> it leaves it up to the implementor to decide which
> references *are* needed for validation, and as far as I can
> see, an implementation that needs *all" references is
> conformant according to the letter.
>
Just to add to this with a concrete example: if a content model refers to
element E, then a validator needs to know all the elements in the
substitution group of E. If there's an element Z in the schema whose
substitition group membership is a dangling reference, do you say "I don't
know whether element Z is in the substitution group of E, therefore I can't
validate", or do you assume that Z isn't in the substitution group of E.
Same applies to a complex type derived by extension from an unknown base
type. I'm pretty sure the spec doesn't give you the answer to this.
Michael Kay
http://www.saxonica.com/
