Joe Conway <mail@joeconway.com> writes:
> The only reason I can think to not apply it, is if you think
> we should completely disallow targetlist set returning functions as part of
> moving to SQL99.
I would like to eventually get rid of targetlist SRF's altogether.
I believe the feature represents a nontrivial drag on executor
performance and reliability even when it's not being used. (Look at all
the isDone cruft in execQual, the TupFromTlist hoop-jumping, the places
that are missing TupFromTlist hoop-jumping and should have it, etc.)
Obviously we can't do that until we have a fully functional alternative,
which FROM-list SRF's aren't. But if there is a chance of getting rid
of them via SQL99 extensions then I'd like to pursue that direction.
In the meantime, I don't see any need to spend any effort on cleaning up
what we're likely to discard altogether later...
regards, tom lane