In MariaDB 5.3 the output from the EXPLAIN of the test case for lp bug #714999 is

MariaDB [test]> EXPLAIN SELECT f2 FROM t3 WHERE (
-> SELECT MAX( pk ) FROM t1
-> WHERE EXISTS (
-> SELECT max(f1)
-> FROM t2 GROUP BY f1
-> )
-> ) IS NULL ;
-------------------------------------------------------------------------------------------------------------+

-------------------------------------------------------------------------------------------------------------+

-------------------------------------------------------------------------------------------------------------+

So the expected number of examined rows from t2 is 2 and that is correct.

Yet in MariaDB 5.5 the EXPLAIN output for the same test case is
MariaDB [test]> EXPLAIN SELECT f2 FROM t3 WHERE (
-> SELECT MAX( pk ) FROM t1
-> WHERE EXISTS (
-> SELECT max(f1)
-> FROM t2 GROUP BY f1
-> )
-> ) IS NULL ;
----------------------------------------------------------------------------------------------------------------------------+

----------------------------------------------------------------------------------------------------------------------------+

----------------------------------------------------------------------------------------------------------------------------+

Here the name of the table in the third line is 'internal_tmp_table' and the expected number of the examined rows is 1.

Apparently this is not correct. This line of the EXPLAIN output is for the plan of the uncorrelated sub-query 'SELECT max(f1)FROM t2 GROUP BY f1'.

This change in the plan appeared after the patch that fixes lp bug #944706.

I consider this as a serious regression bug.