Re: Type and Structure Re: ASN.1 and XML

[James Clark <jjc@jclark.com>]

> Grammars should be an implementation technique,
> not the nub of the question.

Yes. I wouldn't claim that people should necessarily model their information
structures directly in a grammar-based language such as TREX or RELAX.  I
think it's a fine strategy for people to model in some higher-level language
(such as the component-oriented one which you sketched) and then map into a
grammar as one possible implementation technique.  However, whatever
mechanism you use for modelling the conceptual structures, you still need a
mechanism to describe the syntax, so that you can interchange the
representations of these structures.  Grammars are one such mechanism, which
have been proven by more than a decade of experience to be useful for this.
I certainly wouldn't claim that they are the best possible mechanism in all
cases; in some cases, mechanisms like Schematron may be better.

One way to look at a type in the context of XML is simply that it is a set
of XML documents.  In that sense, RELAX, TREX and indeed Schematron all
define types. I am afraid I still don't see how RELAX or TREX are
interposing a type system between the markup structure and the conceptual
model, except in a completely vacuous sense that is equally true of
Schematron.  I see RELAX or TREX as a mechanism for the describing the
syntax that you use to represent a conceptual model in XML.  I don't see how
that is a hindrance to conceptual modelling.

James