SGML objects

Lou Burnard (lou@vax.ox.ac.uk)
Tue, 17 Aug 1993 13:49:00 +0200


A snappish response to Dave R's recent pronouncements on SGML:

> There are mechanisms for
>"public" identifiers (similar to URNs but run by the ISO and probably
>unsuitable for widespread use in the Internet), and "system" identifiers
>supposedly for local files. The "proper" SGML approach forces you to declare
>all such references and to then use local names in subsequent markup. HTML of
>course flouts this by including the URL directly in the elements defining
>hypertext links.

Why do you think ISO public identifiers are unsuitable for widespread
use in the Internet? Is this a case of "not invented here"?

>In general SGML is quite a weak specification language, and not as good
>as simple BNF grammar formalisms.

This is simply meaningless. Not as good in what respect? Weak with
respect to what?

>The supposed ability to use common SGML parsers to interchange document
>formats is dependent on the formats involved having closely matching
>semantic models.

SGML is not about interchange of formats, but about interchange of
document structure. The fact that HTML+ cannot distinguish these
fundamentally different properties of a text is the strongest argument
against its widespread adoption and the basic cause of the endless
confusion evident throughout recent discussion of it.

> The SGML standard makes some provision for including procedural
>information but this is not present in the DTD's for each format.

The SGML standard allows you to associate procedural information with
parts of the document structure within a given dtd if you really want
to (using so-called "processing instructions" is one way; using Linkset
declarations is another). Most people would prefer that all such
information should be standardized in a separate document or database
entirely, using the new ISO 'Document Style Semantics Specification
Language' (DSSSL).

And while I'm being obnoxious: the plural of "DTD" is "DTDs"

Lou Burnard