Re: SD2 - Structured Attributes [fmt]

Tim Bray wrote:
> <editorial>
> I also used to be troubled by the very fuzzy and non-formal lines,
> in SGML, between what's an element and what's an attribute.  In fact,
> one of the early academic groups looking at SGML (led by Mamrak at
> Ohio/Columbus) proposed that there was no reason ever to use attributes
> for anything; I found that sensible, and still do.  Which is to say
> that if I'd been designing SGML, I wouldn't have put attributes in.

If we believe that the grove formalism is the appropriate one for
modelling SGML documents then the argument actually goes the other way.
Attributes are name/value pairs that describe the "attributes" of an
element. In the grove model, everything revolves around name/value pairs
called properties. In this group we are discussing if we can label
certain elements as being "really" attributes. In contrast the grove
model chooses particular properties and labels them as "content." This
suggests to me that if we (SGML people) are to align our language with
our own data model then we DO need structured attributes. That in turn
suggests a radically different syntax for both documents and DTDs. That
in turn suggests to me that 

      ****   THIS IS NOT THE RIGHT TIME OR PLACE TO DO IT ****

We have hard deadlines. We have SGML-86 compatibility targets. We have a
list of several workarounds that have worked for 11 years now. We have
working code. Some of those who want this in XML tend to want it for the
wrong reasons. If we put it in, it should be to fulfill the needs of
*documents*, not *relational databases*.
 
 Paul Prescod

Received on Monday, 19 May 1997 14:43:17 UTC