Message text written by Murata Makoto
>
I don't think you have spoken with database people. When I gave a talk
at WebDB, somebody asked about identity constraints in RELAX NG. When I
said that RELAX NG does not have multi-part keys, he said that such
identity constraints will not make the database community happy. Nobody in
the room disagreed.
Let's face it. The current spec is already very unsatisfactory for
database
guys. We will eventually need a different spec for identity constraints.
If
our mechanism is as simple as ID/IDREF/IDREFS of DTDs, we can say that it
is
only for compatibility with DTDs. If our key/keyRef is extremely
complicated,
we cannot say so.<
>>>>>>>>>>>>>
Murata,
We have to face reality here. We cannot please everyone. Equally we
cannot
swamp the spec' with detail and complexity for this or that fringe - or
non-core
use case / requirement. That defeats the purpose - of having something
simple
and consistent that meets the 80:20 rule. We should say what RELAX is
NOT as
well as what it IS.
If the database guys want all this key stuff - let them go off and develop
RELAX DB
extensions or something.
Frankly I think this database stuff is a red-herring. It's because people
are still
steeped in RDBMS / SQL think - and they have not moved beyond that yet.
The future is adaptive information filtering driven off semantics and the
inherent
XML structural context. The means to implement all this is separate and
distinct from the XML instances and structural layer and should not be
confused
with it.
The real answer for XML accessing within an XML DB is XPath / XQuery et al
and that clearly is the domain of other WG's.
My company has two DB/retrieval products for XML. Both work directly off
the
XML instances, and yet utilize external extracted indexing to actually
drive the
querying. Most every product has copied this approach. And actually the
MORE of IDREFs and cute tricks people use in their raw XML - the less
likely
it is to be maintainable / supported; because they imagine all kinds of
behaviours that are not there / sustainable OR would be more simply and
effectively achieved using an equivalent XPath joined query.
We have more than enough in the spec' already to support RELAX Parser level
referencial content. Anything else is IMHO out of scope.
Sometimes when you give presentations you have to be blunt! It's only
salesmen that say "Oh that? YES! It's in there!".
DW.
