Mark Hollomon <mhh@mindspring.com> writes:
> I think it would be better to move the test further down, to just before we
> actually try to do the update/insert. Maybe into the heap access routines as
> suggested by Andreas.
I'm worried about whether it'll be practical to generate a good error
message from that low a level.
Looking at it from the DBA's viewpoint rather than implementation
details, I haven't seen a good reason *why* we should support
conditional rules for views, as opposed to an unconditional rule with
multiple actions. Seems to me that writing independent rules that you
hope will cover all cases is a great way to build an unreliable system.
regards, tom lane