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

[Henri Sivonen <hsivonen@iki.fi>]

On Nov 30, 2006, at 12:49, Jirka Kosek wrote:

> Henri Sivonen wrote:
>
>> One thing that I have learned in the HTML5 conformance checking  
>> project
>> is that validation when augmented with Schematron and editing with  
>> RELAX
>> NG only call for different schemas.
>
> In what sense different? You have RELAX NG + Schematron and for
> validation you use both schemas, but XML editor will probably use only
> RELAX NG when doing code completition. What's the problem here?

Exclusions are easy to tack on in Schematron. However, for a RELAX NG- 
only editing schema, these need to be factored into the grammar,  
which causes an explosion of parallel productions for a given element  
that differ in terms of what the ancestors are at a given point. Even  
if you managed to generate or write such a schema, the validation  
error messages it would produce would be far less obvious than the  
ones produced by a simpler Schematron refinement.

ID/IDREF may be better than nothing for editing, but for validation,  
you'd want a Schematron schema (or Java code) that checks if the  
referenced element is of the right kind.

Etc.

>>> Well, according to this logic, shouldn't be then proponents of HTML5
>>> accused? ;-D
>>
>> Accused of what?
>
> Of breaking several best practices developed by markup community over
> the years.

Best practices that deny the realities of the Web aren't particularly  
good if applied to the Web.

> For example: defining new markup language instead of reusing existing,

Reuse only makes sense when what is being reused is suitable for the  
purpose for which it would be used.

> refusing schemas, ...

Well, we disagree here. I think it is good that the definition of the  
markup language isn't coupled with a requirement to use the spec  
writer's favorite schema technology or the fad technology of the day.

> I'm wondering if all syntax changes that HTML5 makes couldn't be  
> just achieved by producing more strict SGML declaration for HTML.  
> Does anyone did such analysis?

Defining HTML5 in terms of SGML would be useless, because SGML  
doesn't define a processing model that would be suitable for the real  
Web. The HTML5 spec would need to define its own parsing algorithm  
anyway.

> If not, I think that coming with a new grammar for HTML5 is just an  
> insane.

It is the only sane course of action given the Web realities. (From  
the context, I assume that by "grammar" you mean the low-level  
characters to parse tree syntax and not the high-level what element  
can occur where syntax.)

> Until my validator and editor are able to understand prose text I  
> prefer having schemas or other similar rigor formal definitions.

The WHATWG doesn't outlaw schemas for those purposes. It just makes  
them non-normative. That is, they are implementation details for  
particular applications--not part of the spec.

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