Lists Home |
Date Index |
Of course to turn this on it's head, (the XML As
Complex System and the XML=WAP threads are converging),
we should step back as posters have hinted and
ask, if XML is not self-describing (really,
what does an XML instance say about its application
semantics), how should we be doing that?
Some current candidates:
1. Laissez-faire: send only the message. You
have an agreement or trust the other guy's process.
You might augment this with a checking service,
say one that goes out and compares prices to ensure
the other guy is giving you the best deal (time-sensitive).
2. Schema/DTD travels with message. That is the
SGML Way. It works if you want ultimate one-connect
shopping. It seems to be a big part of some implementations.
Best applied, IMO, to infrequent but noise-critical
communications. You send the test because you
absolutely want the receiver to understand this
3. Ask the Web: use RDF or some other expert system
what is needed. Isn't this sort of a dictionary? It
works as long as you own or accept the ontology of
others. This is Trust and Verify. Advantages?
4. Send the code. Java folks like that. This
is the ultimate, Trust Me, I Own The Process
approach. Augmented with a way to export the
information on request, it is easy for end users
despite any problems of performance, proprietary
If we say that meaning is discoverable, we play
right into the web services paradigm as I understand it.
Get only as much semantic as we recognize easily, negotiate
the rest, then document the negotiation results and hold
the communication process to those results. That this
is still in the main, contract-based document processing
shouldn't disenchant or disappoint. Humans have worked
like this for a very long time very successfully and
all this paradigm does is amplify that and take out
some noise. Given that organizations are competent
at creating their own definitions, the requirement for
standard schemata isn't critical, although it saves a
lot of work and things go quickly. That is pretty
much all any SGML project from CALS or even the old
GenCode and GML projects promised anyway.