Re: HTML is declarative on purpose [was: Web neurons ]

Daniel W. Connolly <connolly@w3.org> wrote in a post script:
> [...] Len Bullard writes:
> >   The HTML
> >model has limits that begin to show as more folks try
> >to add more functionality and cannot agree on it.  HTML
> >should be an architecture, not a DTD.
>
> Fair enough. Yet the burden of proof is still on the folks
> that want to see this happen.

It's already been done as a matter of fact: the latest beta version of
SP includes an "architecture engine" that can (among other things)
convert any SGML document conforming to HTML-as-an-architecture into
normalized HTML-as-a-document-type.  The chief drawback at this point
seems to be that you have to read an as-yet-unpublished annex to the
HyTime standard to figure out how to make an SGML document conform to
HTML-as-an-architecture in the first place :-); and the results won't
win anybody any "Cool Site Of the Year" awards; but other than that it
works great :-)

Actually using the architecture engine is not that hard; for the most
part you just have to specify #FIXED attributes in the DTD or an LPD to map
source element types to HTML element type forms.  It's easiest if the
source DTD is HTML-like to begin with; to make it easier for arbitrary
DTDs the only major changes required in HTML would be to:

    + allow ID attributes on all body elements;
    + introduce a general-purpose phrase-level element analogous
	to the general-purpose block-level element DIV;
    + (possibly) bring back the proposed SPOT element for
	identifying points and/or spans which cross element boundaries
	(this would only be needed to support 'spanloc' and 'dataloc'
	for document types conforming to both HyTime and HTML);
    + refrain from using the same attribute name to mean different
	things on different elements (e.g., TYPE, ALIGN, etc.)
	(this isn't even necessary, but it sure would make things
	easier...)

It would probably also be helpful to formalize the HTML location scheme
in SGML terms (e.g., HREF = system identifier + optional nameloc),
but that would be a matter for the architecture engine, not HTML itself.



--Joe English

  joe@art.com

Received on Wednesday, 29 May 1996 00:35:36 UTC