I checked description of the MySQL bug, and tried to merge the patch, it does not work (maybe wrong merge).

Need to talk to Timour about the problem.

reassiginig as we decided on IRC

I am not sure if MtSQL's fix is correct, but it is incomplete for sure. If the PK is changed to be only over 1 column, MySQL 5.6 produces the same incorrect result.
Example:

CREATE TABLE t1 (pk INT NOT NULL, i INT);
INSERT INTO t1 VALUES (0,10), (1,20), (2,30), (3,40);

CREATE TABLE t2b (pk INT NOT NULL, i INT NOT NULL, PRIMARY KEY);
INSERT INTO t2b VALUES (0,0), (1,1), (2,2), (3,3);

EXPLAIN
SELECT * FROM t1 WHERE NULL NOT IN (SELECT t2b.i FROM t2b WHERE t2b.pk = t1.pk);
SELECT * FROM t1 WHERE NULL NOT IN (SELECT t2b.i FROM t2b WHERE t2b.pk = t1.pk);
SELECT t1.pk, NULL NOT IN (SELECT t2b.i FROM t2b WHERE t2b.pk = t1.pk) FROM t1;

I am trying to understand if MySQL's solution is just incomplete, and the above case needs a separate solution, or if there is some generic solution that will address all cases.

• Re-merged after Monty's push
• Wasted some time to analyze a failure in ps-protocol mode of view.test, found out it fails in main 5.5, at the same time I didn't see the failure in buildbot.
• Pushed the merge into 5.5