Re: issue: type of modelReference attribute?

Joel, 

thanks for this clarification. I think I understand what you suggest. 
I still don't feel comfortable with it, though.

If we put a single modelReference for Printer on a WSDL interface
component, then we are saying that the interface represents a printer.
Is the model that describes Printer supposed to be the complete
description? I don't think we can mandate completeness. If we could
mandate completeness, we can say that multiple modelReferences must be
equivalent, just possibly in different modeling languages. But we don't
need multiple modelReferences, assuming that any modeling language can
assign properties to any URI.

However, without completeness, one modelReference can claim something
else than another, in orthogonal directions. Then the question is what
the combination would be. For example, we could put "ColorPrinter" and
"LaserPrinter" on one interface, one would say it can do color, one
would mean that it prints relatively fast and cheap. 

I believe that the natural operation here is conjunction (or
intersection) of the modelReferences, i.e. both references say true
things about the interface. This would also allow us to combine
different descriptions in languages with different expressivity - a
pointer to a class in product ontology to say it's a printer, a pointer
to process description that will tell us how to communicate with the
printer (the choreography). While they describe the same thing, they
don't say the same statements, therefore they aren't exchangeable, and
the client would choose whatever annotation is appropriate for the task
at hand.

So if we want to allow multiple model references (I haven't seen an
actual use case yet, except for my weak examples above), it should
probably behave like creating a new class, just what you wrote against.

Does it make sense?

Jacek

On Mon, 2006-04-24 at 10:03 -0400, Joel Farrell wrote:
> Hi Jacek,
> 
> I think the confusing part of my message was
> 
> > > We should make
> > > it clear that multiple URIs  do not represent some union or other
> > > combination of the ontological concepts that are referenced.
> 
> I was referring to how a modelReference attribute whose value includes more
> than one URI should be interpreted.  I think the multiple URI's should be
> considered references to be equivalent alternative descriptions.  The tool
> reading the WSDL file should pick one. Each of the URI's would point to a
> particular element in a model.  The list of URI's should not be interpreted
> or used as an aggregate description of the semantics of the WSDL component
> to which the annotation was attached.  So for example, if an input element
> denotes a Nobel Prize winner and a Turing Award winner, you would include a
> modelReferenc to a class called NobelTuringWinners.  You would not include
> a modelReference with two URI's, one pointing to a class called
> NobelWinners and a second pointing to a class called TuringWinners.  The
> modelReference URI list cannot be used to define a new class.
> 
> Regards,
> Joel

Received on Monday, 24 April 2006 19:17:10 UTC