> Or have I missed something and is it more complex than this?
Sometimes, even if the grammar is unambiguous, it is hard to assign
types in a streaming fashion. Consider this:
<start>
<element name="root">
<choice>
<ref name="foo1"/>
<ref name="foo2"/>
</
</
</
<define name="foo1">
<zeroOrMore>
<element name="foo"><empty/></element>
</zeroOrMore>
<element name="one"><empty/></element>
</define>
<define name="foo2">
<zeroOrMore>
<element name="foo"><empty/></element>
</zeroOrMore>
<element name="two"><empty/></element>
</define>
This grammar is unambiguous because a valid instance can only yield one
interpretation. However, it's very hard to keep track of type assignment
to each element.
I guess in the world out there, most of the people write unambiguous
schema. So my feeling is that it's OK if we can guarantee that
1. there is a well-defined subset of RELAX NG,
2. the type assignment of those schemas are guaranteed to be unique, and
2. it can do a type assignment by a streaming processor
regards,
--
Kohsuke KAWAGUCHI
Sun Microsystems
