"Richard Harris" <richard_haris@adp.com> writes:
> Is PG 8.2.4 behavior correct?
Nope :-( ... looks like some sorta bug in the rules for when outer
joins can be rearranged. It's doing the t1a/t1b join first, and
then taking that as the right side of the join to t2, which means
that the t1b column goes to null if there's not a match to t2.pkt2.
Needs some study ...
regards, tom lane