RE: [xml-dev] XML should NOT be a new programming language

> From: Tim Bray [mailto:]
> Sent: Monday, March 04, 2002 1:42 PM

>
> At 10:52 AM 04/03/02 +0000, Francis Norton wrote:
> >Yes - you can implement XML interfaces between components in at
> least four ways:
> >
> >[1] pass an XML document
> >[2] pass an XML DOM
> >[3] pass a schema-equivalent programming language object (using
> tools for automatic schema <-> class conversion)
> >[4] pass an event stream
> >
> >Our experiences suggest that [3] and [4] are best for
> high-volume performance-critical server-side finance applications.
>
> You're buying some performance.  You're giving up a lot of the
> things that make XML worthwhile, in particular no binary
> dependencies on any particular hardware, OS, or whatever.
> Your call.  But it feels like a lousy bargain, architecturally,
> compared to [1].  -Tim

You're right that [2]-[4] are more tightly coupled (Using Larry
Constantine's definition of coupling).  That would lead to the conclusion
that [1] would lead to a better design.  However it's worth observing that
any of [2] and [4] can easily be wrapped to appear as [1].

I've been working with Sean McGrath on the XPipe system.  My hope is that
one outcome of the system will be that if one uses the framework to wrap
serialization/deserialization for DOM or SAX processing, then the framework
can locally optimize the linkage of adjacent pipeline components without
losing the general ability to pass documents.

David