From: Arjun Ray <aray@nmds.com.invalid>
Newsgroups: comp.text.sgml
Subject: How useful is a NOTATION Declared Value?
Date: Wed, 08 Nov 2000 00:29:34 -0500
Organization: FUDGE Dispersal Systems
Message-ID: <5llh0tkp0vrg44tgrtjmvdhbfu9j55t9ho@4ax.com>


ISO 8879 Clause 7.9.4.4 (p336-7 in the Handbook) turns out to be more
confusing than illuminating, at least to me.  The NOTATION declared
value strikes me as broken in a number of ways.

First, there's no explanation of why the attribute value is restricted
to a specific name group of notation names.  How is this useful if the
content is going to be passed off to *some* notation processor anyway?
Shouldn't requiring a declaration of the notation be enough?
 
Second, there seems to be an implicit assumption that the element type
will have a content model of (#PCDATA) or declared content of either
CDATA or RCDATA.  What sense could there be in a NOTATION attribute
for an element type with mixed or element content?  Since the standard
doesn't rule this out, shouldn't there be an explanation?

Actually, it's worse than that.  The example on p.336 of the Handbook
looks like this 

<!NOTATION cgmbin   SYSTEM "Program to interpret CGM binary notation">
<!NOTATION cgmclear SYSTEM "Program to interpret CGM clear text">
<!NOTATION iges     SYSTEM "Program to interpret IGES clear text">

<!ENTITY   fig42    SYSTEM "fig42.cgm" NDATA cgmbin>

<!ELEMENT figure    - o    CDATA>
<!ATTLIST figure
          notation  NOTATION (cgmbin|cgmclear|iges)  #REQUIRED
          location  ENTITY   #CONREF >

With two instance uses:

 <figure notation=iges> Some IGES stuff </figure>

and 

 <figure location=fig42>

Problems:

  1. In the second example, a #REQUIRED attribute gets nuked on the 
     grounds that it's irrelevant/redundant.  Huh?  Why was it 
     #REQUIRED then?  (I think what's happening is that #REQUIRED is
     a kludge for a semantic that really can't be expressed cleanly.)

  2. Obviously 'cgmbin' can't be a value of the attribute for inline
     content (without at least some form of encoding of the binary 
     data.)  On the other hand, the data content notation of a
     #CONREF-ed entity will be binding anyway - and the standard
     doesn't require that notation to be consistent with the NOTATION
     attribute.  So why is 'cgmbin' mentioned in the attribute
     definition at all?

Third, assuming nevertheless that the NOTATION attribute is indeed
saying something about the data content of its associated element
(type) - does the standard actually say this anywhere? - there is no
way to take advantage of the notation's data attributes as one could
with an external entity declaration.  (The DAFE facility requires too
much administrative machinery to be particularly useful, and at any
rate doesn't have a clean answer for the nameclash problem, not that I
can see.)

Is there room to re-engineer this kind of attribute? (Or rethink the
ways in which one might attach notated semantics to an element type in
general?)


-- 
:ar