> I am wondering if another
> answer couldn't be to define an optional order which would be followed
> by a class of Relax NG implementations to produce a deterministic type
> assignments on ambiguous schemas.
Yes, I think this is an possibility. This is what regex implementations do:
they allow non-deterministic models, but they define an order so that \1,
\2,... have well-defined values. Also I believe XSD does this for <union>:
the first choice wins (unless there's an xsi:type).
> I *think* (but may be wrong) that it shouldn't be much more complex than
> defining the order in which patterns should be checked and in the simple
> form of a Relax NG schema there are only 3 patterns with more than one
> non name class sub-pattern: group for which the order is obvious, choice
> and interleave.
I think this would be straightfoward for a naive backtracking implementation
to deal with, but I suspect it would be harder for a derivative-based
streaming implementation, but I haven't thouoght about it enough to be sure.
James
