On ons, 2011-07-13 at 11:26 -0400, Tom Lane wrote:
> Dean Rasheed <dean.a.rasheed@gmail.com> writes:
> >>> On 7/12/11 9:46 AM, Andres Freund wrote:
> >>>> I guess $subject wasn't implemented because plain unique indexes aren't
> >>>> represented in pg_constraint and thus do not have a place to store
> >>>> information about being deferred?
>
> > I agree that expressing that using a UNIQUE constraint would perhaps
> > be more intuitive, but it would be new non-SQL-spec syntax that AFAICS
> > wouldn't actually add any new functionality.
>
> Our standard reason for not implementing UNIQUE constraints on
> expressions has been that then you would have a thing that claims to be
> a UNIQUE constraint but isn't representable in the information_schema
> views that are supposed to show UNIQUE constraints. We avoid this
> objection in the current design by shoving all that functionality into
> EXCLUDE constraints, which are clearly outside the scope of the spec.
I have never heard that reason before, and I think it's a pretty poor
one. There are a lot of other things that are not representable in the
information schema.