Comment 6 for bug 586051

PrafullaT (prafulla-t) wrote :

Analysis:

Fix for Bug 586051

1. test_if_ref method which checks whether predicate is already evaluated
   due to ref/eq_ref access or not was incorrectly removing a predicate
   that was not implicitly evaluated due to ref access (due to presence of filesort ?)
   It was field=NULL predicate.
   Such predicate should be kept and execution engine will filter out rows
   correctly. Removal of such predicate led to returning of rows which had
   NULLs for join/predicate columns.
2. field COMP_OP NULL will always false for all fields except when COMP_OP
   is NULL-safe equality operator. Modified range optimizer to return zero
   row count in such cases.
   Query now does not even run. It returns zero result. As such Fix(1) is not
   required but we might hit that case in some other query (I have not tried it
   yet)
3. Fixed Field::val_str to print "NULL" for literal NULL instead of "0". It
   added lot of confusion while debugging. This has led to some result mismatch
   for subselect.test and might also show up in other tests . We need to update those
   results.