I don't have the original email to reply to. So I'll reply here. Overall, I had not really considered the primitive attribute case, but yeah that's clearly an issue. My mistake. A) I agree that the con here is huge. Not the best option B) Is close to better. We could certainly check this and throw an error. A better logical option is to do similar to what we do for ids and versions... on startup instantiate one of these and grab/store its initial state when freshly instantiated. We can later use those values to perform the empty check. This is more logical, but not sure how "practical" it is given that we do not really have a good place to keep this relative to an embeddable, nor relative to an embedded, aside from CompositeType, but that does not feel right. This is better in 6 as we have an actual runtime model representation of the embedded and embeddable - but of course that does not help us in 5 C) I really hate exposing `ComponentType` on a new SPI interface considering the type system is completely revamped in 6. This would be (1) a very short lived contract in this form and therefore (2) means yet another pain point for user upgrading 5->6. Ultimately I think this is the most promising solution moving forward (possibly coupled with the "expanded" B option).. However, that said, I am not sure we have a choice prior to 6 if we want to go this route - we'd have to expose the CompositeType.. There is no good singular thing prior to 6 to describe a embedded/embeddable. To clarify what sounds like a misunderstanding though, CompositeType is unique to each embedded usage, not embeddable. The CompositeType however does not expose its "role", so not sure if that distinction helps here. To make sure I understand your (D)... you mean to allow users to disable this option globally but to allow this option per specific embedded? Longer term (6) I think this is something else we ought to support as well as the inverse (globally opt-in, but allowing a local opt-out). That's not necessarily a strategy though for dealing with embeddeds that are "opted-in" that happen to use primitives. Yes, it allows the user to more actively and selectively manage that themselves, but if they happen to opt-in an embedded which contains primitives we have the same issue to deal with. Longer term I see allowing a mix of (B) (expanded), (C) and (D). For the short term (5), I think a the (possibly expanded) (B) option is the best. I say that because (a) I prefer to not add a new contract like this for a bug-fix release and (b) the concern about the immediate contract change in 6. If we all deem that this is acceptable, I's be fine with (C) as well. On Fri, Sep 1, 2017 at 4:14 PM Gail Badner <gbadner(a)redhat.com&gt; wrote: