"Mendola Gaetano" <mendola@bigfoot.com> writes:
> but you agree anyway that should not possible delete
> a user that own DB object or at least change the owner
> to postgres?
If you do it the approved way, with DROP USER, you'll find there's
already an interlock:
regression=# create user foo;
CREATE USER
regression=# create database foo owner foo;
CREATE DATABASE
regression=# drop user foo;
ERROR: user "foo" cannot be dropped
DETAIL: The user owns database "foo".
regression=#
When you issue direct UPDATE/DELETEs on the system catalogs, consistency
is your responsibility.
regards, tom lane