Kurt Cagle wrote:
> Validation as a concept is generally poorly understood by the 
> aforementioned mainstream. Schematron in particular illustrates that 
> validation does not have to be complete (i.e., all possible assertions 
> that can be made of a given instance are in fact made), and that there 
> really is no such thing as one unique validation for a given XML 
> instance. XSDL tries to take a monolithic approach and succeeds about 
> 75% of the time, but even the XSDL community is recognizing that there 
> are constraints that simply cannot (and should not) be modeled in the 
> language. (I also keep wondering if Kurt Goedel's theorem isn't just 
> waiting in the wings in all of these discussions of (axiomatic?) 
> assertions).

FWIW, I'm a *big* fan of Schematron and of Relax-NG. I've worked on 
schema extensibility myself, as one of the authors of the Schema Adjunct 
Framework, which had some similarities to Schematron.

But this is all pretty abstract to the mainstream. And adding more 
levels of abstraction tends to lose the audience. After explaining 
validation, we then explain the differences between DTDs, Relax-NG, and 
W3C XML Schema, and why there is a compact syntax and an XML syntax for 
Relax-NG. Then we explain Schematron, followed by a reminder that some 
things do wind up being validated in code.

Of course, these are the same people we recruited for XML by telling 
them that it's all very simple, showing them how concrete it is, 
explaining how reading the documents gives you such a good understanding 
of what's going on. And when we do this, a lot of people conclude that 
XML is just too hard, that there are 5,184 new technologies you have to 
master in order to "really" use XML, and that some of these technologies 
are still being developed.

Jonathan