Greg Stark <gsstark@mit.edu> writes:
> Hm. If you incorrectly mark your function as IMMUTABLE even though it
> has side effects then the planner may indeed collapse this. Does the
> planner know it can't collapse views if the underlying tables aren't
> accessible to the user?
There are no cases where function or view collapsing elides permissions
checks (if you have a counterexample please provide it!!). They could
change the time at which permissions checks are applied, though; which
has the potential for a REVOKE to not disallow execution of already-
planned queries that ideally it should prevent. I believe that this
risk will be fixed by the planned forcing of replanning after schema
changes.
regards, tom lane