Thread: Exclusion constraint with negated operator?

Exclusion constraint with negated operator?

From
Torsten Förtsch
Date:
Hi,

I wanted to create an exclusion constraint like

EXCLUDE (c1 WITH =, c2 with <>)

This gives an error:

operator <>(integer,integer) is not a member of operator family "integer_ops"

I can resolve that by using gist and the btree_gist extension.

My question is can I somehow express something like

EXCLUDE (c1 WITH =, c2 with NOT =)

It seems that's not possible at the moment. But is there any obstacle in principle or is it just not implemented?

We have a ton of overlap operators (geometric, ranges, arrays) but none for the opposite. Sometimes that's useful for exclusion constraints.

Thanks,
Torsten

Re: Exclusion constraint with negated operator?

From
Tom Lane
Date:
=?UTF-8?Q?Torsten_F=C3=B6rtsch?= <tfoertsch123@gmail.com> writes:
> My question is can I somehow express something like
> EXCLUDE (c1 WITH =, c2 with NOT =)
> It seems that's not possible at the moment. But is there any obstacle in
> principle or is it just not implemented?

Well, it'd likely be a bad idea.  Indexes are meant to help you quickly
find a small part of a table that satisfies a condition.  Finding the
probably-much-larger part of the table that doesn't satisfy the condition
is something they are bad at.  This is why "=" is an indexable operator
while "<>" is not.  It's not impossible in principle for "<>" to be an
index operator, but the set of cases where indexing on such a condition
would beat a seqscan is likely to be uselessly small.  By the same
token, EXCLUDE constraints using such a condition would be unpleasantly
inefficient.

            regards, tom lane