Your analysis is correct, Winnie. I'm a wonk from the
days when ISO 8879 was a draft and fought in the
"HTML didn't get it all right" wars. Forgive me
for being tendentious. Given the choices at the
time, XSD was the right technology. If done over
again, I'd take a hard look at alternatives such
as RELAX NG and Schematron, and the new ISO standards
for this.
<offtopic>
Atts are typically used for the domain or for singleton
primtive types such as ID that the XML processor must enforce.
Precisely as you say. Avoiding attributes is one of those myths
propagated in the early part of the XML development
when the parserGorillas dominated the savannah. It's
an old and well documented topic. They do create an
impedance mismatch with object-oriented systems that
serialize and deserialize if the developer wrongly
believes that object fields and element attributes
are isomorphic. It was a major stumbling block for
X3D specification work because it took the OOPMen
a very long time to accept that. VRML was a pure
object-oriented design. Where rendering and
behavioral fidelity are equally important, the
markup design has to be subordinate to the object
design. It isn't unusual to find multiple encodings
for those kinds of specs.
</offtopic>
len
From: Winfield Wagner [mailto:]
Claude Bullard:
You are correct on your analysis. You have to forgive me on the fact
coming from the GJXDM world I have been brain washed into thinking XSD.
Before my GJXDM days and today I would use the attribute method for
solving the reference system identity situation. It is describer for the
domain of the element not a real world element in itself. This has been
a very sticky point in GJXDM. Many of elements are descriptive
variations that define a context of an element (how it is referenced in
a domain) rather than a real world element.
