[Bug 5293] Subsumption

http://www.w3.org/Bugs/Public/show_bug.cgi?id=5293





------- Comment #7 from noah_mendelsohn@us.ibm.com  2007-12-07 01:16 -------
> B: ab?|ba?
> 
> D: a?&b? [count(*)>0]
> 
> where the square brackets indicate an assertion.
> 
> These two types are equivalent (both allow the
> instances a, b, ab, or ba), and as far as I can
> see our current rules require processors to treat
> D as a valid restriction of B. But I don't think
> this can be done without some sophisticated
> theorem proving


Whoa.  Do assertions count in subsumption or restriction checking?  I thought
that after two years of trying to make a generalized restriction requires
subsumption rule, we dropped back to requiring only subsumption of the content
models (plus some other side conditions).  A quick look at the constraints
seems to bear out that they deal only with subsumption of content models, but I
could easily have missed something.

If I'm reading the above example correctly, it presuupposes that we are
checking the net content allowed by the combination of content models and
assertions, and I didn't think that was so.  Am I misremembering our intent?

Also: in some of the earlier comments, it was suggested that we might consider
an intersection-based approach.  Am I correct that proponents of that approach
do want to retain our requirement that types be intensionally as well as
extensionally connected for us to consider one as a derivation of another?  In
other words, whatever we do about content checking (and I'm not necessarily in
favor of the intersection approach but am willing to hear more), we would in
any case continue to require that types be in a parent/child relationship as
well?  I assume that was implicit in the earlier comments, but perhaps not.  I
certainly would not be OK with dropping that.

Thank you.

Noah

Received on Friday, 7 December 2007 01:16:30 UTC