Jerry Sievers <jerry(at)jerrysievers(dot)com> writes:
> I believe what's happening here is that the server doesn't realize
> that the new column is going to have all nulls and that the check
> constraint allows nulls. As such, the check evidently is being
> evaluated for each row of the table.
Yup, that's right. There are some corner cases that make that harder to
optimize than it might look:
* volatile functions in the constraint might possibly deliver different
answers at different rows
* if table is in fact empty, we should not throw an error, nor indeed
evaluate the constraint even once (again, volatile functions...)
regards, tom lane