On May 24, 2008, at 2:18 AM, ando@sys-net.it wrote:

I believe the actual implementation should be... implementation dependent :), provided it is consistent.

I agree there is a slight inconsistency.

A client can delete a listed superclass but not add an superclass of a listed attribute.

One could argue that both be considered invalid operations. However, it would be reasonable to consider an add of an unlisted superclass and a delete of a listed superclass to be valid. However, a add of an already listed class should be invalid and a delete of a unlisted class should always be invalid.

-- Kurt