"Brendan Jurd" <direvus(at)gmail(dot)com> writes:
> So, if I read you correctly, in summary we'd like to:
> * make SET LOCAL local to the transaction (i.e., make it behave as documented),
> * abandon the idea of a subtransaction-local SET, because the new
> function-local SET takes care of the interesting use-cases for that,
> * somehow deal with the incompatibility with the 8.2 "security
> definer" workaround.
> Tom's proposal to handle the latter was that when a function-local SET
> reverts, it overrides any inner SET LOCALs.
> Am I on the right page?
Got it in one, I believe.
regards, tom lane