When starts_with() and the equivalent ^@ operator were added, they
were plugged into the planner in only a rather half-baked way.
Selectivity estimation got taught about the operator, but the
other infrastructure associated with LIKE/regex matching wasn't
updated. This causes these operators to be planned more stupidly
than a functionally-equivalent LIKE/regex pattern [1].
With the (admittedly later) introduction of planner support functions,
it's really quite easy to do better. The attached patch adds a planner
support function for starts_with(), with these benefits:
* A condition such as "textcol ^@ constant" can now use a regular
btree index, not only an SP-GiST index, so long as the index's
collation is C. (This works just like "textcol LIKE 'foo%'".)
* "starts_with(textcol, constant)" can be optimized the same as
"textcol ^@ constant".
I also rejiggered match_pattern_prefix() a bit, with the effect
that fixed-prefix LIKE and regex patterns are now more like
starts_with() in another way: if you apply one to an SPGiST-indexed
column, you'll get an index condition using ^@ rather than two
index conditions with >= and <. That should be more efficient
at runtime, though I didn't try to do any performance testing.
regards, tom lane
[1]
https://www.postgresql.org/message-id/CADT4RqB13KQHOJqqQ%2BWXmYtJrukS2UiFdtfTvT-XA3qYLyB6Cw%40mail.gmail.com
diff --git a/src/backend/utils/adt/like_support.c b/src/backend/utils/adt/like_support.c
index 241e6f0f59..988568825e 100644
--- a/src/backend/utils/adt/like_support.c
+++ b/src/backend/utils/adt/like_support.c
@@ -143,6 +143,14 @@ texticregexeq_support(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(like_regex_support(rawreq, Pattern_Type_Regex_IC));
}
+Datum
+text_starts_with_support(PG_FUNCTION_ARGS)
+{
+ Node *rawreq = (Node *) PG_GETARG_POINTER(0);
+
+ PG_RETURN_POINTER(like_regex_support(rawreq, Pattern_Type_Prefix));
+}
+
/* Common code for the above */
static Node *
like_regex_support(Node *rawreq, Pattern_Type ptype)
@@ -246,6 +254,7 @@ match_pattern_prefix(Node *leftop,
Oid eqopr;
Oid ltopr;
Oid geopr;
+ Oid preopr = InvalidOid;
bool collation_aware;
Expr *expr;
FmgrInfo ltproc;
@@ -302,12 +311,20 @@ match_pattern_prefix(Node *leftop,
switch (ldatatype)
{
case TEXTOID:
- if (opfamily == TEXT_PATTERN_BTREE_FAM_OID ||
- opfamily == TEXT_SPGIST_FAM_OID)
+ if (opfamily == TEXT_PATTERN_BTREE_FAM_OID)
+ {
+ eqopr = TextEqualOperator;
+ ltopr = TextPatternLessOperator;
+ geopr = TextPatternGreaterEqualOperator;
+ collation_aware = false;
+ }
+ else if (opfamily == TEXT_SPGIST_FAM_OID)
{
eqopr = TextEqualOperator;
ltopr = TextPatternLessOperator;
geopr = TextPatternGreaterEqualOperator;
+ /* This opfamily has direct support for prefixing */
+ preopr = TextPrefixOperator;
collation_aware = false;
}
else
@@ -360,20 +377,6 @@ match_pattern_prefix(Node *leftop,
return NIL;
}
- /*
- * If necessary, verify that the index's collation behavior is compatible.
- * For an exact-match case, we don't have to be picky. Otherwise, insist
- * that the index collation be "C". Note that here we are looking at the
- * index's collation, not the expression's collation -- this test is *not*
- * dependent on the LIKE/regex operator's collation.
- */
- if (collation_aware)
- {
- if (!(pstatus == Pattern_Prefix_Exact ||
- lc_collate_is_c(indexcollation)))
- return NIL;
- }
-
/*
* If necessary, coerce the prefix constant to the right type. The given
* prefix constant is either text or bytea type, therefore the only case
@@ -409,8 +412,31 @@ match_pattern_prefix(Node *leftop,
}
/*
- * Otherwise, we have a nonempty required prefix of the values.
- *
+ * Otherwise, we have a nonempty required prefix of the values. Some
+ * opclasses support prefix checks directly, otherwise we'll try to
+ * generate a range constraint.
+ */
+ if (OidIsValid(preopr) && op_in_opfamily(preopr, opfamily))
+ {
+ expr = make_opclause(preopr, BOOLOID, false,
+ (Expr *) leftop, (Expr *) prefix,
+ InvalidOid, indexcollation);
+ result = list_make1(expr);
+ return result;
+ }
+
+ /*
+ * Since we need a range constraint, it's only going to work reliably if
+ * the index is collation-insensitive or has "C" collation. Note that
+ * here we are looking at the index's collation, not the expression's
+ * collation -- this test is *not* dependent on the LIKE/regex operator's
+ * collation.
+ */
+ if (collation_aware &&
+ !lc_collate_is_c(indexcollation))
+ return NIL;
+
+ /*
* We can always say "x >= prefix".
*/
if (!op_in_opfamily(geopr, opfamily))
@@ -1165,7 +1191,6 @@ pattern_fixed_prefix(Const *patt, Pattern_Type ptype, Oid collation,
case Pattern_Type_Prefix:
/* Prefix type work is trivial. */
result = Pattern_Prefix_Partial;
- *rest_selec = 1.0; /* all */
*prefix = makeConst(patt->consttype,
patt->consttypmod,
patt->constcollid,
@@ -1175,6 +1200,8 @@ pattern_fixed_prefix(Const *patt, Pattern_Type ptype, Oid collation,
patt->constlen),
patt->constisnull,
patt->constbyval);
+ if (rest_selec != NULL)
+ *rest_selec = 1.0; /* all */
break;
default:
elog(ERROR, "unrecognized ptype: %d", (int) ptype);
diff --git a/src/include/catalog/pg_operator.dat b/src/include/catalog/pg_operator.dat
index 89c73acd68..0075a02f32 100644
--- a/src/include/catalog/pg_operator.dat
+++ b/src/include/catalog/pg_operator.dat
@@ -102,7 +102,7 @@
oprright => 'text', oprresult => 'bool', oprcom => '=(text,text)',
oprnegate => '<>(text,text)', oprcode => 'texteq', oprrest => 'eqsel',
oprjoin => 'eqjoinsel' },
-{ oid => '3877', descr => 'starts with',
+{ oid => '3877', oid_symbol => 'TextPrefixOperator', descr => 'starts with',
oprname => '^@', oprleft => 'text', oprright => 'text', oprresult => 'bool',
oprcode => 'starts_with', oprrest => 'prefixsel',
oprjoin => 'prefixjoinsel' },
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index d068d6532e..6412f369f1 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -167,8 +167,12 @@
proname => 'texteq', proleakproof => 't', prorettype => 'bool',
proargtypes => 'text text', prosrc => 'texteq' },
{ oid => '3696',
- proname => 'starts_with', proleakproof => 't', prorettype => 'bool',
- proargtypes => 'text text', prosrc => 'text_starts_with' },
+ proname => 'starts_with', prosupport => 'text_starts_with_support',
+ proleakproof => 't', prorettype => 'bool', proargtypes => 'text text',
+ prosrc => 'text_starts_with' },
+{ oid => '8923', descr => 'planner support for text_starts_with',
+ proname => 'text_starts_with_support', prorettype => 'internal',
+ proargtypes => 'internal', prosrc => 'text_starts_with_support' },
{ oid => '68',
proname => 'xideq', proleakproof => 't', prorettype => 'bool',
proargtypes => 'xid xid', prosrc => 'xideq' },
diff --git a/src/test/regress/expected/create_index_spgist.out b/src/test/regress/expected/create_index_spgist.out
index f4fb08a289..5c04df9c01 100644
--- a/src/test/regress/expected/create_index_spgist.out
+++ b/src/test/regress/expected/create_index_spgist.out
@@ -804,6 +804,22 @@ SELECT count(*) FROM radix_text_tbl WHERE t ^@ 'Worth';
2
(1 row)
+EXPLAIN (COSTS OFF)
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+ QUERY PLAN
+------------------------------------------------------------
+ Aggregate
+ -> Index Only Scan using sp_radix_ind on radix_text_tbl
+ Index Cond: (t ^@ 'Worth'::text)
+ Filter: starts_with(t, 'Worth'::text)
+(4 rows)
+
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+ count
+-------
+ 2
+(1 row)
+
-- Now check the results from bitmap indexscan
SET enable_seqscan = OFF;
SET enable_indexscan = OFF;
@@ -1333,6 +1349,23 @@ SELECT count(*) FROM radix_text_tbl WHERE t ^@ 'Worth';
2
(1 row)
+EXPLAIN (COSTS OFF)
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+ QUERY PLAN
+------------------------------------------------
+ Aggregate
+ -> Bitmap Heap Scan on radix_text_tbl
+ Filter: starts_with(t, 'Worth'::text)
+ -> Bitmap Index Scan on sp_radix_ind
+ Index Cond: (t ^@ 'Worth'::text)
+(5 rows)
+
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+ count
+-------
+ 2
+(1 row)
+
RESET enable_seqscan;
RESET enable_indexscan;
RESET enable_bitmapscan;
diff --git a/src/test/regress/sql/create_index_spgist.sql b/src/test/regress/sql/create_index_spgist.sql
index b126dae629..660bfc6193 100644
--- a/src/test/regress/sql/create_index_spgist.sql
+++ b/src/test/regress/sql/create_index_spgist.sql
@@ -295,6 +295,10 @@ EXPLAIN (COSTS OFF)
SELECT count(*) FROM radix_text_tbl WHERE t ^@ 'Worth';
SELECT count(*) FROM radix_text_tbl WHERE t ^@ 'Worth';
+EXPLAIN (COSTS OFF)
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+
-- Now check the results from bitmap indexscan
SET enable_seqscan = OFF;
SET enable_indexscan = OFF;
@@ -424,6 +428,10 @@ EXPLAIN (COSTS OFF)
SELECT count(*) FROM radix_text_tbl WHERE t ^@ 'Worth';
SELECT count(*) FROM radix_text_tbl WHERE t ^@ 'Worth';
+EXPLAIN (COSTS OFF)
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+SELECT count(*) FROM radix_text_tbl WHERE starts_with(t, 'Worth');
+
RESET enable_seqscan;
RESET enable_indexscan;
RESET enable_bitmapscan;
