RE: Proposed restatement of syntax-based interoperability princip le ( was RE: Action item on syntax-based interoperability)

> From: Tim Bray [mailto:tbray@textuality.com]
> 
> I'm certainly not arguing that APIs and data models aren't 
> useful and necessary citizens of the software universe.  I'm 
> just saying that they do not provide a basis for 
> interoperability in networked information systems.  And the 
> evidence is on my side. -Tim

Others are saying that syntax alone, without some model of what it maps onto
and what the equivalence classes are, concrete syntax doesn't seem to
provide all that great a basis for interoperablity either.  NOBODY (at least
on this thread) has argued that data model or APIs alone can do the job, and
others have mentioned that COM/CORBA do have concrete syntaxes for the bits
on the wire specified, so their very real interoperability problems can't be
blamed on them trying to use APIs and data models as the basis for interop.


Where does that leave this discussion?  "You need to define a concrete wire
format specification to have data be usefully exchanged across
processor/platform boundaries" is not controversial.  "XQuery as a real
basis for interoperability in a large-scale networked information system is
doomed because it is based on a shared data model rather than concrete
syntax" is extremely controversial, but it seems to be an implication of the
draft text. (As I see it, XQuery has a real possiblity of interoperating
over the Web because it has the concrete syntax of the language itself and
the XML serialization of the results, and these are made meaningful by their
reference to the shared data model). And as Jonathan has pointed out, one
can't duck the issue by asserting that XQuery is not meant for the Web,
since it is explicitly chartered to "extract data from real and virtual
documents on the Web."

Sure, COM, CORBA, and SQL (not to mention implementations of the XQuery
drafts!) are not terribly portable/interoperable in actual practice.  Is
that due to some architectural limitation because they go too far in
assuming a shared data model?  Interesting hypothesis, but the null
hypothesis that they don't interoperate because there are multiple versions
of the standards out there, vendors have little business incentive to make
implementations interoperate, and lots of incentive to embrace and extend
the standards, can't be rejected with the evidence at hand.

Received on Saturday, 25 October 2003 13:52:16 UTC