From: Arjun Ray <aray@nmds.com.invalid>
Newsgroups: comp.text.sgml
Subject: Generalizing IDLINK
Date: Sat, 02 Dec 2000 14:26:59 -0500
Organization: FUDGE Dispersal Systems
Message-ID: <hlhi2tstiqmfh94eesok1tp4421elj3rfg@4ax.com>


[ Hello?  Anyone home?  ;-) ]

The following straw proposal is motivated by two considerations:

1. IDLINKs in LPDs have the obvious disadvantage that they make sense
only in internal subsets.  They exist, however, only to provide a
means for greater specificity in link rule selection than that offered
by just the GI.  IOW, it's the specificity that's important: using an
ID as the hook could be overkill.

2. A link set can have more than one rule for a GI.  This is to let
the application decide, on the basis of the relevant attributes, which
particular rule should apply.  Quite often, though, this decision can
be based on a single attribute and at that, can amount to mechanical
inspection of the value only.  (This is analogous to the way in which
the CLASS attribute in HTML is "specialized" for CSS.)

So, just as an IDLINK declaration looks like this (for implicit links)

 <!IDLINK 
     foo  some-gi [ ... ]
     bar  some-gi [ ... ] >

We could have a declaration that looks like this

 <!ATTLINK  attname
     attval1  gi1  [ ... ]
     attval2  gi2  [ ... ] >

or, more formally, comparable to [168.1]:

  Attribute link set specification =
    MDO,
    "ATTLINK",
    ps+,
    name,
    ( ps+,
      attribute value specification,
      link rule )+,
    ps*,
    MDC

[Parenthetically, why is the (source) GI needed in an IDLINK -
shouldn't it be #IMPLIED when the ID uniquely identifies it anyway?]

Since there aren't too many implementations of the LINK feature
anyway, might this not be an opportunity to reconsider its design?
 

-- 
:ar