xml-dev - Re: Question About Namespaces and DTDs

[Rick JELLIFFE <ricko@geotempo.com>]

Norman Walsh wrote:
> 
> / "Paul W. Abrahams" <abrahams@valinet.com> was heard to say:
> | I like the idea of what you're trying to do, but giving particular PIs an
> | essential function in XML seems to me the wrong way to go.
> 
> Can someone point me to some arguments against PIs? There seems to be
> strong sentiment against them, and I don't understand why. Some things
> are exactly that, expectations that you wish to pass on to a processor.

The general arguments against PIs seem to be:
 - they are not elements (all we need are trees)
 - they are declarations (all we need are element start & end-tags)
 - they are not defined (how can we have interoperability if the meaning
is not known?)
 - they may be stripped (how can a generic transformation system know
whether to keep or
    discard a PI?)
 - they cannot have IDs, hence their addressability and nameability is
weak for
    hyperlink purposes
 - their scope is point (there is no mechanism to tie them to hierarchy)
and
   that limits the effect to piggybacking on containing structures
(entities
   in particular.)  

The general arguments for PIs seem to be:
  - they are not elements
  - they are declarations
  - they are not defined 
  - they may be stripped (they are useful point-to-point not end-to-end,
    in particular for server-side PIs)
  - they cannot have IDs (their effect is dynamic not static)
  - their scope is point-based.

Note that SGML'86 PIs do not have the same limitations as XML PIs: in 
particular, there is an attribute type PI which can appear on 
element and entity declarations. This allows PIs to have structured
scope.

There was a little session on Schemas v. DTDs at the WWW7 Developers
Day.
John Bosak asked the question whether we should augment DTDs or go
with instance syntax.  I was trying to defend DTDs in my presentation,
but the point was that there were so many additions to DTDs that we
might
need: I thought it was better to keep DTDs as they are, a nice 75/25%
solution
for structural schemas, and develop an instance syntax.  

I still think that
an instance-based syntax is correct, but the DOM WG would have saved
everyone
a lot of problems if they developed a DOM for DTDs as part of 
DOM 1: I think that has been one of the biggest things for holding back
XML development, and it entrenched the connection in people's mind
between
DTDs as a syntax and DTDs as a modeling technique.  If every DOM
provided
content-modeling structures, it would make deployment of schemas of
various
kinds much easier.

Rick Jelliffe