From a00591629687fe281826c5979a11b46a93d8e817 Mon Sep 17 00:00:00 2001 From: Chengpeng Yan Date: Wed, 18 Nov 2026 08:07:42 +0800 Subject: [PATCH v1] Count characters, not bytes, in pattern selectivity like_selectivity() and regex_selectivity_sub() applied FIXED_CHAR_SEL once per byte, so a multibyte character was charged several times and the estimate for the same pattern depended on the database encoding. Advance by whole characters for text patterns, and pass the regex fixed-prefix length in characters to match. Bytea patterns remain byte-oriented. --- src/backend/utils/adt/like_support.c | 19 +++++++++++++----- src/test/regress/expected/collate.utf8.out | 23 ++++++++++++++++++++++ src/test/regress/sql/collate.utf8.sql | 11 +++++++++++ 3 files changed, 48 insertions(+), 5 deletions(-) diff --git a/src/backend/utils/adt/like_support.c b/src/backend/utils/adt/like_support.c index 588425b3812..8cae0d74c90 100644 --- a/src/backend/utils/adt/like_support.c +++ b/src/backend/utils/adt/like_support.c @@ -99,7 +99,7 @@ static Selectivity prefix_selectivity(PlannerInfo *root, Oid collation, Const *prefixcon); static Selectivity like_selectivity(const char *patt, int pattlen, - bool case_insensitive); + bool is_bytea); static Selectivity regex_selectivity(const char *patt, int pattlen, bool case_insensitive, int fixed_prefix_len); @@ -1052,7 +1052,8 @@ like_fixed_prefix(Const *patt_const, Const **prefix_const, *prefix_const = string_to_bytea_const(match, match_pos); if (rest_selec != NULL) - *rest_selec = like_selectivity(&patt[pos], pattlen - pos, false); + *rest_selec = like_selectivity(&patt[pos], pattlen - pos, + typeid == BYTEAOID); pfree(patt); pfree(match); @@ -1157,7 +1158,7 @@ like_fixed_prefix_ci(Const *patt_const, Oid collation, Const **prefix_const, rest = palloc(pg_database_encoding_max_length() * wrestlen + 1); rest_mblen = pg_wchar2mb_with_len(&wpatt[wpos], rest, wrestlen); - *rest_selec = like_selectivity(rest, rest_mblen, true); + *rest_selec = like_selectivity(rest, rest_mblen, false); pfree(rest); } @@ -1228,7 +1229,7 @@ regex_fixed_prefix(Const *patt_const, bool case_insensitive, Oid collation, *rest_selec = regex_selectivity(patt, strlen(patt), case_insensitive, - strlen(prefix)); + pg_mbstrlen(prefix)); pfree(patt); } } @@ -1397,7 +1398,7 @@ prefix_selectivity(PlannerInfo *root, VariableStatData *vardata, #define PARTIAL_WILDCARD_SEL 2.0 static Selectivity -like_selectivity(const char *patt, int pattlen, bool case_insensitive) +like_selectivity(const char *patt, int pattlen, bool is_bytea) { Selectivity sel = 1.0; int pos; @@ -1423,9 +1424,15 @@ like_selectivity(const char *patt, int pattlen, bool case_insensitive) if (pos >= pattlen) break; sel *= FIXED_CHAR_SEL; + if (!is_bytea) + pos += pg_mblen_range(patt + pos, patt + pattlen) - 1; } else + { sel *= FIXED_CHAR_SEL; + if (!is_bytea) + pos += pg_mblen_range(patt + pos, patt + pattlen) - 1; + } } /* Could get sel > 1 if multiple wildcards */ if (sel > 1.0) @@ -1515,11 +1522,13 @@ regex_selectivity_sub(const char *patt, int pattlen, bool case_insensitive) break; if (paren_depth == 0) sel *= FIXED_CHAR_SEL; + pos += pg_mblen_range(patt + pos, patt + pattlen) - 1; } else { if (paren_depth == 0) sel *= FIXED_CHAR_SEL; + pos += pg_mblen_range(patt + pos, patt + pattlen) - 1; } } /* Could get sel > 1 if multiple wildcards */ diff --git a/src/test/regress/expected/collate.utf8.out b/src/test/regress/expected/collate.utf8.out index cdd1a37ba18..b1ea81638c1 100644 --- a/src/test/regress/expected/collate.utf8.out +++ b/src/test/regress/expected/collate.utf8.out @@ -41,6 +41,29 @@ SELECT 'café' ILIKE 'caf_' COLLATE "C"; t (1 row) +-- pattern selectivity counts characters, not encoded bytes +EXPLAIN +SELECT * FROM generate_series(1, 2000) g(i) WHERE i::text LIKE '%é' +UNION ALL +SELECT * FROM generate_series(1, 2000) g(i) WHERE i::text ILIKE '%é' +UNION ALL +SELECT * FROM generate_series(1, 2000) g(i) WHERE i::text ~ 'é$' +UNION ALL +SELECT * FROM generate_series(1, 2000) g(i) + WHERE convert_to(i::text, 'UTF8') LIKE decode('25c3a9', 'hex'); + QUERY PLAN +--------------------------------------------------------------------------------- + Append (cost=0.00..151.41 rows=1280 width=4) + -> Function Scan on generate_series g (cost=0.00..35.00 rows=400 width=4) + Filter: ((i)::text ~~ '%é'::text) + -> Function Scan on generate_series g_1 (cost=0.00..35.00 rows=400 width=4) + Filter: ((i)::text ~~* '%é'::text) + -> Function Scan on generate_series g_2 (cost=0.00..35.00 rows=400 width=4) + Filter: ((i)::text ~ 'é$'::text) + -> Function Scan on generate_series g_3 (cost=0.00..40.00 rows=80 width=4) + Filter: (convert_to((i)::text, 'UTF8'::name) ~~ '\x25c3a9'::bytea) +(9 rows) + -- -- Test PG_C_UTF8 -- diff --git a/src/test/regress/sql/collate.utf8.sql b/src/test/regress/sql/collate.utf8.sql index 52cf068dd0c..fd9adf367ca 100644 --- a/src/test/regress/sql/collate.utf8.sql +++ b/src/test/regress/sql/collate.utf8.sql @@ -29,6 +29,17 @@ DROP COLLATION regress_builtin_c; -- a '_' matches a multibyte character SELECT 'café' ILIKE 'caf_' COLLATE "C"; +-- pattern selectivity counts characters, not encoded bytes +EXPLAIN +SELECT * FROM generate_series(1, 2000) g(i) WHERE i::text LIKE '%é' +UNION ALL +SELECT * FROM generate_series(1, 2000) g(i) WHERE i::text ILIKE '%é' +UNION ALL +SELECT * FROM generate_series(1, 2000) g(i) WHERE i::text ~ 'é$' +UNION ALL +SELECT * FROM generate_series(1, 2000) g(i) + WHERE convert_to(i::text, 'UTF8') LIKE decode('25c3a9', 'hex'); + -- -- Test PG_C_UTF8 -- -- 2.50.1 (Apple Git-155)