mirror of
https://github.com/MariaDB/server.git
synced 2025-01-27 01:04:19 +01:00
e987b9350c
(Review input addressed)
(Added handling of UPDATE/DELETE and partitioning w/o index)
If the properties of the used collation allow, do the following
equivalent rewrites:
1. UPPER(key_col)=expr -> key_col=expr
expr=UPPER(key_col) -> expr=key_col
(also rewrite both sides of the equality at the same time)
2. UPPER(key_col) IN (constant-list) -> key_col IN (constant-list)
- Mark utf8mb{3,4}_general_ci as collations that allow this.
- Add optimizer_switch='sargable_casefold=ON' to control this.
(ON by default in this patch)
- Cover the rewrite in Optimizer Trace, rewrite name is
"sargable_casefold_removal".
141 lines
3.6 KiB
Text
141 lines
3.6 KiB
Text
#
|
|
# MDEV-31496 Make optimizer handle UCASE(varchar_col)=...
|
|
#
|
|
--source include/have_sequence.inc
|
|
|
|
set
|
|
@tmp_switch_sarg_casefold=@@optimizer_switch,
|
|
optimizer_switch='sargable_casefold=on';
|
|
|
|
let $collation=utf8mb3_general_ci;
|
|
source include/sargable_casefold.inc;
|
|
|
|
--echo # must not be rewritten:
|
|
explain select * from t1 where ucase(col1 collate utf8mb3_bin)='a-3';
|
|
|
|
--echo # Will not do the rewrite due to collation mismatch:
|
|
explain select * from t1 where ucase(col1)=_utf8mb3'abc' COLLATE utf8mb3_bin;
|
|
drop table t1;
|
|
|
|
let $collation=utf8mb4_general_ci;
|
|
source include/sargable_casefold.inc;
|
|
|
|
--echo # must not be rewritten:
|
|
explain select * from t1 where ucase(col1 collate utf8mb4_bin)='a-3';
|
|
|
|
--echo # Will not do the rewrite due to collation mismatch:
|
|
explain select * from t1 where ucase(col1)=_utf8mb4'abc' COLLATE utf8mb4_bin;
|
|
|
|
--echo #
|
|
--echo # Check if optimizer_switch turns the rewrite off:
|
|
--echo #
|
|
|
|
set
|
|
@save_os=@@optimizer_switch,
|
|
optimizer_switch='sargable_casefold=off';
|
|
explain select * from t1 where upper(col1)='A-3';
|
|
explain select * from t1 where ucase(col1)='a-3';
|
|
|
|
set optimizer_switch=@save_os;
|
|
|
|
--echo # The following will not do the rewrite because the comparison
|
|
--echo # is done as DOUBLEs. Come to think of it, it won't harm to do
|
|
--echo # the rewrite but it is outside of the scope of this patch:
|
|
explain select * from t1 where ucase(col1)=123.456;
|
|
|
|
select
|
|
coercibility(upper(col1))
|
|
from t1 limit 1;
|
|
select coercibility(_utf8mb3'abc' COLLATE utf8mb3_bin);
|
|
|
|
|
|
--echo # This is transformed too even if it doesn't create any new
|
|
--echo # [potential] access paths:
|
|
explain format=json select * from t1 where upper(col1)=upper(col2);
|
|
|
|
|
|
--echo #
|
|
--echo # Check if ref access works
|
|
--echo #
|
|
create table t2 (
|
|
a varchar(32),
|
|
non_key varchar(32),
|
|
key(a)
|
|
) collate utf8mb4_general_ci;
|
|
|
|
insert into t2
|
|
select
|
|
concat('A-', seq),
|
|
concat('A-', seq)
|
|
from seq_1_to_10;
|
|
|
|
--echo # Must use ref access for t1:
|
|
explain select * from t1, t2 where upper(t1.col1)= t2.non_key;
|
|
|
|
# Check the interplay with equality propagation
|
|
|
|
create table t3 (
|
|
a varchar(32),
|
|
b varchar(32),
|
|
key(a),
|
|
key(b)
|
|
) collate utf8mb3_general_ci;
|
|
insert into t3 values ('abc','ABC'), ('xyz','XYZ');
|
|
|
|
explain extended
|
|
select a from t3 ignore index(a) where a=b and upper(b)='ABC';
|
|
|
|
--echo #
|
|
--echo # Check that rewrite isn't applied for non-applicable collations
|
|
--echo #
|
|
create table t4 (
|
|
col1 varchar(32) collate utf8mb3_bin,
|
|
col2 varchar(32) collate utf8mb3_czech_ci,
|
|
col3 varchar(32) collate latin1_bin,
|
|
key(col1),
|
|
key(col2),
|
|
key(col3)
|
|
);
|
|
|
|
insert into t4
|
|
select
|
|
concat('A-', seq),
|
|
concat('A-', seq),
|
|
concat('A-', seq)
|
|
from seq_1_to_100;
|
|
|
|
analyze table t4 persistent for all;
|
|
|
|
--echo # None should use ref access:
|
|
explain select * from t4 where upper(col1)='A-3';
|
|
explain select * from t4 where upper(col2)='a-3';
|
|
explain select * from t4 where upper(col3)='a-3';
|
|
|
|
--echo #
|
|
--echo # Check that rewrite works for UPPER(col) IN (const-list)
|
|
--echo #
|
|
set
|
|
@tmp_ot= @@optimizer_trace,
|
|
optimizer_trace=1;
|
|
|
|
--echo # must use range:
|
|
explain
|
|
select * from t1 where upper(col1) IN ('A-3','A-4','a-5');
|
|
select * from t1 where upper(col1) IN ('A-3','A-4','a-5');
|
|
|
|
--echo # Will not use the rewrite:
|
|
explain
|
|
select * from t1 where upper(col1) IN ('A-3','A-4',col2);
|
|
|
|
--echo #
|
|
--echo # MDEV-31946: Optimizer handle UCASE(varchar_col)=... does not work for UPDATE/DELETE
|
|
--echo #
|
|
explain delete from t1 where upper(col1)='A';
|
|
explain delete from t1 where upper(col1) IN ('A','B');
|
|
|
|
explain update t1 set col2='ABC' where upper(col1)='A';
|
|
explain update t1 set col2='ABC' where upper(col1) IN ('A','B');
|
|
|
|
drop table t1,t2,t3,t4;
|
|
|
|
set optimizer_switch=@tmp_switch_sarg_casefold;
|