Re: Personal reply to Edd Dumbill's XML Hack Article wrt W3C XML Schema

[Eric van der Vlist <vdv@dyomedea.com>]

Matthew Gertner wrote:
> 
> > I like this vision that is so coherent with the processing
> > model of XML
> > through pipelines of XSLT transformations.
> >
> > This is adding value without breaking anything!
> 
> While it's true that the inventors of XML were prescient in providing the
> option to use well-formed documents independently of schemas or DTDs
> (considering the apparent difficulty in getting anything resembling a
> universally accepted schema spec out the door), the kind of approach that
> you are proposing runs counter to any kind of coherent long-term view of
> what XML is all about. I can't see any scenario where semantics as
> fundamental as data types would vary for a given XML element type across
> instances.

This is a "may", not a "should" and data types may already vary between
instances since I can already attach different schemas to a single
document.

> Element types have specific semantics that don't very across instances. I
> would have thought that this would be one of the few universally accepted
> truths in the XML world (wishful thinking, I guess). 

I wonder if I have been clear ;=) ...

By specifying the data types in the instance, I wasn't thinking to ask
the users to type it (except, maybe in very specific cases, but it's
already possible using a xsi:type attribute), but rather defining the
validation process as a transformation that would add this information's
as attributes (or elements).

This would enable using other techniques than W3C XML Schema (including
hand written XSLT transformations) to construct a "typed infoset" and
this typed infoset could even never been materialized but only
constructed as SAX events or DOM trees.

> Specifying data types in the instance for convenience runs directly 
> counter to this. In this case I can't even, say, generate Java classes 
> from schemas to process XML instances conformant to the schema (a la Castor, 
> Commerce One XDK, Alphaworks Beanmaker and so forth), since the data types 
> might vary across instances. Say good-bye to one of the few truly 
> forward-looking XML-based processing models. And there's an infinite amount 
> of additional semantic information that I can include in a schema to bind to 
> data sources, web forms, SOAP-enabled applications, etc., etc. Including all 
> of this as attributes in each instance hardly seems practical for too many reasons to
> even start contemplating.

No, you could still do all this, this typed infoset would just be
another way to convey type information.

> Are people truly attracted to this idea, or is it just an exaggerated
> response to the admittedly frustrating warts on the XML Schema spec?

I do think it would make sense.

Eric
 
> Matthew

-- 
Rendez-vous à Paris pour net2001.               http://www.mynet2001.net
------------------------------------------------------------------------
Eric van der Vlist       Dyomedea                    http://dyomedea.com
http://xmlfr.org         http://4xt.org              http://ducotede.com
------------------------------------------------------------------------