"Simon Riggs" <simon@2ndquadrant.com> writes:
> On Tue, 2007-01-30 at 17:34 -0500, Tom Lane wrote:
>> Since that is unlikely to be the case, I can't see that this is worth
>> implementing...
> Integers are typically used as keys...
Yeah, in the form of sequences, so you have a hole for every failed
insert. If the key isn't coming from a sequence then there's still
not any very good reason to suppose it's exactly contiguous. People
do delete entries.
> What would be wrong with checking for a NOT NULL constraint? Thats how
> other planners cope with it. Or are you thinking about lack of plan
> invalidation?
Yup, without that, depending on constraints for plan correctness is
pretty risky.
Basically what I see here is a whole lot of work and new executor
infrastructure for something that will be a win in a very narrow
use-case and a significant loss the rest of the time. I think there
are more productive ways to spend our development effort.
regards, tom lane