On 03/14/2018 06:19 AM, Jeremy Finzel wrote: > Hello! From all that I can tell, it is not possible using a btree_gist > index as a primary key. If so, why not? I have a table with this gist > index which truly ought to be its primary key. as_of_date is of range > date type: > > EXCLUDE USING gist (id WITH =, as_of_date WITH &&)
I'm curious why you need a primary key on this table, especially if the exclusion constraint is already preventing duplicate/overlapping records?
Technically I think an exclusion constraint (or at least this one) fulfills the formal requirements of a primary key (is unique, isn't null), but maybe there are other primary-key duties it doesn't meet, like defining foreign keys that reference it. I've been on-and-off building an extension for temporal foreign keys at [1]. That is pretty new, but perhaps it will be useful/interesting to you. And if you have any feedback, I'd love to hear it!
But anyway, maybe if you shared why the table needs a real PRIMARY KEY, people here can suggest something.