Bolt-in Type Systems (RE: [xml-dev] Relax NG and Web Services (formerly Joining the church))

["Bullard, Claude L (Len)" <clbullar@ingr.com>]

Somewhere in these interrelated threads, someone mentioned 
that the RELAX NG approach to type systems was to enable a 
designer to add their own.  How does that work?  Are there 
examples?

Some can pretend type issues don't exist or are purely private 
local affairs.  That really doesn't work for languages that 
initially are defined as object models and treat the XML 
or other encodings as just that, the bits on the wire 
representation of what an implementation of that object 
model needs to operate.  True extensibility for these is 
in the object model, and true interoperability as well. 

The X in XML is pretty worthless for that.  It must be 
able to express that an extension to the object model 
is needed and there are ways to do that (eg, modify 
the schema or DTD, add a namespace, add application 
language constructs such as X3D PROTOs), but using 
XML to design extensibility is backwards.  Using it 
to detect it in the document is not.  We shouldn't 
confuse the diet of the animal with the animal.

We can't get away from types.  Because it seems that 
for the object model languages, extensibility has to 
be worked out in the object model first then and only 
then worked out as an expression in the encodings such 
as XML application languages or in terms of the metalanguage 
(eg, namespaces), bolt-in types seem highly desirable and 
this could be a crucial feature for choosing which schema 
application languages (eg, XSD, NG, DTD) to normatively 
specify.

len