Heikki Linnakangas <heikki.linnakangas@enterprisedb.com> writes:
> I can see two possible interpretations for this:
> 1. The subquery pull-up code is broken, the transformation of a
> whole-row reference to ROW(...) is not valid.
I think the problem is probably that we need a PlaceHolderVar wrapper
around the ROW() constructor.
> 2. The semantics of STRICT with row arguments is broken. It should be
> made consistent with IS NULL.
Well, that's a different argument. The immediate problem is that this
case doesn't behave consistently with pre-8.4 behavior, which was not
an intended change --- so I think we'd better make it work like before.
regards, tom lane