> One question I'm struggling with is whether to make most of
> the elements optional.
There doesn't have to be one schema.
It can be a bit tricky to manage multiple schemas for the same namespace
(because there are parts of the spec that seem to assume that by specifying
the namespace, you have identified the schema), but it's definitely part of
the design thinking that the validation rules you want to impose might be
stricter in some circumstances, weaker in others.
The right solution here might be a meta-schema that can be transformed
(using a set of input parameters) into the target schema you actually use
for validation. For example, this allows the list of enumeration values to
be varied.
Michael Kay
http://www.saxonica.com/
