Skip to content

M
mariadb
Commit 54a713aa authored by sergefp@mysql.com's avatar sergefp@mysql.com
Browse files
Options

BUG#8804: wrong results for NULL IN (SELECT ...) 

Evaluate "NULL IN (SELECT ...)" in a special way: Disable pushed-down 
conditions and their "consequences": 
 = Do full table scans instead of unique_[index_subquery] lookups.
 = Change appropriate "ref_or_null" accesses to full table scans in
   subquery's joins.
Also cache value of NULL IN (SELECT ...) if the SELECT is not correlated 
wrt any upper select.
parent 2ab34bac
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 900 additions and 104 deletions
mysql-test/r/subselect.result
View file @ 54a713aa
......@@ -744,7 +744,7 @@ id select_type table type possible_keys key key_len ref rows Extra
3 DEPENDENT UNION NULL NULL NULL NULL NULL NULL NULL No tables used
NULL UNION RESULT <union2,3> ALL NULL NULL NULL NULL NULL
Warnings:
Note 1003 select `test`.`t2`.`id` AS `id` from `test`.`t2` where <in_optimizer>(`test`.`t2`.`id`,<exists>(select 1 AS `1` having (<cache>(`test`.`t2`.`id`) = <ref_null_helper>(1)) union select 3 AS `3` having (<cache>(`test`.`t2`.`id`) = <ref_null_helper>(3))))
Note 1003 select `test`.`t2`.`id` AS `id` from `test`.`t2` where <in_optimizer>(`test`.`t2`.`id`,<exists>(select 1 AS `1` having trigcond((<cache>(`test`.`t2`.`id`) = <ref_null_helper>(1))) union select 3 AS `3` having trigcond((<cache>(`test`.`t2`.`id`) = <ref_null_helper>(3)))))
SELECT * FROM t2 WHERE id IN (SELECT 5 UNION SELECT 3);
id
SELECT * FROM t2 WHERE id IN (SELECT 5 UNION SELECT 2);
......@@ -907,7 +907,7 @@ id select_type table type possible_keys key key_len ref rows Extra
2 DEPENDENT SUBQUERY t2 ref_or_null a a 5 func 2 Using where; Using index
2 DEPENDENT SUBQUERY t3 ALL NULL NULL NULL NULL 3 Using where
Warnings:
Note 1003 select `test`.`t1`.`a` AS `a`,<in_optimizer>(`test`.`t1`.`a`,<exists>(select 1 AS `Not_used` from `test`.`t2` join `test`.`t3` where ((`test`.`t3`.`a` = `test`.`t2`.`a`) and ((<cache>(`test`.`t1`.`a`) = `test`.`t2`.`a`) or isnull(`test`.`t2`.`a`))) having <is_not_null_test>(`test`.`t2`.`a`))) AS `t1.a in (select t2.a from t2,t3 where t3.a=t2.a)` from `test`.`t1`
Note 1003 select `test`.`t1`.`a` AS `a`,<in_optimizer>(`test`.`t1`.`a`,<exists>(select 1 AS `Not_used` from `test`.`t2` join `test`.`t3` where ((`test`.`t3`.`a` = `test`.`t2`.`a`) and trigcond(((<cache>(`test`.`t1`.`a`) = `test`.`t2`.`a`) or isnull(`test`.`t2`.`a`)))) having trigcond(<is_not_null_test>(`test`.`t2`.`a`)))) AS `t1.a in (select t2.a from t2,t3 where t3.a=t2.a)` from `test`.`t1`
drop table t1,t2,t3;
create table t1 (a float);
select 10.5 IN (SELECT * from t1 LIMIT 1);
......@@ -2817,19 +2817,19 @@ id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t1 ALL NULL NULL NULL NULL 8
2 DEPENDENT SUBQUERY t2 ALL NULL NULL NULL NULL 9 Using where
Warnings:
Note 1003 select `test`.`t1`.`one` AS `one`,`test`.`t1`.`two` AS `two`,<in_optimizer>((`test`.`t1`.`one`,`test`.`t1`.`two`),<exists>(select `test`.`t2`.`one` AS `one`,`test`.`t2`.`two` AS `two` from `test`.`t2` where ((`test`.`t2`.`flag` = _latin1'0') and ((<cache>(`test`.`t1`.`one`) = `test`.`t2`.`one`) or isnull(`test`.`t2`.`one`)) and ((<cache>(`test`.`t1`.`two`) = `test`.`t2`.`two`) or isnull(`test`.`t2`.`two`))) having (<is_not_null_test>(`test`.`t2`.`one`) and <is_not_null_test>(`test`.`t2`.`two`)))) AS `test` from `test`.`t1`
Note 1003 select `test`.`t1`.`one` AS `one`,`test`.`t1`.`two` AS `two`,<in_optimizer>((`test`.`t1`.`one`,`test`.`t1`.`two`),<exists>(select `test`.`t2`.`one` AS `one`,`test`.`t2`.`two` AS `two` from `test`.`t2` where ((`test`.`t2`.`flag` = _latin1'0') and trigcond((((<cache>(`test`.`t1`.`one`) = `test`.`t2`.`one`) or isnull(`test`.`t2`.`one`)) and ((<cache>(`test`.`t1`.`two`) = `test`.`t2`.`two`) or isnull(`test`.`t2`.`two`))))) having trigcond((<is_not_null_test>(`test`.`t2`.`one`) and <is_not_null_test>(`test`.`t2`.`two`))))) AS `test` from `test`.`t1`
explain extended SELECT one,two from t1 where ROW(one,two) IN (SELECT one,two FROM t2 WHERE flag = 'N');
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t1 ALL NULL NULL NULL NULL 8 Using where
2 DEPENDENT SUBQUERY t2 ALL NULL NULL NULL NULL 9 Using where
Warnings:
Note 1003 select `test`.`t1`.`one` AS `one`,`test`.`t1`.`two` AS `two` from `test`.`t1` where <in_optimizer>((`test`.`t1`.`one`,`test`.`t1`.`two`),<exists>(select `test`.`t2`.`one` AS `one`,`test`.`t2`.`two` AS `two` from `test`.`t2` where ((`test`.`t2`.`flag` = _latin1'N') and (<cache>(`test`.`t1`.`one`) = `test`.`t2`.`one`) and (<cache>(`test`.`t1`.`two`) = `test`.`t2`.`two`))))
Note 1003 select `test`.`t1`.`one` AS `one`,`test`.`t1`.`two` AS `two` from `test`.`t1` where <in_optimizer>((`test`.`t1`.`one`,`test`.`t1`.`two`),<exists>(select `test`.`t2`.`one` AS `one`,`test`.`t2`.`two` AS `two` from `test`.`t2` where ((`test`.`t2`.`flag` = _latin1'N') and trigcond(((<cache>(`test`.`t1`.`one`) = `test`.`t2`.`one`) and (<cache>(`test`.`t1`.`two`) = `test`.`t2`.`two`))))))
explain extended SELECT one,two,ROW(one,two) IN (SELECT one,two FROM t2 WHERE flag = '0' group by one,two) as 'test' from t1;
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t1 ALL NULL NULL NULL NULL 8
2 DEPENDENT SUBQUERY t2 ALL NULL NULL NULL NULL 9 Using where; Using temporary; Using filesort
Warnings:
Note 1003 select `test`.`t1`.`one` AS `one`,`test`.`t1`.`two` AS `two`,<in_optimizer>((`test`.`t1`.`one`,`test`.`t1`.`two`),<exists>(select `test`.`t2`.`one` AS `one`,`test`.`t2`.`two` AS `two` from `test`.`t2` where (`test`.`t2`.`flag` = _latin1'0') group by `test`.`t2`.`one`,`test`.`t2`.`two` having (((<cache>(`test`.`t1`.`one`) = `test`.`t2`.`one`) or isnull(`test`.`t2`.`one`)) and ((<cache>(`test`.`t1`.`two`) = `test`.`t2`.`two`) or isnull(`test`.`t2`.`two`)) and <is_not_null_test>(`test`.`t2`.`one`) and <is_not_null_test>(`test`.`t2`.`two`)))) AS `test` from `test`.`t1`
Note 1003 select `test`.`t1`.`one` AS `one`,`test`.`t1`.`two` AS `two`,<in_optimizer>((`test`.`t1`.`one`,`test`.`t1`.`two`),<exists>(select `test`.`t2`.`one` AS `one`,`test`.`t2`.`two` AS `two` from `test`.`t2` where (`test`.`t2`.`flag` = _latin1'0') group by `test`.`t2`.`one`,`test`.`t2`.`two` having trigcond((((<cache>(`test`.`t1`.`one`) = `test`.`t2`.`one`) or isnull(`test`.`t2`.`one`)) and ((<cache>(`test`.`t1`.`two`) = `test`.`t2`.`two`) or isnull(`test`.`t2`.`two`)) and <is_not_null_test>(`test`.`t2`.`one`) and <is_not_null_test>(`test`.`t2`.`two`))))) AS `test` from `test`.`t1`
DROP TABLE t1,t2;
CREATE TABLE t1 (a char(5), b char(5));
INSERT INTO t1 VALUES (NULL,'aaa'), ('aaa','aaa');
......
mysql-test/r/subselect3.result 0 → 100644
View file @ 54a713aa
drop table if exists t0, t1, t2, t3, t4;
create table t1 (oref int, grp int, ie int) ;
insert into t1 (oref, grp, ie) values
(1, 1, 1),
(1, 1, 1),
(1, 2, NULL),
(2, 1, 3),
(3, 1, 4),
(3, 2, NULL);
create table t2 (oref int, a int);
insert into t2 values
(1, 1),
(2, 2),
(3, 3),
(4, NULL),
(2, NULL);
select a, oref, a in (select max(ie)
from t1 where oref=t2.oref group by grp) from t2;
a oref a in (select max(ie)
from t1 where oref=t2.oref group by grp)
1 1 1
2 2 0
3 3 NULL
NULL 4 0
NULL 2 NULL
explain extended
select a, oref, a in (select max(ie)
from t1 where oref=t2.oref group by grp) from t2;
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t2 ALL NULL NULL NULL NULL 5
2 DEPENDENT SUBQUERY t1 ALL NULL NULL NULL NULL 6 Using where; Using temporary; Using filesort
Warnings:
Note 1276 Field or reference 't2.oref' of SELECT #2 was resolved in SELECT #1
Note 1003 select `test`.`t2`.`a` AS `a`,`test`.`t2`.`oref` AS `oref`,<in_optimizer>(`test`.`t2`.`a`,<exists>(select max(`test`.`t1`.`ie`) AS `max(ie)` from `test`.`t1` where (`test`.`t1`.`oref` = `test`.`t2`.`oref`) group by `test`.`t1`.`grp` having trigcond((<cache>(`test`.`t2`.`a`) = <ref_null_helper>(max(`test`.`t1`.`ie`)))))) AS `a in (select max(ie)
from t1 where oref=t2.oref group by grp)` from `test`.`t2`
explain extended
select a, oref from t2
where a in (select max(ie) from t1 where oref=t2.oref group by grp);
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t2 ALL NULL NULL NULL NULL 5 Using where
2 DEPENDENT SUBQUERY t1 ALL NULL NULL NULL NULL 6 Using where; Using temporary; Using filesort
Warnings:
Note 1276 Field or reference 't2.oref' of SELECT #2 was resolved in SELECT #1
Note 1003 select `test`.`t2`.`a` AS `a`,`test`.`t2`.`oref` AS `oref` from `test`.`t2` where <in_optimizer>(`test`.`t2`.`a`,<exists>(select max(`test`.`t1`.`ie`) AS `max(ie)` from `test`.`t1` where (`test`.`t1`.`oref` = `test`.`t2`.`oref`) group by `test`.`t1`.`grp` having (<cache>(`test`.`t2`.`a`) = <ref_null_helper>(max(`test`.`t1`.`ie`)))))
create table t3 (a int);
insert into t3 values (NULL), (NULL);
flush status;
select a in (select max(ie) from t1 where oref=4 group by grp) from t3;
a in (select max(ie) from t1 where oref=4 group by grp)
0
0
show status like 'Handler_read_rnd_next';
Variable_name Value
Handler_read_rnd_next 11
select ' ^ This must show 11' Z;
Z
^ This must show 11
explain extended select a in (select max(ie) from t1 where oref=4 group by grp) from t3;
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t3 ALL NULL NULL NULL NULL 2
2 DEPENDENT SUBQUERY t1 ALL NULL NULL NULL NULL 6 Using where; Using temporary; Using filesort
Warnings:
Note 1003 select <in_optimizer>(`test`.`t3`.`a`,<exists>(select max(`test`.`t1`.`ie`) AS `max(ie)` from `test`.`t1` where (`test`.`t1`.`oref` = 4) group by `test`.`t1`.`grp` having trigcond((<cache>(`test`.`t3`.`a`) = <ref_null_helper>(max(`test`.`t1`.`ie`)))))) AS `a in (select max(ie) from t1 where oref=4 group by grp)` from `test`.`t3`
drop table t1, t2, t3;
create table t1 (a int, oref int, key(a));
insert into t1 values
(1, 1),
(1, NULL),
(2, 3),
(2, NULL),
(3, NULL);
create table t2 (a int, oref int);
insert into t2 values (1, 1), (2,2), (NULL, 3), (NULL, 4);
select oref, a, a in (select a from t1 where oref=t2.oref) Z from t2;
oref a Z
1 1 1
2 2 0
3 NULL NULL
4 NULL 0
explain extended
select oref, a, a in (select a from t1 where oref=t2.oref) Z from t2;
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t2 ALL NULL NULL NULL NULL 4
2 DEPENDENT SUBQUERY t1 index_subquery a a 5 func 2 Using index; Using where
Warnings:
Note 1276 Field or reference 't2.oref' of SELECT #2 was resolved in SELECT #1
Note 1003 select `test`.`t2`.`oref` AS `oref`,`test`.`t2`.`a` AS `a`,<in_optimizer>(`test`.`t2`.`a`,<exists>(<index_lookup>(<cache>(`test`.`t2`.`a`) in t1 on a checking NULL where (`test`.`t1`.`oref` = `test`.`t2`.`oref`)))) AS `Z` from `test`.`t2`
flush status;
select oref, a from t2 where a in (select a from t1 where oref=t2.oref);
oref a
1 1
show status like '%Handler_read_rnd_next';
Variable_name Value
Handler_read_rnd_next 5
delete from t2;
insert into t2 values (NULL, 0),(NULL, 0), (NULL, 0), (NULL, 0);
flush status;
select oref, a, a in (select a from t1 where oref=t2.oref) Z from t2;
oref a Z
0 NULL 0
0 NULL 0
0 NULL 0
0 NULL 0
show status like '%Handler_read%';
Variable_name Value
Handler_read_first 0
Handler_read_key 0
Handler_read_next 0
Handler_read_prev 0
Handler_read_rnd 0
Handler_read_rnd_next 29
select 'No key lookups, seq reads: 29= 5 reads from t2 + 4 * 6 reads from t1.' Z;
Z
No key lookups, seq reads: 29= 5 reads from t2 + 4 * 6 reads from t1.
drop table t1, t2;
create table t1 (a int, b int, primary key (a));
insert into t1 values (1,1), (3,1),(100,1);
create table t2 (a int, b int);
insert into t2 values (1,1),(2,1),(NULL,1),(NULL,0);
select a,b, a in (select a from t1 where t1.b = t2.b) Z from t2 ;
a b Z
1 1 1
2 1 0
NULL 1 NULL
NULL 0 0
drop table t1, t2;
create table t1 (a int, b int, key(a));
insert into t1 values
(0,0),(1,1),(2,2),(3,3),(4,4),(5,5),(6,6),(7,7),(8,8),(9,9);
create table t2 like t1;
insert into t2 select * from t1;
update t2 set b=1;
create table t3 (a int, oref int);
insert into t3 values (1, 1), (NULL,1), (NULL,0);
select a, oref,
t3.a in (select t1.a from t1, t2 where t1.b=t2.a and t2.b=t3.oref) Z
from t3;
a oref Z
1 1 1
NULL 1 NULL
NULL 0 0
explain extended
select a, oref,
t3.a in (select t1.a from t1, t2 where t1.b=t2.a and t2.b=t3.oref) Z
from t3;
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY t3 ALL NULL NULL NULL NULL 3
2 DEPENDENT SUBQUERY t1 ref_or_null a a 5 func 4 Using where
2 DEPENDENT SUBQUERY t2 ref a a 5 test.t1.b 1 Using where
Warnings:
Note 1276 Field or reference 't3.oref' of SELECT #2 was resolved in SELECT #1
Note 1003 select `test`.`t3`.`a` AS `a`,`test`.`t3`.`oref` AS `oref`,<in_optimizer>(`test`.`t3`.`a`,<exists>(select 1 AS `Not_used` from `test`.`t1` join `test`.`t2` where ((`test`.`t2`.`a` = `test`.`t1`.`b`) and (`test`.`t2`.`b` = `test`.`t3`.`oref`) and trigcond(((<cache>(`test`.`t3`.`a`) = `test`.`t1`.`a`) or isnull(`test`.`t1`.`a`)))) having trigcond(<is_not_null_test>(`test`.`t1`.`a`)))) AS `Z` from `test`.`t3`
drop table t1, t2, t3;
mysql-test/t/subselect3.test 0 → 100644
View file @ 54a713aa
--disable_warnings
drop table if exists t0, t1, t2, t3, t4;
--enable_warnings
#
# 1. Subquery with GROUP/HAVING
#
create table t1 (oref int, grp int, ie int) ;
insert into t1 (oref, grp, ie) values
(1, 1, 1),
(1, 1, 1),
(1, 2, NULL),
(2, 1, 3),
(3, 1, 4),
(3, 2, NULL);
# Ok, for
# select max(ie) from t1 where oref=PARAM group by grp
# we'll have:
# 1 -> (1, NULL) matching + NULL
# 2 -> (3) non-matching
# 3 -> (3, NULL) non-matching + NULL
# 4 -> () nothing.
create table t2 (oref int, a int);
insert into t2 values
(1, 1),
(2, 2),
(3, 3),
(4, NULL),
(2, NULL);
# true, false, null, false, null
select a, oref, a in (select max(ie)
from t1 where oref=t2.oref group by grp) from t2;
# This must have a trigcond
explain extended
select a, oref, a in (select max(ie)
from t1 where oref=t2.oref group by grp) from t2;
# This must not have a trigcond:
explain extended
select a, oref from t2
where a in (select max(ie) from t1 where oref=t2.oref group by grp);
# Non-correlated subquery, 2 NULL evaluations
create table t3 (a int);
insert into t3 values (NULL), (NULL);
flush status;
select a in (select max(ie) from t1 where oref=4 group by grp) from t3;
show status like 'Handler_read_rnd_next';
select ' ^ This must show 11' Z;
# This must show trigcond:
explain extended select a in (select max(ie) from t1 where oref=4 group by grp) from t3;
drop table t1, t2, t3;
#
# 2. Subquery handled with 'index_subquery':
#
create table t1 (a int, oref int, key(a));
insert into t1 values
(1, 1),
(1, NULL),
(2, 3),
(2, NULL),
(3, NULL);
create table t2 (a int, oref int);
insert into t2 values (1, 1), (2,2), (NULL, 3), (NULL, 4);
select oref, a, a in (select a from t1 where oref=t2.oref) Z from t2;
# The next explain shows "using index" but that is just incorrect display
# (there is a bug filed about this).
explain extended
select oref, a, a in (select a from t1 where oref=t2.oref) Z from t2;
flush status;
select oref, a from t2 where a in (select a from t1 where oref=t2.oref);
# This will only show access to t2:
show status like '%Handler_read_rnd_next';
# Check that repeated NULL-scans are not cached (subq. is not correlated):
delete from t2;
insert into t2 values (NULL, 0),(NULL, 0), (NULL, 0), (NULL, 0);
flush status;
select oref, a, a in (select a from t1 where oref=t2.oref) Z from t2;
show status like '%Handler_read%';
select 'No key lookups, seq reads: 29= 5 reads from t2 + 4 * 6 reads from t1.' Z;
drop table t1, t2;
#
# 3. Subquery handled with 'unique_index_subquery':
#
create table t1 (a int, b int, primary key (a));
insert into t1 values (1,1), (3,1),(100,1);
create table t2 (a int, b int);
insert into t2 values (1,1),(2,1),(NULL,1),(NULL,0);
select a,b, a in (select a from t1 where t1.b = t2.b) Z from t2 ;
drop table t1, t2;
#
# 4. Subquery that is a join, with ref access
#
create table t1 (a int, b int, key(a));
insert into t1 values
(0,0),(1,1),(2,2),(3,3),(4,4),(5,5),(6,6),(7,7),(8,8),(9,9);
create table t2 like t1;
insert into t2 select * from t1;
update t2 set b=1;
create table t3 (a int, oref int);
insert into t3 values (1, 1), (NULL,1), (NULL,0);
select a, oref,
t3.a in (select t1.a from t1, t2 where t1.b=t2.a and t2.b=t3.oref) Z
from t3;
# This must have trigcond in WHERE and HAVING:
explain extended
select a, oref,
t3.a in (select t1.a from t1, t2 where t1.b=t2.a and t2.b=t3.oref) Z
from t3;
drop table t1, t2, t3;
sql/item.h
View file @ 54a713aa
......@@ -1959,6 +1959,16 @@ public:
class Item_in_subselect;
/*
An object of this class:
- Converts val_XXX() calls to ref->val_XXX_result() calls, like Item_ref.
- Sets owner->was_null=TRUE if it has returned a NULL value from any
val_XXX() function. This allows to inject an Item_ref_null_helper
object into subquery and then check if the subquery has produced a row
with NULL value.
*/
class Item_ref_null_helper: public Item_ref
{
protected:
......
sql/item_cmpfunc.cc
View file @ 54a713aa
......@@ -786,9 +786,41 @@ longlong Item_in_optimizer::val_int()
{
DBUG_ASSERT(fixed == 1);
cache->store(args[0]);
if (cache->null_value)
{
null_value= 1;
if (((Item_in_subselect*)args[1])->is_top_level_item())
{
/*
We're evaluating "NULL IN (SELECT ...)". The result can be NULL or
FALSE, and we can return one instead of another. Just return NULL.
*/
null_value= 1;
}
else
{
if (!((Item_in_subselect*)args[1])->is_correlated &&
result_for_null_param != UNKNOWN)
{
/* Use cached value from previous execution */
null_value= result_for_null_param;
}
else
{
/*
We're evaluating "NULL IN (SELECT ...)". The result is:
FALSE if SELECT produces an empty set, or
NULL otherwise.
We disable the predicates we've pushed down into subselect, run the
subselect and see if it has produced any rows.
*/
((Item_in_subselect*)args[1])->enable_pushed_conds= FALSE;
longlong tmp= args[1]->val_bool_result();
result_for_null_param= null_value=
!((Item_in_subselect*)args[1])->engine->no_rows();
((Item_in_subselect*)args[1])->enable_pushed_conds= TRUE;
}
}
return 0;
}
bool tmp= args[1]->val_bool_result();
......
sql/item_cmpfunc.h
View file @ 54a713aa
......@@ -100,25 +100,44 @@ public:
};
class Item_cache;
#define UNKNOWN ((my_bool)-1)
/*
Item_in_optimizer(left_expr, Item_in_subselect(...))
Item_in_optimizer is used to wrap an instance of Item_in_subselect. This
class does the following:
- Evaluate the left expression and store it in Item_cache_* object (to
avoid re-evaluating it many times during subquery execution)
- Shortcut the evaluation of "NULL IN (...)" to NULL in the cases where we
don't care if the result is NULL or FALSE.
NOTE
It is not quite clear why the above listed functionality should be
placed into a separate class called 'Item_in_optimizer'.
*/
class Item_in_optimizer: public Item_bool_func
{
protected:
Item_cache *cache;
bool save_cache;
/*
Stores the value of "NULL IN (SELECT ...)" for uncorrelated subqueries:
UNKNOWN - "NULL in (SELECT ...)" has not yet been evaluated
FALSE - result is FALSE
TRUE - result is NULL
*/
my_bool result_for_null_param;
public:
Item_in_optimizer(Item *a, Item_in_subselect *b):
Item_bool_func(a, my_reinterpret_cast(Item *)(b)), cache(0), save_cache(0)
Item_bool_func(a, my_reinterpret_cast(Item *)(b)), cache(0),
save_cache(0), result_for_null_param(UNKNOWN)
{}
bool fix_fields(THD *, Item **);
bool fix_left(THD *thd, Item **ref);
bool is_null();
/*
Item_in_optimizer item is special boolean function. On value request
(one of val, val_int or val_str methods) it evaluate left expression
of IN by storing it value in cache item (one of Item_cache* items),
then it test cache is it NULL. If left expression (cache) is NULL then
Item_in_optimizer return NULL, else it evaluate Item_in_subselect.
*/
longlong val_int();
void cleanup();
const char *func_name() const { return "<in_optimizer>"; }
......@@ -256,9 +275,11 @@ public:
class Item_maxmin_subselect;
/*
trigcond<param>(arg) ::= param? arg : TRUE
The class Item_func_trig_cond is used for guarded predicates
which are employed only for internal purposes.
A guarded predicates is an object consisting of an a regular or
A guarded predicate is an object consisting of an a regular or
a guarded predicate P and a pointer to a boolean guard variable g.
A guarded predicate P/g is evaluated to true if the value of the
guard g is false, otherwise it is evaluated to the same value that
......@@ -276,6 +297,10 @@ class Item_maxmin_subselect;
Objects of this class are built only for query execution after
the execution plan has been already selected. That's why this
class needs only val_int out of generic methods.
Current uses of Item_func_trig_cond objects:
- To wrap selection conditions when executing outer joins
- To wrap condition that is pushed down into subquery
*/
class Item_func_trig_cond: public Item_bool_func
......@@ -1019,6 +1044,11 @@ public:
/* Functions used by HAVING for rewriting IN subquery */
class Item_in_subselect;
/*
This is like IS NOT NULL but it also remembers if it ever has
encountered a NULL.
*/
class Item_is_not_null_test :public Item_func_isnull
{
Item_in_subselect* owner;
......
sql/item_subselect.cc
View file @ 54a713aa
This diff is collapsed.
sql/item_subselect.h
View file @ 54a713aa
......@@ -60,6 +60,9 @@ public:
/* subquery is transformed */
bool changed;
/* TRUE <=> The underlying SELECT is correlated w.r.t some ancestor select */
bool is_correlated;
enum trans_res {RES_OK, RES_REDUCE, RES_ERROR};
enum subs_type {UNKNOWN_SUBS, SINGLEROW_SUBS,
EXISTS_SUBS, IN_SUBS, ALL_SUBS, ANY_SUBS};
......@@ -92,7 +95,7 @@ public:
return null_value;
}
bool fix_fields(THD *thd, Item **ref);
virtual bool exec();
virtual bool exec(bool full_scan);
virtual void fix_length_and_dec();
table_map used_tables() const;
table_map not_null_tables() const { return 0; }
......@@ -215,7 +218,20 @@ public:
friend class subselect_indexsubquery_engine;
};
/* IN subselect */
/*
IN subselect: this represents "left_exr IN (SELECT ...)"
This class has:
- (as a descendant of Item_subselect) a "subquery execution engine" which
allows it to evaluate subqueries. (and this class participates in
execution by having was_null variable where part of execution result
is stored.
- Transformation methods (todo: more on this).
This class is not used directly, it is "wrapped" into Item_in_optimizer
which provides some small bits of subquery evaluation.
*/
class Item_in_subselect :public Item_exists_subselect
{
......@@ -231,12 +247,14 @@ protected:
bool abort_on_null;
bool transformed;
public:
/* Used to trigger on/off conditions that were pushed down to subselect */
bool enable_pushed_conds;
Item_func_not_all *upper_item; // point on NOT/NOP before ALL/SOME subquery
Item_in_subselect(Item * left_expr, st_select_lex *select_lex);
Item_in_subselect()
:Item_exists_subselect(), optimizer(0), abort_on_null(0), transformed(0),
upper_item(0)
enable_pushed_conds(TRUE), upper_item(0)
{}
subs_type substype() { return IN_SUBS; }
......@@ -256,6 +274,7 @@ public:
my_decimal *val_decimal(my_decimal *);
bool val_bool();
void top_level_item() { abort_on_null=1; }
inline bool is_top_level_item() { return abort_on_null; }
bool test_limit(st_select_lex_unit *unit);
void print(String *str);
bool fix_fields(THD *thd, Item **ref);
......@@ -313,7 +332,28 @@ public:
THD * get_thd() { return thd; }
virtual int prepare()= 0;
virtual void fix_length_and_dec(Item_cache** row)= 0;
virtual int exec()= 0;
/*
Execute the engine
SYNOPSIS
exec()
full_scan TRUE - Pushed-down predicates are disabled, the engine
must disable made based on those predicates.
FALSE - Pushed-down predicates are in effect.
DESCRIPTION
Execute the engine. The result of execution is subquery value that is
either captured by previously set up select_result-based 'sink' or
stored somewhere by the exec() method itself.
A required side effect: if full_scan==TRUE, subselect_engine->no_rows()
should return correct result.
RETURN
0 - OK
1 - Either an execution error, or the engine was be "changed", and
caller should call exec() again for the new engine.
*/
virtual int exec(bool full_scan)= 0;
virtual uint cols()= 0; /* return number of columns in select */
virtual uint8 uncacheable()= 0; /* query is uncacheable */
enum Item_result type() { return res_type; }
......@@ -325,6 +365,8 @@ public:
virtual bool change_result(Item_subselect *si, select_subselect *result)= 0;
virtual bool no_tables()= 0;
virtual bool is_executed() const { return FALSE; }
/* Check if subquery produced any rows during last query execution */
virtual bool no_rows() = 0;
};
......@@ -342,7 +384,7 @@ public:
void cleanup();
int prepare();
void fix_length_and_dec(Item_cache** row);
int exec();
int exec(bool full_scan);
uint cols();
uint8 uncacheable();
void exclude();
......@@ -351,6 +393,7 @@ public:
bool change_result(Item_subselect *si, select_subselect *result);
bool no_tables();
bool is_executed() const { return executed; }
bool no_rows();
};
......@@ -364,7 +407,7 @@ public:
void cleanup();
int prepare();
void fix_length_and_dec(Item_cache** row);
int exec();
int exec(bool full_scan);
uint cols();
uint8 uncacheable();
void exclude();
......@@ -373,6 +416,7 @@ public:
bool change_result(Item_subselect *si, select_subselect *result);
bool no_tables();
bool is_executed() const;
bool no_rows();
};
......@@ -382,6 +426,12 @@ class subselect_uniquesubquery_engine: public subselect_engine
protected:
st_join_table *tab;
Item *cond;
/*
TRUE<=> last execution produced empty set. Valid only when left
expression is NULL.
*/
bool empty_result_set;
bool null_keypart; /* TRUE <=> constructed search tuple has a NULL */
public:
// constructor can assign THD because it will be called after JOIN::prepare
......@@ -395,7 +445,7 @@ public:
void cleanup();
int prepare();
void fix_length_and_dec(Item_cache** row);
int exec();
int exec(bool full_scan);
uint cols() { return 1; }
uint8 uncacheable() { return UNCACHEABLE_DEPENDENT; }
void exclude();
......@@ -403,11 +453,15 @@ public:
void print (String *str);
bool change_result(Item_subselect *si, select_subselect *result);
bool no_tables();
int scan_table();
bool copy_ref_key();
bool no_rows() { return empty_result_set; }
};
class subselect_indexsubquery_engine: public subselect_uniquesubquery_engine
{
/* FALSE for 'ref', TRUE for 'ref-or-null'. */
bool check_null;
public:
......@@ -418,7 +472,7 @@ public:
:subselect_uniquesubquery_engine(thd, tab_arg, subs, where),
check_null(chk_null)
{}
int exec();
int exec(bool full_scan);
void print (String *str);
};
......
sql/records.cc
View file @ 54a713aa
......@@ -20,7 +20,7 @@
#include "mysql_priv.h"
static int rr_quick(READ_RECORD *info);
static int rr_sequential(READ_RECORD *info);
int rr_sequential(READ_RECORD *info);
static int rr_from_tempfile(READ_RECORD *info);
static int rr_unpack_from_tempfile(READ_RECORD *info);
static int rr_unpack_from_buffer(READ_RECORD *info);
......@@ -184,6 +184,7 @@ void init_read_record(READ_RECORD *info,THD *thd, TABLE *table,
} /* init_read_record */
void end_read_record(READ_RECORD *info)
{ /* free cache if used */
if (info->cache)
......@@ -289,7 +290,7 @@ static int rr_index(READ_RECORD *info)
}
static int rr_sequential(READ_RECORD *info)
int rr_sequential(READ_RECORD *info)
{
int tmp;
while ((tmp=info->file->rnd_next(info->record)))
......
sql/sql_lex.cc
View file @ 54a713aa
......@@ -1179,7 +1179,7 @@ void st_select_lex::init_select()
select_limit= 0; /* denotes the default limit = HA_POS_ERROR */
offset_limit= 0; /* denotes the default offset = 0 */
with_sum_func= 0;
is_correlated= 0;
}
/*
......@@ -1373,6 +1373,8 @@ void st_select_lex::mark_as_dependent(SELECT_LEX *last)
SELECT_LEX_UNIT *munit= s->master_unit();
munit->uncacheable|= UNCACHEABLE_DEPENDENT;
}
is_correlated= TRUE;
this->master_unit()->item->is_correlated= TRUE;
}
bool st_select_lex_node::set_braces(bool value) { return 1; }
......
sql/sql_lex.h
View file @ 54a713aa
......@@ -470,7 +470,7 @@ public:
void set_thd(THD *thd_arg) { thd= thd_arg; }
friend void lex_start(THD *thd, uchar *buf, uint length);
friend int subselect_union_engine::exec();
friend int subselect_union_engine::exec(bool);
List<Item> *get_unit_column_types();
};
......@@ -562,6 +562,8 @@ public:
query processing end even if we use temporary table
*/
bool subquery_in_having;
/* TRUE <=> this SELECT is correlated w.r.t. some ancestor select */
bool is_correlated;
/*
This variable is required to ensure proper work of subqueries and
stored procedures. Generally, one should use the states of
......
sql/sql_select.cc
View file @ 54a713aa
......@@ -158,8 +158,8 @@ static int join_read_prev_same(READ_RECORD *info);
static int join_read_prev(READ_RECORD *info);
static int join_ft_read_first(JOIN_TAB *tab);
static int join_ft_read_next(READ_RECORD *info);
static int join_read_always_key_or_null(JOIN_TAB *tab);
static int join_read_next_same_or_null(READ_RECORD *info);
int join_read_always_key_or_null(JOIN_TAB *tab);
int join_read_next_same_or_null(READ_RECORD *info);
static COND *make_cond_for_table(COND *cond,table_map table,
table_map used_table);
static Item* part_of_refkey(TABLE *form,Field *field);
......@@ -512,11 +512,12 @@ err:
DBUG_RETURN(-1); /* purecov: inspected */
}
/*
test if it is known for optimisation IN subquery
SYNOPSYS
JOIN::test_in_subselect
SYNOPSIS
JOIN::test_in_subselect()
where - pointer for variable in which conditions should be
stored if subquery is known
......@@ -550,6 +551,35 @@ bool JOIN::test_in_subselect(Item **where)
}
/*
Check if the passed HAVING clause is a clause added by subquery optimizer
SYNOPSIS
is_having_subq_predicates()
having Having clause
RETURN
TRUE The passed HAVING clause was added by the subquery optimizer
FALSE Otherwise
*/
bool is_having_subq_predicates(Item *having)
{
if (having->type() == Item::FUNC_ITEM)
{
if (((Item_func *) having)->functype() == Item_func::ISNOTNULLTEST_FUNC)
return TRUE;
if (((Item_func *) having)->functype() == Item_func::TRIG_COND_FUNC)
{
having= ((Item_func*)having)->arguments()[0];
if (((Item_func *) having)->functype() == Item_func::ISNOTNULLTEST_FUNC)
return TRUE;
}
return TRUE;
}
return FALSE;
}
/*
global select optimisation.
return 0 - success
......@@ -1016,9 +1046,7 @@ JOIN::optimize()
}
} else if (join_tab[0].type == JT_REF_OR_NULL &&
join_tab[0].ref.items[0]->name == in_left_expr_name &&
having->type() == Item::FUNC_ITEM &&
((Item_func *) having)->functype() ==
Item_func::ISNOTNULLTEST_FUNC)
is_having_subq_predicates(having))
{
join_tab[0].type= JT_INDEX_SUBQUERY;
error= 0;
......@@ -2512,6 +2540,9 @@ typedef struct key_field_t { // Used when finding key fields
when val IS NULL.
*/
bool null_rejecting;
/* TRUE<=> This ref access is an outer subquery reference access */
bool outer_ref;
} KEY_FIELD;
/* Values in optimize */
......@@ -2810,6 +2841,7 @@ add_key_field(KEY_FIELD **key_fields,uint and_level, Item_func *cond,
cond->functype() == Item_func::MULT_EQUAL_FUNC) &&
((*value)->type() == Item::FIELD_ITEM) &&
((Item_field*)*value)->field->maybe_null());
(*key_fields)->outer_ref= FALSE;
(*key_fields)++;
}
......@@ -2868,7 +2900,7 @@ add_key_equal_fields(KEY_FIELD **key_fields, uint and_level,
}
static void
add_key_fields(KEY_FIELD **key_fields,uint *and_level,
add_key_fields(JOIN *join, KEY_FIELD **key_fields, uint *and_level,
COND *cond, table_map usable_tables,
SARGABLE_PARAM **sargables)
{
......@@ -2881,28 +2913,54 @@ add_key_fields(KEY_FIELD **key_fields,uint *and_level,
{
Item *item;
while ((item=li++))
add_key_fields(key_fields,and_level,item,usable_tables,sargables);
add_key_fields(join, key_fields, and_level, item, usable_tables,
sargables);
for (; org_key_fields != *key_fields ; org_key_fields++)
org_key_fields->level= *and_level;
}
else
{
(*and_level)++;
add_key_fields(key_fields,and_level,li++,usable_tables,sargables);
add_key_fields(join, key_fields, and_level, li++, usable_tables,
sargables);
Item *item;
while ((item=li++))
{
KEY_FIELD *start_key_fields= *key_fields;
(*and_level)++;
add_key_fields(key_fields,and_level,item,usable_tables,sargables);
add_key_fields(join, key_fields, and_level, item, usable_tables,
sargables);
*key_fields=merge_key_fields(org_key_fields,start_key_fields,
*key_fields,++(*and_level));
}
}
return;
}
/* If item is of type 'field op field/constant' add it to key_fields */
/*
Subquery optimization: check if the encountered condition is one
added by condition push down into subquery.
*/
{
if (cond->type() == Item::FUNC_ITEM &&
((Item_func*)cond)->functype() == Item_func::TRIG_COND_FUNC)
{
cond= ((Item_func*)cond)->arguments()[0];
if (!join->group_list && !join->order &&
join->unit->item &&
join->unit->item->substype() == Item_subselect::IN_SUBS &&
!join->unit->first_select()->next_select())
{
add_key_fields(join, key_fields, and_level, cond, usable_tables,
sargables);
// Indicate that this ref access candidate is for subquery lookup:
(*key_fields)[-1].outer_ref= TRUE;
}
return;
}
}
/* If item is of type 'field op field/constant' add it to key_fields */
if (cond->type() != Item::FUNC_ITEM)
return;
Item_func *cond_func= (Item_func*) cond;
......@@ -3076,6 +3134,7 @@ add_key_part(DYNAMIC_ARRAY *keyuse_array,KEY_FIELD *key_field)
keyuse.used_tables=key_field->val->used_tables();
keyuse.optimize= key_field->optimize & KEY_OPTIMIZE_REF_OR_NULL;
keyuse.null_rejecting= key_field->null_rejecting;
keyuse.outer_ref= key_field->outer_ref;
VOID(insert_dynamic(keyuse_array,(gptr) &keyuse));
}
}
......@@ -3198,7 +3257,7 @@ sort_keyuse(KEYUSE *a,KEYUSE *b)
Here we can add 'ref' access candidates for t1 and t2, but not for t3.
*/
static void add_key_fields_for_nj(TABLE_LIST *nested_join_table,
static void add_key_fields_for_nj(JOIN *join, TABLE_LIST *nested_join_table,
KEY_FIELD **end, uint *and_level,
SARGABLE_PARAM **sargables)
{
......@@ -3210,12 +3269,13 @@ static void add_key_fields_for_nj(TABLE_LIST *nested_join_table,
while ((table= li++))
{
if (table->nested_join)
add_key_fields_for_nj(table, end, and_level, sargables);
add_key_fields_for_nj(join, table, end, and_level, sargables);
else
if (!table->on_expr)
tables |= table->table->map;
}
add_key_fields(end, and_level, nested_join_table->on_expr, tables, sargables);
add_key_fields(join, end, and_level, nested_join_table->on_expr, tables,
sargables);
}
......@@ -3290,7 +3350,8 @@ update_ref_and_keys(THD *thd, DYNAMIC_ARRAY *keyuse,JOIN_TAB *join_tab,
return TRUE;
if (cond)
{
add_key_fields(&end,&and_level,cond,normal_tables,sargables);
add_key_fields(join_tab->join, &end, &and_level, cond, normal_tables,
sargables);
for (; field != end ; field++)
{
add_key_part(keyuse,field);
......@@ -3312,8 +3373,9 @@ update_ref_and_keys(THD *thd, DYNAMIC_ARRAY *keyuse,JOIN_TAB *join_tab,
into account as well.
*/
if (*join_tab[i].on_expr_ref)
add_key_fields(&end,&and_level,*join_tab[i].on_expr_ref,
join_tab[i].table->map,sargables);
add_key_fields(join_tab->join, &end, &and_level,
*join_tab[i].on_expr_ref,
join_tab[i].table->map, sargables);
}
/* Process ON conditions for the nested joins */
......@@ -3323,7 +3385,8 @@ update_ref_and_keys(THD *thd, DYNAMIC_ARRAY *keyuse,JOIN_TAB *join_tab,
while ((table= li++))
{
if (table->nested_join)
add_key_fields_for_nj(table, &end, &and_level, sargables);
add_key_fields_for_nj(join_tab->join, table, &end, &and_level,
sargables);
}
}
......@@ -10784,6 +10847,13 @@ join_init_quick_read_record(JOIN_TAB *tab)
}
int rr_sequential(READ_RECORD *info);
int init_read_record_seq(JOIN_TAB *tab)
{
tab->read_record.read_record= rr_sequential;
return tab->read_record.file->ha_rnd_init(1);
}
static int
test_if_quick_select(JOIN_TAB *tab)
{
......@@ -10912,7 +10982,7 @@ join_ft_read_next(READ_RECORD *info)
Reading of key with key reference and one part that may be NULL
*/
static int
int
join_read_always_key_or_null(JOIN_TAB *tab)
{
int res;
......@@ -10928,7 +10998,7 @@ join_read_always_key_or_null(JOIN_TAB *tab)
}
static int
int
join_read_next_same_or_null(READ_RECORD *info)
{
int error;
......
sql/sql_select.h
View file @ 54a713aa
......@@ -36,6 +36,8 @@ typedef struct keyuse_t {
satisfied if val has NULL 'value'.
*/
bool null_rejecting;
/* TRUE<=> This ref access is an outer subquery reference access */
bool outer_ref;
} KEYUSE;
class store_key;
......@@ -455,10 +457,11 @@ class store_key :public Sql_alloc
Field *to_field; // Store data here
char *null_ptr;
char err;
public:
public:
bool null_key; /* TRUE <=> the value of the key has a null part */
enum store_key_result { STORE_KEY_OK, STORE_KEY_FATAL, STORE_KEY_CONV };
store_key(THD *thd, Field *field_arg, char *ptr, char *null, uint length)
:null_ptr(null),err(0)
:null_ptr(null), err(0), null_key(0)
{
if (field_arg->type() == FIELD_TYPE_BLOB)
{
......@@ -496,6 +499,7 @@ class store_key_field: public store_key
enum store_key_result copy()
{
copy_field.do_copy(&copy_field);
null_key= to_field->is_null();
return err != 0 ? STORE_KEY_FATAL : STORE_KEY_OK;
}
const char *name() const { return field_name; }
......@@ -516,8 +520,8 @@ public:
enum store_key_result copy()
{
int res= item->save_in_field(to_field, 1);
null_key= to_field->is_null() || item->null_value;
return (err != 0 || res > 2 ? STORE_KEY_FATAL : (store_key_result) res);
}
const char *name() const { return "func"; }
};
......@@ -547,6 +551,7 @@ public:
err= res;
}
}
null_key= to_field->is_null() || item->null_value;
return (err > 2 ? STORE_KEY_FATAL : (store_key_result) err);
}
const char *name() const { return "const"; }
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7