Lists Home |
Date Index |
Simon St.Laurent <firstname.lastname@example.org> responds:
> >Hmm, I might even grant integrity constraint conformation (99.99% or
> >the time). What I wouldn't necessarily expect is
> normalizations that
> >conforms to what experts in the current relational world
> might expect:
> >I'm starting to believe that data normalization and metadata
> >normalization are orthogonal to each other.
> I think I'm with Peter on this one as far as normalization.
Wow, I've been buried so deep in this problem for so long I was beginning to
wonder if my experience had anything to do with the rest of the world....
> At the same time, though, I think there's a huge difference
> between the expectations of relational databases - which
> really demand a schema upfront before you're allowed to work
> with data - and XML, which has no such requirements. No
> rules, no violation - no harm, no foul.
That is currently true, but in my other response to this thread I sort of
point out that this may be less and less true going forward. If you want to
have any half way decent treatment of your XML (say even within your
application) you may need a schema just to tell a parser how to optimally
parse your XML. This schema may come about after the fact, but by the time
you get to doing data exchange it should be hanging around. Certainly, it
seems an unreasonable expectation that one can have much in the way of
automated (or efficient manual) generalized mappings without a schema (or
equivalent metadata) on both sides of the fence.
I just don't expect you're ever going to get good interchange between
"media" without doing some real work. I think that means well defined and
robust schema. I can't see that something like regex pattern matching on
tags and/or column names will ever work in general (which is more of a
response to Jeff's post, but I'm too lazy to reply twice) even ignoring I18N
types of issues.
> One of these days I'd like to figure out if the math
> underlying RELAX NG and the math underlying relations can be
> made compatible. That seems like a plausible path forward
> toward an easily processable and vaguely cross-media world,
> but I still don't think it'll cover all the differences.
This would certainly seem like a worthwhile pursuit, but it may perhaps only
be of academic interest; even if you do arrive at a common foundation, given
the differences in implementation of supposedly standard DDL among
relational vendors (as just one example of the scope of the problem) you are
(as you note) still a very long way from having anything that's going to
actually make cross platform, cross media metadata easy. I would pin a
little more expectation on the top down modeling approach. That's probably
just my bias from having done a lot of top down design, but I've done it in
both the OO and Relational worlds enough to believe it not unreasonable...