UML22 — inconsistency wrt UML2 classifier behavior

Figure 13.6 - Common Behavior (page 412 of formal/05-07-04) shows BehavioredClassifier's ownedBehavior as a composition (black diamond) and it shows classifierBehavior as a directed association (no diamond).

no problem so far.

But then the figure also shows classifierBehavior subsets ownedBehavior, and the text says (page 420, section 13.3.4 BehavioredClassifer|Associations) that classifierBehavior specializes BehavioredClassifier.ownedBehavior).

If classifierBehavior is a specialization and the set of its instances is a subset, then the metaassociation denoting classifierBehavior should have the same association type as the superset, in other words for conssitency, both or neither should be black diamond.

My assumption here is a form of the covariance thesis, a subset and specialization of a composition must also be a composition.

In the UML 2.2 Specification, Behaviored::classifierBehavior is no longer specified in the text as "specializes BehavioredClassifier.ownedBehavior". Instead, it simply notes "subsets BehavioredClassifier::ownedBehavior", which is consistent with the diagram.
The subset classifierBehavior association doesn't need to be composite, because it implies the superset ownedBehavior one (for the same behavior instance), which is composite, so composite semantics will apply anyway. Deleting an M1 classifier instance will delete the M1 behavior instance linked by the subset association, because that M1 classifier is also linked by the superset composite association.
Revised Text:
None.
Disposition: Closed No Change