"Alex Hunsaker" <badalex@gmail.com> writes:
> Currently this loops through all the constraints for a relation (old
> behavior of MergeAttributesIntoExisting)... Do you think its worth
> adding a non-unique index to speed this up?
No. If we were to refactor pg_constraint as I mentioned earlier,
then it could have a natural primary key (reloid, constrname)
(replacing the existing nonunique index on reloid) and then a number
of things could be sped up. But just piling more indexes on a
fundamentally bad design doesn't appeal to me ...
Will review the revised patch today.
regards, tom lane