SDD Specifications Document

[Bob Morris]

Kevin Thiele wrote:

> Dear list_eners

> <XML data modeling spec proposal>...

I'm not competent to address the proposal, but I would like
to try to influence the directions that may emerge from it in
several ways:

(a).On theoretical and practical grounds, DTD's
are now understood to be a weak way to describe the schemata
(small 'S') of XML documents. The specialists should argue
about whether RDF or XML-Schema's (uppercase 'S') work
better (or try each on for size) and if there is a compelling
reason to have a DTD then derive one from the Schema. Aside
from (or perhaps actually because of) the lack of typing,
inheritance, and ordering mechanisms, DTDs add restrictions
on the kinds of schema inference that can be made from
particular XML documents that might be a query result, and
because of that make it less straightforward to mediate
between the answers to queries on two different databases.
These issues are explored in the book "Data on the Web" by
Abitoul et al., and in the literature behind it.

(b). Mixed content (a mixture of text and elements) is a really
bad idea, as is well known in the SGML community. Our own
limited experience has been that dealing with it in display code
necessitates to code obscurely. I can't tell whether anything
in the proposal would lead this way. Maybe some pseudo-code
XML-like examples would enlighten.

(c). One should keep in mind that queries can be constructed
by software, not just by people. The latter are far better than
the former at resolving ambiguity and supplying reasonable
defaults. So for example, we should always ask at the end if
we have given enough to formulate queries and interpret their
answers in a non-interactive fashion.

Put another way, data exchange is a minor reason to
adopt XML compared to its utility for the construction of
queries and the interpretation of their answers. (Here's a trivial
example, soon to be easily fixed: the data returned by the
experimental XML facility at ITIS*ca conforms to a DTD in all
but one case: that where there is no data matching the query.
In that case, it returns the same HTML "sorry no answer"
page that the html interface does, thereby requiring an XML
application to have special knowledge of this case---which
would not bother a human at all but sure startled our
little application.)

Bob Morris