On Fri, Feb 6, 2009 at 9:50 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> The UNION arms have to all be the same data type in order to have
> restrictions pushed down through the UNION. You did not show us
> the table declarations for your first example, but I bet that updateid
> isn't the same type in both. (And yes, a domain is different from its
> underlying type for this purpose.)
I think you're right. The domain's in both cases (updateid and uri)
are bigints default nextval('something') not null;
and the r.history table's ones are just bigints not null. Same
underlying type, but not a domain. I'll try to alter it to domain
type, and see.
thanks.
--
GJ