF-8 and we are not using the C collation, complex
1862  * hacks are required. We don't currently have a comparator that
1863  * handles that case, so we fall back on the slow method of having the
1864  * sort code invoke bttextcmp() (in the case of text) via the fmgr
1865  * trampoline. ICU locales work just the same on Windows, however.
1866  */
1867 #ifdef WIN32
1868  if (GetDatabaseEncoding() == PG_UTF8 &&
1869  !(locale && locale->provider == COLLPROVIDER_ICU))
1870  return;
1871 #endif
1872 
1874  }
1875 
1876  /*
1877  * Unfortunately, it seems that abbreviation for non-C collations is
1878  * broken on many common platforms; testing of multiple versions of glibc
1879  * reveals that, for many locales, strcoll() and strxfrm() do not return
1880  * consistent results, which is fatal to this optimization. While no
1881  * other libc other than Cygwin has so far been shown to have a problem,
1882  * we take the conservative course of action for right now and disable
1883  * this categorically. (Users who are certain this isn't a problem on
1884  * their system can define TRUST_STRXFRM.)
1885  *
1886  * Even apart from the risk of broken locales, it's possible that there
1887  * are platforms where the use of abbreviated keys should be disabled at
1888  * compile time. Having only 4 byte datums could make worst-case
1889  * performance drastically more likely, for example. Moreover, macOS's
1890  * strxfrm() implementation is known to not effectively concentrate a
1891  * significant amount of entropy from the original string in earlier
1892  * transformed blobs. It's possible that other supported platforms are
1893  * similarly encumbered. So, if we ever get past disabling this
1894  * categorically, we may still want or need to disable it for particular
1895  * platforms.
1896  */
1897 #ifndef TRUST_STRXFRM
1898  if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
1899  abbreviate = false;
1900 #endif
1901 
1902  /*
1903  * If we're using abbreviated keys, or if we're using a locale-aware
1904  * comparison, we need to initialize a StringSortSupport object. Both
1905  * cases will make use of the temporary buffers we initialize here for
1906  * scratch space (and to detect requirement for BpChar semantics from
1907  * caller), and the abbreviation case requires additional state.
1908  */
1909  if (abbreviate || !collate_c)
1910  {
1911  sss = palloc(sizeof(VarStringSortSupport));
1912  sss->buf1 = palloc(TEXTBUFLEN);
1913  sss->buflen1 = TEXTBUFLEN;
1914  sss->buf2 = palloc(TEXTBUFLEN);
1915  sss->buflen2 = TEXTBUFLEN;
1916  /* Start with invalid values */
1917  sss->last_len1 = -1;
1918  sss->last_len2 = -1;
1919  /* Initialize */
1920  sss->last_returned = 0;
1921  sss->locale = locale;
1922 
1923  /*
1924  * To avoid somehow confusing a strxfrm() blob and an original string,
1925  * constantly keep track of the variety of data that buf1 and buf2
1926  * currently contain.
1927  *
1928  * Comparisons may be interleaved with conversion calls. Frequently,
1929  * conversions and comparisons are batched into two distinct phases,
1930  * but the correctness of caching cannot hinge upon this. For
1931  * comparison caching, buffer state is only trusted if cache_blob is
1932  * found set to false, whereas strxfrm() caching only trusts the state
1933  * when cache_blob is found set to true.
1934  *
1935  * Arbitrarily initialize cache_blob to true.
1936  */
1937  sss->cache_blob = true;
1938  sss->collate_c = collate_c;
1939  sss->bpchar = bpchar;
1940  ssup->ssup_extra = sss;
1941 
1942  /*
1943  * If possible, plan to use the abbreviated keys optimization. The
1944  * core code may switch back to authoritative comparator should
1945  * abbreviation be aborted.
1946  */
1947  if (abbreviate)
1948  {
1949  sss->prop_card = 0.20;
1950  initHyperLogLog(&sss->abbr_card, 10);
1951  initHyperLogLog(&sss->full_card, 10);
1952  ssup->abbrev_full_comparator = ssup->comparator;
1953  ssup->comparator = varstrcmp_abbrev;
1956  }
1957  }
1958 }
static int varstrfastcmp_locale(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2034
#define COLLPROVIDER_ICU
Definition: pg_collation.h:85
static int bpcharfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2001
int errhint(const char *fmt,...)
Definition: elog.c:987
static Datum varstr_abbrev_convert(Datum original, SortSupport ssup)
Definition: varlena.c:2237
static int varstrcmp_abbrev(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:2212
int errcode(int sqlerrcode)
Definition: elog.c:575
#define OidIsValid(objectId)
Definition: c.h:576
static int varstrfastcmp_c(Datum x, Datum y, SortSupport ssup)
Definition: varlena.c:1964
void initHyperLogLog(hyperLogLogState *cState, uint8 bwidth)
Definition: hyperloglog.c:65
#define ERROR
Definition: elog.h:43
bool lc_collate_is_c(Oid collation)
Definition: pg_locale.c:1128
int(* comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:107
hyperLogLogState abbr_card
Definition: varlena.c:73
pg_locale_t locale
Definition: varlena.c:76
int(* abbrev_full_comparator)(Datum x, Datum y, SortSupport ssup)
Definition: sortsupport.h:192
#define DEFAULT_COLLATION_OID
Definition: pg_collation.h:75
Datum(* abbrev_converter)(Datum original, SortSupport ssup)
Definition: sortsupport.h:173
void * ssup_extra
Definition: sortsupport.h:87
#define ereport(elevel, rest)
Definition: elog.h:122
pg_locale_t pg_newlocale_from_collation(Oid collid)
Definition: pg_locale.c:1275
int GetDatabaseEncoding(void)
Definition: mbutils.c:1004
hyperLogLogState full_card
Definition: varlena.c:74
bool(* abbrev_abort)(int memtupcount, SortSupport ssup)
Definition: sortsupport.h:183
static bool varstr_abbrev_abort(int memtupcount, SortSupport ssup)
Definition: varlena.c:2477
#define TEXTBUFLEN
Definition: varlena.c:83
void * palloc(Size size)
Definition: mcxt.c:848
int errmsg(const char *fmt,...)
Definition: elog.c:797
static char * locale
Definition: initdb.c:124
Datum bpchar(PG_FUNCTION_ARGS)
Definition: varchar.c:267