RE: optional, but at least one required

> c) It makes the worst-case time complexity of subsumption checking
>    between two content models linear (as opposed to exponential).


Which makes me wonder just how important this very general capability to
restrict content models is. Particularly as the price you pay is to spell
out the parts of the model that you want to keep in the subtype, rather than
only saying which parts you want to drop, which makes it a maintenance
nightmare.

Sometimes I think that all you really need in practice is the ability to
modify individual element particles either (a) to prohibit them from
appearing at all, or (b) to restrict the type of their own content. 

Michael Kay
http://www.saxonica.com/