Jeff Trout <threshar(at)threshar(dot)is-a-geek(dot)com> writes:
> I just ran across this, and I do not think it is entirely a PG bug or
> even something that the backend can detect and handle.
> The problem stems from swapping a table definition from under a
> function.
Hmm. This should yield an error (SQL function not returning the type
it claims to), and we probably should plug the hole by invoking
check_sql_fn_retval every time not just at creation.
I thought you were about to complain about plpgsql, which has much worse
problems due to plan caching...
regards, tom lane