🚀 go-pugleaf

  I have read Erik's latest missive, and the sections of Goldfarb pointed 
out by him, with some attention.  To use his own phrase, however, I can't
make head or tails of it.  In some places he appears to be reading as much
of what he wants into Goldfarb as he does with my postings:

>An SGML document is, by the very definition of "SGML document" (see
>definition 4.282, and Chapter 4, pp 132ff), application-dependent, and
>it has to be.  

  Definition 4.282 says absolutely nothing about applications, let alone 
that SGML documents are application-dependent.  Moreover, at least some 
advocates of descriptive markup think they're building documents that are 
application-independent, at least to some extent (if memory serves, the 
Coombs et al. CACM paper claims this).

>An SGML document includes a reference to (or includes) the document
>type declaration, which, in sufficient detail to allow parsers to
>process the document, defines the application.  

  This also seems to be wrong.  4.279 says that the application is the 
"rules that apply SGML to a text processing application", it does not 
say that the DTD defines the application.  In fact in the notes it quite 
explicitly mentions several other things, such as link process definitions.

  Apart from minutiae, Erik seems to have missed the whole point of the 
discussion between Eliot and myself.  I wanted to draw attention to the
fact that SGML enforces certain relationships and permits arbitrary
other ones outside of itself, and that the latter should not be counted
as "being represented in SGML" (hence the overlap problem).  I wanted to 
make the further point that structure moved outside what SGML can enforce
is structure that is potentially application-dependent (hence the claim
that SGML would lead to non-portable documents, non-portable because
their structure may have meaning only to a single application, perhaps
even only a single implementation). 
  
  If this problem is all in my head, as Erik seems to think, then why did 
both Steve and Eliot propose upgrades to SGML to address the overlap example?
And did anyone else notice that neither of them found it useful to resort
to HyTime or some other standard?  It makes me wonder if people are really
thinking when they dismiss the problems I raise by saying "SGML wasn't
meant to do that, do it in X".

-Darrell.