Re: [xml-dev] Victory has been declared in the schema wars ...

[Henri Sivonen <hsivonen@iki.fi>]

On Nov 29, 2006, at 18:52, Jirka Kosek wrote:

> Henri Sivonen wrote:
>
>> I think not having a normative schema for HTML5 is a very good idea.
>
> I don't think so.

Why not? Isn't it good that a schema can freely be improved to more  
closely meet the normative prose?

Suppose the schemas James Clark wrote for XHTML 1.0 had been  
normative. Would it have been good it Petr Nálevka had been accused  
of breaking a normative schema when he made bug fixes and tightened  
the schemas with Schematron? Would it be a good idea if I'd be  
accused of breaking a normative schema if I tightened the schemas  
further by e.g. using the datatype library that I've been developing  
for HTML5? ( http://hsivonen.iki.fi/html5-datatypes/ )

>> None of the current schema languages (not even Schematron) are  
>> adequate
>> for expressing all the conformance requirements of HTML5.  
>> Experience has
>> shown that people become too schema-focused if there is a normative
>> schema: DTDs aren't adequate for expressing the conformance  
>> requirements
>> of HTML 4.01, but still people behave as though they had checked for
>> conformance when they have validated against a DTD.
>
> I think that problem is in education here -- it is not a big deal to
> show and explain difference between valid and conformant document.

To me, it seems that the WHATWG is interested in people producing  
conforming documents but not interested in a lesser notion of  
validity that is based on the arbitrary limitations of a particular  
schema language.

>> Still, HTML5 can be counted as a bullet point in favor of the  
>> RELAX NG
>> camp. The only schema for HTML5 in the works is a RELAX NG (Compact
>> Syntax) schema with companion Schematron assertions (
>> http://syntax.whattf.org/ ). A conformance checker for HTML5 requires
>> non-schema checking code to fill the gaps that RELAX NG and  
>> Schematron
>> can't fill (conveniently or at all). For example,
>> http://hsivonen.iki.fi/validator/html5/ checks for table cell  
>> overlap in
>> Java code--not in RNG, XSD or Schematron.
>
> Even with custom Java validation code you can't get 100% coverage.

By definition, machine-checkable conformance criteria can be checked  
with a Turing-complete programming language.

> Many requirements in HTML specification are too soft and have to be  
> interpreted by humans.

An HTML5 conformance checker isn't expected to check conformance  
criteria that is not machine-checkable, because doing so would be  
impossible. Still, an HTML5 conformance checker can do *much* better  
than a traditional HTML validator.

>> A lot of people use XSD for data binding. RNG doesn't do data binding
>> *by design*, so telling those people to use RNG for data binding  
>> doesn't
>> seem to help.
>
> RNG can do databinding and there are many tools that implement  
> this. And
> if you want do databinding you probably do not have ambiguous schemas.

I am aware of such tools, but the use them, grammar ambiguity has to  
be sacrificed, and grammar ambiguity is an excellent feature for  
validation. That is, only a subset of RELAX NG is suitable for  
databinding (the unambiguous subset), so if a schema is to be used  
for databinding, the schema has to be written specifically for that  
purpose at which point you are dealing with an application-specific  
dialect.

> On its latest meeting, RELAX NG Technical Committee was discussing
> whether RELAX NG 1.0 should be improved and developed in future
> (http://lists.oasis-open.org/archives/relax-ng/200611/ 
> msg00022.html). If
> you think that RNG should have better support for data-binding you can
> submit this request to TC.

I don't have databinding needs, so I have no intention to request  
changes in this area.

-- 
Henri Sivonen
hsivonen@iki.fi
http://hsivonen.iki.fi/