RE: breaking up?

[Michael Brennan <Michael_Brennan@allegis.com>]

> From: Simon St.Laurent [mailto:simonstl@simonstl.com]

<snip/>

> I used to think that data folks gravitated toward the Infoset, but it
> doesn't seem to be that simple.  I've talked with a few too many data
> people who find the layers built on top of XML 1.0 to be excessively
> complex to believe that statement works generally any more.  (I also
> know some document people who are very fond of the Infoset view of the
> universe, though not necessarily the PSVI.)
> 
> Among data people, there seemed to be two main concerns:
> 
> In some of the cases they were shuttling information between 
> dissimilar
> systems with wildly varying levels of support for schemas of 
> any flavor,
> and found that their expectations based on the stories of what schema
> will do were just plain wrong. (I wrote some of those stories 
> myself, of
> course.)
> 
> In other cases they found that it was easier for them to use 
> XML purely
> as a transfer syntax for labelled content, preferring their own
> validation and semantics mechanisms.  Sometimes they had lots of
> constraints which weren't easily expressed (except perhaps by
> Schematron, though some were just plain mathematically 
> difficult), while
> other times they had constraints they didn't want explicitly 
> published.
> Some just didn't like any of the options they had.
> 
> I've also talked with people who think that XML is great as a
> data-exchange syntax but that all of the processing beyond that simple
> syntax should be done with domain-specific tools, or UML, or similar
> non-XML systems.  They didn't seem to find UML any harder to share or
> use than W3C XML Schema for this work.

Well, given that my predominate use of XML is for application integration,
I'd say I'm in the "data" camp. I agree with all of your observations above,
though, except that I never had very high expectations for XML Schema. I see
its utility, but I was concerned with its complexity very early on. I really
wish the W3C had given us something like RELAX NG (combined with simple data
types), instead. Just because many applications are binding XML structures
to data types doesn't mean that those data types are somehow intrinsically
there in the document instance. But then, maybe I've just been influenced in
my thinking by others on this list.

I think a big part of the problem is people who conflate XML with particular
uses of XML. A clearer architectural vision that distinguishes between the
foundations of XML -- which should be purely based on syntax and the
contents of a document instance (no PSVI) -- and particular uses of XML
which may involve additional annotations and transformations would put us in
better shape, IMO. In this respect, I see the PSVI and XSLT as just two
extremes of the same spectrum. I don't see why adding things such as
defaulted attributes can't be viewed as a lightweight transformation to suit
a particular use of the XML (which means, of course, that they can't be
viewed as something intrinsic to the document instance, itself).
Unfortunately, XML 1.0 already set the precedent; it gave us DTDs with
defaulted attributes (among other things). In my mind, the problem starts
there, and I see no hope of that being fixed.

Too bad we can't go back and start over with the benefit of hindsight.