Robert Haas <robertmhaas@gmail.com> writes:
> I get the impression that you think that there's a problem not only
> with the approach but with any approach whatsoever to that underlying
> problem.
Let's just say that the approaches proposed so far have performance
and/or functionality and/or code maintenance penalties that are utterly
unacceptable from the standpoint of people who don't need RLS. I don't
know if there is a workable solution, but I do know I've not seen one.
> With respect to selectivity estimation, do we have a live bug there
> now?
No, I don't believe so. Given that you'd like to get the planner to
call function XYZ, you could create an operator using XYZ and attach to
it one of the estimation functions that will actually call the
underlying function --- but you have to have call permission on the
function in order to create the operator.
regards, tom lane