I should explain a bit more... I'm absolutely open to the possibility of learning that there's something about the OO connection that makes one or another XSD approach more viable or optimal, and if such features are weighed against the benefits of referenceable (global) element declarations and found more important by the group, fine.

But, just as an example, in my work on the SAML schemas, I haven't heard any feedback from my JAXB colleagues that the choice of global vs. local makes much of a difference in the results of pushbutton class generation. (On the other hand, I have heard from them on things like wildcards and substitution groups vs. choice groups, none of which seems to be an issue here.) I should have thought to check with them before now, and will do so and report what I learn.

I have to admit that the discussion we've been having about global vs. local is making me think that I would have done SAML differently; currently all the elements there are global. What I might have done instead is make only the elements we deem reusable (assertions, requests, and responses mainly) global, with the rest being local and qualified; you can currently reuse an element in the middle of SAML's model, but doing so has no standardized semantics and is forbidden in prose.

However, since UBL is supposed to be a reusable library of well-defined semantics, any global element bound to a type has at least that much meaning, and could be useful for customizers in building new document types. (Do we need to better validate this use case?) But if OO practice doesn't interfere with it, then I would still be on the side of enabling element-level reuse.

Eve

Dan Vint wrote: