RE: Convert DTD to IDL ?

["Frank A. Christoph" <christo@xxxxxxxxxxxxxxxxxx>]

>> Yes, I am working in a similar vein, I have been thinking of the
>> document instance as the state information for instantiating an
>> object and the CORBA/idl as defining the interface.


This is a very good idea that I have toyed around with myself.  The second
biggest problem with SGML in my mind is that it defines a data format rather
than an object interface.  This is precisely the reason that entities
references between documents are handled so badly, and HyTime tree pathlocs,
etc. are so useless.  If you define an interface to the document, then it is
possible to change the internal structure (alter the implementation) without
affecting document clients.  I think it is also possible to do something
like this with so-called architectures, but architectures are a mess without
even a hint of a well-defined semantics.

If this notion of document interfaces had existed when DSSSL was drawn up,
maybe DSSSL construction rules would be dispatching on the methods rather
than matching on the element names, etc.  DSSSL style sheets would be a
client of the document interface.

There is, incidentally, a dual way of attacking this problem; you abstract
over the "inputs" to the document rather than its "outputs" (the methods),
which turns it into sort of a very large lambda-function.  From this point
of view, DSSSL's construction rules are more sensible, because they resemble
pattern-matching as found in functional languages.  In fact, I have even
seen some work done on context-dependent pattern-matching, which looks a
little bit like a construction rule that matches a qualified element.

--FC




 DSSSList info and archive:  http://www.mulberrytech.com/dsssl/dssslist