2
0
mirror of https://git.postgresql.org/git/postgresql.git synced 2025-01-06 15:24:56 +08:00

Optimization for single byte encodings.

This commit is contained in:
Tatsuo Ishii 2001-10-04 02:15:47 +00:00
parent a2a0f7b79b
commit 219bd1e4fb
2 changed files with 433 additions and 314 deletions
src/backend/utils/adt

View File

@ -11,7 +11,7 @@
* Portions Copyright (c) 1994, Regents of the University of California
*
* IDENTIFICATION
* $Header: /cvsroot/pgsql/src/backend/utils/adt/like.c,v 1.46 2001/09/14 17:46:40 momjian Exp $
* $Header: /cvsroot/pgsql/src/backend/utils/adt/like.c,v 1.47 2001/10/04 02:15:47 ishii Exp $
*
*-------------------------------------------------------------------------
*/
@ -36,9 +36,15 @@ static int MatchTextIC(unsigned char *t, int tlen,
unsigned char *p, int plen);
static int MatchBytea(unsigned char *t, int tlen,
unsigned char *p, int plen);
static text *do_like_escape(text *, text *);
#ifdef MULTIBYTE
static int MBMatchText(unsigned char *t, int tlen,
unsigned char *p, int plen);
static int MBMatchTextIC(unsigned char *t, int tlen,
unsigned char *p, int plen);
static text *MB_do_like_escape(text *, text *);
/*--------------------
* Support routine for MatchText. Compares given multibyte streams
* as wide characters. If they match, returns 1 otherwise returns 0.
@ -115,16 +121,29 @@ iwchareq(unsigned char *p1, unsigned char *p2)
while (__l-- > 0) \
*(dst)++ = *(src)++; \
} while (0)
#else
#define MatchText MBMatchText
#define MatchTextIC MBMatchTextIC
#define do_like_escape MB_do_like_escape
#include "like_match.c"
#undef CHAREQ
#undef ICHAREQ
#undef NextChar
#undef CopyAdvChar
#undef MatchText
#undef MatchTextIC
#undef do_like_escape
#endif
#define CHAREQ(p1, p2) (*(p1) == *(p2))
#define ICHAREQ(p1, p2) (tolower(*(p1)) == tolower(*(p2)))
#define NextChar(p, plen) ((p)++, (plen)--)
#define CopyAdvChar(dst, src, srclen) (*(dst)++ = *(src)++, (srclen)--)
#endif
#define BYTEA_CHAREQ(p1, p2) (*(p1) == *(p2))
#define BYTEA_NextChar(p, plen) ((p)++, (plen)--)
#define BYTEA_CopyAdvChar(dst, src, srclen) (*(dst)++ = *(src)++, (srclen)--)
#include "like_match.c"
/*
* interface routines called by the function manager
@ -146,7 +165,14 @@ namelike(PG_FUNCTION_ARGS)
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchText(s, slen, p, plen) == LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchText(s, slen, p, plen) == LIKE_TRUE);
else
result = (MBMatchText(s, slen, p, plen) == LIKE_TRUE);
#else
result = (MatchText(s, slen, p, plen) == LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -167,7 +193,14 @@ namenlike(PG_FUNCTION_ARGS)
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchText(s, slen, p, plen) != LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchText(s, slen, p, plen) != LIKE_TRUE);
else
result = (MBMatchText(s, slen, p, plen) != LIKE_TRUE);
#else
result = (MatchText(s, slen, p, plen) != LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -182,13 +215,19 @@ textlike(PG_FUNCTION_ARGS)
*p;
int slen,
plen;
s = VARDATA(str);
slen = (VARSIZE(str) - VARHDRSZ);
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchText(s, slen, p, plen) == LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchText(s, slen, p, plen) == LIKE_TRUE);
else
result = (MBMatchText(s, slen, p, plen) == LIKE_TRUE);
#else
result = (MatchText(s, slen, p, plen) == LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -209,7 +248,14 @@ textnlike(PG_FUNCTION_ARGS)
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchText(s, slen, p, plen) != LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchText(s, slen, p, plen) != LIKE_TRUE);
else
result = (MBMatchText(s, slen, p, plen) != LIKE_TRUE);
#else
result = (MatchText(s, slen, p, plen) != LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -276,7 +322,14 @@ nameiclike(PG_FUNCTION_ARGS)
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchTextIC(s, slen, p, plen) == LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchTextIC(s, slen, p, plen) == LIKE_TRUE);
else
result = (MBMatchTextIC(s, slen, p, plen) == LIKE_TRUE);
#else
result = (MatchTextIC(s, slen, p, plen) == LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -297,7 +350,14 @@ nameicnlike(PG_FUNCTION_ARGS)
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchTextIC(s, slen, p, plen) != LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchTextIC(s, slen, p, plen) != LIKE_TRUE);
else
result = (MBMatchTextIC(s, slen, p, plen) != LIKE_TRUE);
#else
result = (MatchTextIC(s, slen, p, plen) != LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -318,7 +378,14 @@ texticlike(PG_FUNCTION_ARGS)
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchTextIC(s, slen, p, plen) == LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchTextIC(s, slen, p, plen) == LIKE_TRUE);
else
result = (MBMatchTextIC(s, slen, p, plen) == LIKE_TRUE);
#else
result = (MatchTextIC(s, slen, p, plen) == LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -339,7 +406,14 @@ texticnlike(PG_FUNCTION_ARGS)
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
result = (MatchTextIC(s, slen, p, plen) != LIKE_TRUE);
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = (MatchTextIC(s, slen, p, plen) != LIKE_TRUE);
else
result = (MBMatchTextIC(s, slen, p, plen) != LIKE_TRUE);
#else
result = (MatchTextIC(s, slen, p, plen) != LIKE_TRUE);
#endif
PG_RETURN_BOOL(result);
}
@ -354,90 +428,15 @@ like_escape(PG_FUNCTION_ARGS)
text *pat = PG_GETARG_TEXT_P(0);
text *esc = PG_GETARG_TEXT_P(1);
text *result;
unsigned char *p,
*e,
*r;
int plen,
elen;
bool afterescape;
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
e = VARDATA(esc);
elen = (VARSIZE(esc) - VARHDRSZ);
/*
* Worst-case pattern growth is 2x --- unlikely, but it's hardly worth
* trying to calculate the size more accurately than that.
*/
result = (text *) palloc(plen * 2 + VARHDRSZ);
r = VARDATA(result);
if (elen == 0)
{
/*
* No escape character is wanted. Double any backslashes in the
* pattern to make them act like ordinary characters.
*/
while (plen > 0)
{
if (*p == '\\')
*r++ = '\\';
CopyAdvChar(r, p, plen);
}
}
#ifdef MULTIBYTE
if (pg_database_encoding_max_length() == 1)
result = do_like_escape(pat, esc);
else
{
/*
* The specified escape must be only a single character.
*/
NextChar(e, elen);
if (elen != 0)
elog(ERROR, "ESCAPE string must be empty or one character");
e = VARDATA(esc);
/*
* If specified escape is '\', just copy the pattern as-is.
*/
if (*e == '\\')
{
memcpy(result, pat, VARSIZE(pat));
PG_RETURN_TEXT_P(result);
}
/*
* Otherwise, convert occurrences of the specified escape
* character to '\', and double occurrences of '\' --- unless they
* immediately follow an escape character!
*/
afterescape = false;
while (plen > 0)
{
if (CHAREQ(p, e) && !afterescape)
{
*r++ = '\\';
NextChar(p, plen);
afterescape = true;
}
else if (*p == '\\')
{
*r++ = '\\';
if (!afterescape)
*r++ = '\\';
NextChar(p, plen);
afterescape = false;
}
else
{
CopyAdvChar(r, p, plen);
afterescape = false;
}
}
}
VARATT_SIZEP(result) = r - ((unsigned char *) result);
result = MB_do_like_escape(pat, esc);
#else
result = do_like_escape(pat, esc);
#endif
PG_RETURN_TEXT_P(result);
}
@ -540,225 +539,6 @@ like_escape_bytea(PG_FUNCTION_ARGS)
PG_RETURN_BYTEA_P(result);
}
/*
** Originally written by Rich $alz, mirror!rs, Wed Nov 26 19:03:17 EST 1986.
** Rich $alz is now <rsalz@bbn.com>.
** Special thanks to Lars Mathiesen <thorinn@diku.dk> for the LABORT code.
**
** This code was shamelessly stolen from the "pql" code by myself and
** slightly modified :)
**
** All references to the word "star" were replaced by "percent"
** All references to the word "wild" were replaced by "like"
**
** All the nice shell RE matching stuff was replaced by just "_" and "%"
**
** As I don't have a copy of the SQL standard handy I wasn't sure whether
** to leave in the '\' escape character handling.
**
** Keith Parks. <keith@mtcc.demon.co.uk>
**
** SQL92 lets you specify the escape character by saying
** LIKE <pattern> ESCAPE <escape character>. We are a small operation
** so we force you to use '\'. - ay 7/95
**
** Now we have the like_escape() function that converts patterns with
** any specified escape character (or none at all) to the internal
** default escape character, which is still '\'. - tgl 9/2000
**
** The code is rewritten to avoid requiring null-terminated strings,
** which in turn allows us to leave out some memcpy() operations.
** This code should be faster and take less memory, but no promises...
** - thomas 2000-08-06
**
*/
/*--------------------
* Match text and p, return LIKE_TRUE, LIKE_FALSE, or LIKE_ABORT.
*
* LIKE_TRUE: they match
* LIKE_FALSE: they don't match
* LIKE_ABORT: not only don't they match, but the text is too short.
*
* If LIKE_ABORT is returned, then no suffix of the text can match the
* pattern either, so an upper-level % scan can stop scanning now.
*--------------------
*/
static int
MatchText(unsigned char *t, int tlen, unsigned char *p, int plen)
{
/* Fast path for match-everything pattern */
if ((plen == 1) && (*p == '%'))
return LIKE_TRUE;
while ((tlen > 0) && (plen > 0))
{
if (*p == '\\')
{
/* Next pattern char must match literally, whatever it is */
NextChar(p, plen);
if ((plen <= 0) || !CHAREQ(t, p))
return LIKE_FALSE;
}
else if (*p == '%')
{
/* %% is the same as % according to the SQL standard */
/* Advance past all %'s */
while ((plen > 0) && (*p == '%'))
NextChar(p, plen);
/* Trailing percent matches everything. */
if (plen <= 0)
return LIKE_TRUE;
/*
* Otherwise, scan for a text position at which we can match
* the rest of the pattern.
*/
while (tlen > 0)
{
/*
* Optimization to prevent most recursion: don't recurse
* unless first pattern char might match this text char.
*/
if (CHAREQ(t, p) || (*p == '\\') || (*p == '_'))
{
int matched = MatchText(t, tlen, p, plen);
if (matched != LIKE_FALSE)
return matched; /* TRUE or ABORT */
}
NextChar(t, tlen);
}
/*
* End of text with no match, so no point in trying later
* places to start matching this pattern.
*/
return LIKE_ABORT;
}
else if ((*p != '_') && !CHAREQ(t, p))
{
/*
* Not the single-character wildcard and no explicit match?
* Then time to quit...
*/
return LIKE_FALSE;
}
NextChar(t, tlen);
NextChar(p, plen);
}
if (tlen > 0)
return LIKE_FALSE; /* end of pattern, but not of text */
/* End of input string. Do we have matching pattern remaining? */
while ((plen > 0) && (*p == '%')) /* allow multiple %'s at end of
* pattern */
NextChar(p, plen);
if (plen <= 0)
return LIKE_TRUE;
/*
* End of text with no match, so no point in trying later places to
* start matching this pattern.
*/
return LIKE_ABORT;
} /* MatchText() */
/*
* Same as above, but ignore case
*/
static int
MatchTextIC(unsigned char *t, int tlen, unsigned char *p, int plen)
{
/* Fast path for match-everything pattern */
if ((plen == 1) && (*p == '%'))
return LIKE_TRUE;
while ((tlen > 0) && (plen > 0))
{
if (*p == '\\')
{
/* Next pattern char must match literally, whatever it is */
NextChar(p, plen);
if ((plen <= 0) || !ICHAREQ(t, p))
return LIKE_FALSE;
}
else if (*p == '%')
{
/* %% is the same as % according to the SQL standard */
/* Advance past all %'s */
while ((plen > 0) && (*p == '%'))
NextChar(p, plen);
/* Trailing percent matches everything. */
if (plen <= 0)
return LIKE_TRUE;
/*
* Otherwise, scan for a text position at which we can match
* the rest of the pattern.
*/
while (tlen > 0)
{
/*
* Optimization to prevent most recursion: don't recurse
* unless first pattern char might match this text char.
*/
if (ICHAREQ(t, p) || (*p == '\\') || (*p == '_'))
{
int matched = MatchTextIC(t, tlen, p, plen);
if (matched != LIKE_FALSE)
return matched; /* TRUE or ABORT */
}
NextChar(t, tlen);
}
/*
* End of text with no match, so no point in trying later
* places to start matching this pattern.
*/
return LIKE_ABORT;
}
else if ((*p != '_') && !ICHAREQ(t, p))
{
/*
* Not the single-character wildcard and no explicit match?
* Then time to quit...
*/
return LIKE_FALSE;
}
NextChar(t, tlen);
NextChar(p, plen);
}
if (tlen > 0)
return LIKE_FALSE; /* end of pattern, but not of text */
/* End of input string. Do we have matching pattern remaining? */
while ((plen > 0) && (*p == '%')) /* allow multiple %'s at end of
* pattern */
NextChar(p, plen);
if (plen <= 0)
return LIKE_TRUE;
/*
* End of text with no match, so no point in trying later places to
* start matching this pattern.
*/
return LIKE_ABORT;
} /* MatchTextIC() */
/*
* Same as above, but specifically for bytea (binary) datatype
*/

View File

@ -0,0 +1,339 @@
/*-------------------------------------------------------------------------
*
* like_match.c
* like expression handling internal code.
*
* This file is included by like.c *twice* if multibyte is enabled.
* This is for an optimization of single byte encodings.
* Before the inclusion, we need to define following macros:
*
* CHAREQ
* ICHAREQ
* NextChar
* CopyAdvChar
* MatchText (MBMatchText)
* MatchTextIC (MBMatchTextIC)
* do_like_escape (MB_do_like_escape)
*
* Copyright (c) 1996-2001, PostgreSQL Global Development Group
*
* IDENTIFICATION
* $Header: /cvsroot/pgsql/src/backend/utils/adt/like_match.c,v 1.1 2001/10/04 02:15:47 ishii Exp $
*
*-------------------------------------------------------------------------
*/
/*
** Originally written by Rich $alz, mirror!rs, Wed Nov 26 19:03:17 EST 1986.
** Rich $alz is now <rsalz@bbn.com>.
** Special thanks to Lars Mathiesen <thorinn@diku.dk> for the LABORT code.
**
** This code was shamelessly stolen from the "pql" code by myself and
** slightly modified :)
**
** All references to the word "star" were replaced by "percent"
** All references to the word "wild" were replaced by "like"
**
** All the nice shell RE matching stuff was replaced by just "_" and "%"
**
** As I don't have a copy of the SQL standard handy I wasn't sure whether
** to leave in the '\' escape character handling.
**
** Keith Parks. <keith@mtcc.demon.co.uk>
**
** SQL92 lets you specify the escape character by saying
** LIKE <pattern> ESCAPE <escape character>. We are a small operation
** so we force you to use '\'. - ay 7/95
**
** Now we have the like_escape() function that converts patterns with
** any specified escape character (or none at all) to the internal
** default escape character, which is still '\'. - tgl 9/2000
**
** The code is rewritten to avoid requiring null-terminated strings,
** which in turn allows us to leave out some memcpy() operations.
** This code should be faster and take less memory, but no promises...
** - thomas 2000-08-06
**
*/
/*--------------------
* Match text and p, return LIKE_TRUE, LIKE_FALSE, or LIKE_ABORT.
*
* LIKE_TRUE: they match
* LIKE_FALSE: they don't match
* LIKE_ABORT: not only don't they match, but the text is too short.
*
* If LIKE_ABORT is returned, then no suffix of the text can match the
* pattern either, so an upper-level % scan can stop scanning now.
*--------------------
*/
static int
MatchText(unsigned char *t, int tlen, unsigned char *p, int plen)
{
/* Fast path for match-everything pattern */
if ((plen == 1) && (*p == '%'))
return LIKE_TRUE;
while ((tlen > 0) && (plen > 0))
{
if (*p == '\\')
{
/* Next pattern char must match literally, whatever it is */
NextChar(p, plen);
if ((plen <= 0) || !CHAREQ(t, p))
return LIKE_FALSE;
}
else if (*p == '%')
{
/* %% is the same as % according to the SQL standard */
/* Advance past all %'s */
while ((plen > 0) && (*p == '%'))
NextChar(p, plen);
/* Trailing percent matches everything. */
if (plen <= 0)
return LIKE_TRUE;
/*
* Otherwise, scan for a text position at which we can match
* the rest of the pattern.
*/
while (tlen > 0)
{
/*
* Optimization to prevent most recursion: don't recurse
* unless first pattern char might match this text char.
*/
if (CHAREQ(t, p) || (*p == '\\') || (*p == '_'))
{
int matched = MatchText(t, tlen, p, plen);
if (matched != LIKE_FALSE)
return matched; /* TRUE or ABORT */
}
NextChar(t, tlen);
}
/*
* End of text with no match, so no point in trying later
* places to start matching this pattern.
*/
return LIKE_ABORT;
}
else if ((*p != '_') && !CHAREQ(t, p))
{
/*
* Not the single-character wildcard and no explicit match?
* Then time to quit...
*/
return LIKE_FALSE;
}
NextChar(t, tlen);
NextChar(p, plen);
}
if (tlen > 0)
return LIKE_FALSE; /* end of pattern, but not of text */
/* End of input string. Do we have matching pattern remaining? */
while ((plen > 0) && (*p == '%')) /* allow multiple %'s at end of
* pattern */
NextChar(p, plen);
if (plen <= 0)
return LIKE_TRUE;
/*
* End of text with no match, so no point in trying later places to
* start matching this pattern.
*/
return LIKE_ABORT;
} /* MatchText() */
/*
* Same as above, but ignore case
*/
static int
MatchTextIC(unsigned char *t, int tlen, unsigned char *p, int plen)
{
/* Fast path for match-everything pattern */
if ((plen == 1) && (*p == '%'))
return LIKE_TRUE;
while ((tlen > 0) && (plen > 0))
{
if (*p == '\\')
{
/* Next pattern char must match literally, whatever it is */
NextChar(p, plen);
if ((plen <= 0) || !ICHAREQ(t, p))
return LIKE_FALSE;
}
else if (*p == '%')
{
/* %% is the same as % according to the SQL standard */
/* Advance past all %'s */
while ((plen > 0) && (*p == '%'))
NextChar(p, plen);
/* Trailing percent matches everything. */
if (plen <= 0)
return LIKE_TRUE;
/*
* Otherwise, scan for a text position at which we can match
* the rest of the pattern.
*/
while (tlen > 0)
{
/*
* Optimization to prevent most recursion: don't recurse
* unless first pattern char might match this text char.
*/
if (ICHAREQ(t, p) || (*p == '\\') || (*p == '_'))
{
int matched = MatchTextIC(t, tlen, p, plen);
if (matched != LIKE_FALSE)
return matched; /* TRUE or ABORT */
}
NextChar(t, tlen);
}
/*
* End of text with no match, so no point in trying later
* places to start matching this pattern.
*/
return LIKE_ABORT;
}
else if ((*p != '_') && !ICHAREQ(t, p))
{
/*
* Not the single-character wildcard and no explicit match?
* Then time to quit...
*/
return LIKE_FALSE;
}
NextChar(t, tlen);
NextChar(p, plen);
}
if (tlen > 0)
return LIKE_FALSE; /* end of pattern, but not of text */
/* End of input string. Do we have matching pattern remaining? */
while ((plen > 0) && (*p == '%')) /* allow multiple %'s at end of
* pattern */
NextChar(p, plen);
if (plen <= 0)
return LIKE_TRUE;
/*
* End of text with no match, so no point in trying later places to
* start matching this pattern.
*/
return LIKE_ABORT;
} /* MatchTextIC() */
/*
* like_escape() --- given a pattern and an ESCAPE string,
* convert the pattern to use Postgres' standard backslash escape convention.
*/
static text *
do_like_escape(text *pat, text *esc)
{
text *result;
unsigned char *p,
*e,
*r;
int plen,
elen;
bool afterescape;
p = VARDATA(pat);
plen = (VARSIZE(pat) - VARHDRSZ);
e = VARDATA(esc);
elen = (VARSIZE(esc) - VARHDRSZ);
/*
* Worst-case pattern growth is 2x --- unlikely, but it's hardly worth
* trying to calculate the size more accurately than that.
*/
result = (text *) palloc(plen * 2 + VARHDRSZ);
r = VARDATA(result);
if (elen == 0)
{
/*
* No escape character is wanted. Double any backslashes in the
* pattern to make them act like ordinary characters.
*/
while (plen > 0)
{
if (*p == '\\')
*r++ = '\\';
CopyAdvChar(r, p, plen);
}
}
else
{
/*
* The specified escape must be only a single character.
*/
NextChar(e, elen);
if (elen != 0)
elog(ERROR, "ESCAPE string must be empty or one character");
e = VARDATA(esc);
/*
* If specified escape is '\', just copy the pattern as-is.
*/
if (*e == '\\')
{
memcpy(result, pat, VARSIZE(pat));
return result;
}
/*
* Otherwise, convert occurrences of the specified escape
* character to '\', and double occurrences of '\' --- unless they
* immediately follow an escape character!
*/
afterescape = false;
while (plen > 0)
{
if (CHAREQ(p, e) && !afterescape)
{
*r++ = '\\';
NextChar(p, plen);
afterescape = true;
}
else if (*p == '\\')
{
*r++ = '\\';
if (!afterescape)
*r++ = '\\';
NextChar(p, plen);
afterescape = false;
}
else
{
CopyAdvChar(r, p, plen);
afterescape = false;
}
}
}
VARATT_SIZEP(result) = r - ((unsigned char *) result);
return result;
}