Suppose we have a query plan with table
tblX being accessed via the
ref access method:
tblX.key_part1 = expr1 AND tblX.key_part2 = expr2 AND ...
Before performing an index lookup, we determine whether any of
the expr values
is iNULL. If it is, we don't perform the
lookup, but rather immediately return that the matching tuple
is not found.
This optimization reuses the null_rejecting
attribute produced by the early NULLs
filtering code (see
Section 7.3.1.1, “Early NULLs Filtering”). The check
itself is located in the function
join_read_always_key().