I'm afraid this is what the current wording of "5.2.4. Assignability of raw and parameterized types" requires us to do.