>> Because the semantics of the schema language used, W3C Schema
>> 1.0, are not rich enough to express all of the constraints
>> for XSLT as an XML vocabulary.  I believe the RELAX-NG schema
>> semantics are rich enough.
>
> I don't think Relax-NG is rich enough to describe XPath expressions or
> XSLT
> patterns as a data type. So you will only ever describe a subset of the
> constraints.

I presume Ken was talking about element/attribute occurrence rules of the
kind that would be useful (adequate?) for an general interactive editor to
present to its human when editing XSLT. (Without agreeing or disagreeing
with him.)

> Also, there's a fuzzy boundary between syntactic constraints and semantic
> constraints (like: you can't call a template that hasn't been defined.).

Yes, reference rules have all sorts of interesting properties that are
rarely found in element/attribute occurrence rules, though of course value
rules are even less tame.

And I think everyone agrees that semantics has too much meaning.

> Any
> schema language can define all the syntactic constraints if you define
> "syntactic constraint" to mean "a constraint that can be expressed in the
> schema", and deem everything else to be semantic.

:-)

(The problem with grammars schema languages, as I see it, is that none of
them have taken the plunge into greater generality: as logical expressions
with cursors and location steps. RELAX NG has gone furthest here, in that
it treats element, attributes, fixed text fragments, empty and notAllowed
identically as particles.)

I think there are still some useful objective tests possible for judging
schema languages and their fit with a particular XML
language/vocabulary/document type/namespace. For example: does a schema
language allow optimal storage declarations to be derived from it? To test
this, take the schema and write C structs. Does a schema language allow
all rules concerning the context in which the element or attribute may
legitimately appear be expressed? To test this, highlight everything in
the normative text related to context that cannot be expressed and count
the sections.

(Indeed, we can count up the number of rules expressable in different
schema languages and judge between them. And it will all do no
good...until we reach the end of the new capital/development cycle,
because people won't trade their Edsel in for a Prius until they have
tried to squeeze the value out of the Edsel.)

So Ken's claim can be refuted by pointing out element/attribute occurrence
rules of an XSLT document that a (typical? state of the art?) interactive
editor would give to its users that could not be loaded from a RELAX
NG/NVDL schema. I don't know that this is unworkably prone to fuzziness,
is it?

It seems clear that XSD 1.0 proved itself inadequate to handle the
structures that are idiomatic in documents which are interpreted rather
than stored (SVG, XSLT, ODF, ...) But that should be no slur against XSD
unless we hold on to the myth of the one true schema language, which can
do everything.

Cheers
Rick Jelliffe