Re: RELAX NG and W3C XML Schema

[John Cowan <jcowan@xxxxxxxxxxxxxxxxx>]

Eric Sedlar scripsit:

> My point was to illustrate the primacy of market acceptance vs. the
> particular language in question.
> DTDs are harder to read than either Schema or RNG, imho.

But there are a lot more people who know DTDs backwards and forwards
than for any other schema language.  

> Writing schemas by hand is very feasible too and what we do in my developmen
> t group.  However, my point is that most users use tools when practicable.

I think rather that most people use tools when *necessary*.

> The "why" of the market acceptance is irrelevant.  My point is that market
> acceptance and general understanding is much more important than the
> technical merits of the language, assuming it meets the minimal
> requirements.  RelaxNG & Schema do, & DTD does not.

In the long run you are right.  But this is still the short run, despite
casual talk about "Internet time".  These schema languages are barely a
year old: babbling infants.

> No, my comparison is more around the mindset required to write effective
> code (recursive definitions vs. a more traditional structure metaphor).

I don't see it.  If you use only the DTD-compatible subset of XS (plus
the datatypes which are common to XS and RNG), then maybe.  Otherwise,
there is just as much recursion in XS as in RNG, except that now and
again you bump up against an arbitrary rule.

I just finished developing an RNG schema for XLink, probably the most
difficult one I've ever done, because XLink allows arbitrary element
names as well as arbitrary children and arbitrary foreign attributes.
It took me about an hour to get it past the validator's front end.
I'm looking forward to comparing it with Henry Thompson's XS version.

> Design by committee tends to cause the kitchen-sink of features feel to the
> specification, which plague both C++ & Schema.

But not always.

> I think that "feelings" are the basis behind a lot of the comments on this
> list feeling that Schema "sucks" or is "grotesque".

You will note that I have refrained from such denigrations.

> I never said that RNG is not implemented, just not WIDELY implemented.

This is to some extent a chicken-and-egg problem.  Still, how widely
implemented is Perl?  There Is Only One.

> > Anyone was free to participate who wished to, unlike the situation with
> > XML Schema.
> >
> 
> Yet most people wanted to...  Why?

I'm not sure if you're saying that most people wanted to participate in
XML Schema work but couldn't, or contrariwise that most people didn't want
to participate in RNG work.  In the latter case, I can only point to the
generally lower profile of OASIS or even ISO relative to the W3C, which is
perceived as owning the space except in areas (like SAX) that they have
completely stayed out of.

> Lack of vendor support.  I can tell you from Oracle's perspective, XML
> Schema is much easier to support on top of our existing infrastructure than
> RNG would be, because it is more SQL-like, and we don't have to support data
> models that don't map well to our existing storage technology.

I would be interested in the details of this.  Note that it can't be a
matter of datatypes, because those are shared between XS and RNG.

> Leveraging
> existing technology & installed base is always nice.  There's a reason why
> all the database vendors are supporting Schema over RNG.

"Sunk costs are sunk costs."  --David Friedman

> > If you want ASN.1, you know where to find it.
> >
> 
> "Whether you mean it to be or not, this is a smear tactic"

:-)

> The #1 complaint I hear about XML is the size of the textual representation.
> Just reporting what I hear.

If you want hard-coded binary formats, you know where to find them too.
X.25 types used to complain bitterly about the tremendous waste of bits
associated with giving each and every IP packet its own header.
Where are they now?

> > 1) I don't think you can get agreement that market acceptance is
> everything;
> > 2) The market acceptance of XML Schema is being driven from the top down;
> > 3) There is no real market acceptance of XML Schema.
> >
> 
> Support for this?  I think most people want something better than DTD and
> RNG & Schema both meet those requirements, and they want that thing to be
> supported everywhere.

Fair enough.  But what has support in databases got to do with the
specification of wire protocols as such?

> Just responding to James's point that specifying default values doesn't
> belong in a XML schema language.

No more they do.  It's a kitchen-sink feature, retained in a pointless
effort to subsume DTDs completely.

> Implementations clearly need to maintain their own reference copy of the
> schema rather than pick one up off the Internet each time you get an
> instance to validate.  We use the schemaLocation as a unique key to identify
> the schema rather than as a location to dereference, like XML namespaces.
> More implementation experience is needed here, though, but the schema spec
> is flexible on this, but still providing for interop.

RNG is even more flexible: make the association any way you like.  For
that matter, you could write a validator that looks at an rngSchemaName
attribute on the root element and decides to use one of a set of canned
schemas: it would be easy to layer such a thing.

-- 
John Cowan <jcowan@xxxxxxxxxxxxxxxxx>     http://www.reutershealth.com
I amar prestar aen, han mathon ne nen,    http://www.ccil.org/~cowan
han mathon ne chae, a han noston ne 'wilith.  --Galadriel, _LOTR:FOTR_