W3C home > Mailing lists > Public > xmlschema-dev@w3.org > September 2006

Re: the UPA-constraint and danish word division

From: C.M.Sperberg-McQueen <cmsmcq@acm.org>
Date: Wed, 20 Sep 2006 09:25:07 -0600
Message-Id: <ED996A19-202D-47CB-A5A3-1D90278C8AB3@acm.org>
Cc: "C. M. Sperberg-McQueen" <cmsmcq@acm.org>, "Marie Bilde Rasmussen" <mariebilderas@gmail.com>, "Xan Gregg" <xan.gregg@jmp.com>, xmlschema-dev@w3.org
To: noah_mendelsohn@us.ibm.com

On 20 Sep 2006, at 06:21 , noah_mendelsohn@us.ibm.com wrote:

> So I think the right question isn't:  would there be useful  
> grammars you
> can't write today that you could write if UPA were eliminated.  The  
> answer
> is clearly yes.  I think it's also clearly useful at times to build
> schema-driven or certainly schema-aware editing systems.  The  
> question is
> whether the advantages of expressing your application constraints  
> exactly
> in the schema language outweigh the advantages that some (though  
> not all)
> users of schema and implementors of validators derive from the strict
> particle to instance matching that's given by UPA.

That would be a more persuasive argument if, at some point over
the past twenty years, anyone had actually identified any advantages.
So far, I haven't  heard anything that I believe actually counts as an
advantage, still less an advantage strong enough to outweigh the
disadvantage of not being able to express regular languages in
a content model.

Your quotation from the spec seems to me to apply with even more
force to those people who wish to preserve UPA for the sake of
applications they believe that someone, someday, might write, that
could surely, surely derive benefit from UPA.  The schema language
doesn't need to be all things to all people.  But it could do with  
a constraint which has no demonstrated benefit but plenty of
demonstrated problems:  the absence of closure properties, the
well known difficulty in maintaining or revising schemas, difficulty in
adding wildcards where one wants them, and the divergence from well
known automata theory, so that instead of looking up algorithms and
theorems and properties in textbooks, implementers get to play
computer scientist and derive them from scratch for a language which
is similar to, but not quite, the same as, regular expressions.

The fact that other schema languages manage perfectly well without
requiring deterministic content models seems to me to show pretty
clearly that the determinism rules are not actually necessary in  
any more than they are in theory.

Received on Wednesday, 20 September 2006 15:27:45 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 14:56:10 UTC