Comment 3 for bug 727667

Exact location of the problem:

test_if_ref()/part_of_refkey() will indicate to make_cond_for_table() that condition "f3 = 83" will be always true, which actually is not satisfied by the (NULL,'r'), which will be returned by ref-or-null access.

when make_cond_for_table() considers a WHERE clause of:

   (f3 = 83) OR (f10 = 'z' AND f3 IS NULL)

and it is told that (f3 = 83) is universally true, it (correctly) concludes that entire WHERE condition is universally true, so there is no point in checking it.