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.


mysql-test/r/subselect.result:
  BUG#8804: wrong results for NULL IN (SELECT ...): 
   - Updated test results
sql/item.h:
  BUG#8804: wrong results for NULL IN (SELECT ...): 
   - Added comments
sql/item_cmpfunc.cc:
  BUG#8804: wrong results for NULL IN (SELECT ...): 
  Made Item_in_optimizer to:
  - cache the value of "NULL IN (uncorrelated select)"
  - Turn off pushed-down predicates when evaluating "NULL IN (SELECT ...)"
sql/item_cmpfunc.h:
  BUG#8804: wrong results for NULL IN (SELECT ...): 
  - Made Item_in_optimizer cache the value of "NULL IN (uncorrelated select)"
  - Added comments
sql/item_subselect.cc:
  BUG#8804: wrong results for NULL IN (SELECT ...):
  - When needed, wrap the predicates we push into subquery into an 
    Item_func_trig_cond so we're able to turn them off when evaluating 
    NULL IN (SELECT ...).
  - Added code to evaluate NULL IN (SELECT ...) in a special way:
    = In [unique_]index_subquery, do full table scan to see if there 
      are any rows.
    = For other subqueries, change ref[_or_null] to ALL if the
      ref[_or_null] was created from pushed-down predicate.
sql/item_subselect.h:
  BUG#8804: wrong results for NULL IN (SELECT ...):
  - Added Item_subselect::is_correlated
  - Added comments
sql/records.cc:
  BUG#8804: wrong results for NULL IN (SELECT ...):
  - Make rr_sequential() non-static
sql/sql_lex.cc:
  BUG#8804: wrong results for NULL IN (SELECT ...):
  - Added st_select_lex::is_correlated and Item_subselect::is_correlated.
sql/sql_lex.h:
  BUG#8804: wrong results for NULL IN (SELECT ...):
  - Added st_select_lex::is_correlated
sql/sql_select.cc:
  BUG#8804: wrong results for NULL IN (SELECT ...):
  - Added KEY_FIELD::outer_ref to keep track of which ref accesses are 
    created from predicates that were pushed down into the subquery.
sql/sql_select.h:
  BUG#8804: wrong results for NULL IN (SELECT ...):
  - Added KEYUSE::outer_ref
mysql-test/r/subselect3.result:
  New BitKeeper file ``mysql-test/r/subselect3.result''
mysql-test/t/subselect3.test:
  New BitKeeper file ``mysql-test/t/subselect3.test''

mysql-test/r/subselect.result

@@ -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

+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

+--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

@@ -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

@@ -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

@@ -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.h

@@ -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

@@ -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

@@ -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

@@ -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

@@ -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

@@ -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"; }