Re: Personal reply to Edd Dumbill's XML Hack Article wrt W3C XML Schema

[Rick Jelliffe <ricko@allette.com.au>]

From: Henry S. Thompson <ht@cogsci.ed.ac.uk>

>James Clark <jjc@jclark.com> writes:

>> ...I am content to have XML Schema as it stands now, so long
>> as that doesn't prevent the XML community getting something better in
>> the future.  In particular,  I hope that:
>>
>> (a) overly tight coupling between other W3C specs and XML Schema 1.0
>> will be avoided
>
>I hope so.

Not a dry eye in the house!  But what is "over-coupling"?

I suggest that requiring a subsequent general spec (i.e. XPath and XSLT but
not XML Queries) to use _any_ PSVI features that cannot be expressed as
elements or attributes is over-coupling.  I would guess Henry would tend to
judge no, and, from the TREX record, I would guess James would tend to judge
yes.

If XML Schemas 1.0 had weighed more highly issues of idiom, localisation,
notations, modularization, richer selectors, co-occurrence constraints,
verbosity, etc, and all the other points it is currently weak in, it would
be more suitable as _the_ basis for future general specs.

But it has not, and there are some good-enough reasons why it has not. The
best reason is that it is better to put out something sooner rather than
later, and notify everyone that XML Schemas 1.1 will be come hot on its
heels.

XML Schemas 1.0 has been under development for almost 3 years, and in the
last year at least there has been extensive consultation with the public: I
don't believe there is any other W3C spec for which there has been even 1/8
as much public discussion--the file of Last Call comments took over 1/2 meg
and there have been copious comments before and after. Indeed, I would be
surprised if any open source spec has had the kind of thorough public
comment and formal answer mechanism that XML Schemas has had.

(That could be criticized as "too much, too late" in that the requirements
were already set and the basic approach already taken, but even at the start
the early drafts reflected public awareness by synthesizing existing schema
languages such as SOX etc. so there was some kind of reflection of 1998
conceptions of schema languages. )

During 2000 I repeated called on XML-DEV asking people to make comments, so
there has been no secrecy to this process, and there should be no surprise
that the review process must end. So I hope we do not manage to bog
ourselves down in discussing things that are out of our hands to control
(i.e. XML Schemas 1.0) rather than the things that we can constructively
influence (PSVI usage, XML Schemas 1.1, XML Schemas 2.0, XML Schemas
subsetting and enhancement, XPath, XSLT).

If you don't like XML Schemas 1.0, make up your own schema language! I did:
there are lots of different data models on which you could create your own
schema language still--relational grammars, entity-relation,
everything-is-a-tree, instance-specification-of-facets etc.  If you prefer
XDR/SOX/RELAX/TREX/DT4DTD's approach, use them. And then, if you have the
energy to be co-operative, articulate and lobby why XML Schemas 1.1 or 2.0
should be reformulated to cope with them.

>> (b) future versions of XML Schema will not be constrained to be 100%
>> compatible with XML Schema 1.0; the requirement for any future version
>> should be that it be possible automatically to translate XML Schema 1.0
>> into that future version
>
>Speaking personally, I don't believe any such requirement has been
>stated, and for a range of reasons, including the one below, I agree
>it should not be.

Since the details of XML Schemas took so long to nut out, I don't think
anyone involved is surprised if the public finds the results non-obvious.
Finding a comprehesive solution to schemas is not a simple task.  But it is
made impossible if people have the expectation that XML Schemas is a
Universal Schema Language-to-end-all-Schema-Languages.

If people are confused that there are various "rival" approaches and
technologies around, the heart of their confusion lies in that expectation:
universality rather than comprehensiveness.  That there may be more than one
schema language in existence and use should be as surprising as there being
both cars and trucks on the road. But it is difficult to hose down people's
unrealistic expectations if, at the same time, there is the push at W3C to
use Schemas PSVI as some sort if universal basis for the new generation of
general-purpose technology.

The bottom line: we (users, W3C, marketers) should treat XML Schemas 1.0 as
a well-made, interregnal, comprehensive schema language not the mandatory,
ultimate, be-all-and-end-all, universal schema language of fantasy.

Cheers
Rick Jelliffe