RE: is the uniqueness constraint on top level components sufficient?

I totally understand where you were going with the problem.  To me, it falls
in the camp of "don't do stupid things and nobody gets hurt".  I just can't
see putting untestable requirements into a spec like wsdl.

If you as an ns owner want to make sure that a:foo is unique, then you
better have some process to make sure.  Now, where I could see this being a
little more possible to do what you want is something like a web best
practice that said "NS owners that want to ensure unique QNames should
provide some mechanism or policy to ensure such uniqueness.  An example of
QNames that might need uniqueness is WSDL interface names.  An example of
such a mechanism is a process within the organization involving the ns owner
approving publication of wsdl definitions.".

I think that part of this problem has to do with the use of Qnames rather
than uris, but that is indeed a different issue.

Cheers,
Dave



> -----Original Message-----
> From: Sanjiva Weerawarana [mailto:sanjiva@watson.ibm.com]
> Sent: Friday, September 19, 2003 4:32 AM
> To: David Orchard; 'Roberto Chinnici'
> Cc: 'Martin Gudgin'; www-ws-desc@w3.org
> Subject: Re: is the uniqueness constraint on top level components
> sufficient?
>
>
> Looks like I've been out-voted, but let me whether I can explain
> why I think we need to put some more clarification. I am *not*
> looking for a way to check or test whether QNames are indeed
> unique, but it is fully within our purview to say that they
> must be. If we say that then someone receiving a WSDL component
> QName and resolving it to a component with that name has some
> reason to believe they've got the right one: the WSDL spec says
> it must be so. If we don't require it, there is no such confidence.
>
> Let's say I'm writing a BPEL thing and want to refer to a WSDL
> interface (portType today). BPEL will refer to it by QName, a:foo.
> Now "somehow" the BPEL process will find the portType named a:foo
> from a WSDL. How does it know it indeed got the "right" one named
> a:foo? If the WSDL spec says that there must only be one portType
> named a:foo then that provides some confidence that it did indeed
> get the right one. Now, the owner of the a: namespace could've
> of course screwed up and allowed two a:foo's to exist but that's
> that guy's fault. At least the WSDL spec says that's not legal.
>
> The wording we currently have doesn't preclude one from naming
> two different interfaces a:foo. What I'd like to do is tighten
> that wording so that the spec says that QNames of top level WSDL
> components MUST be unique period. I am happy to even add a note
> saying "we realize this is not testable and that its totally
> up to the owner of the namespace to enforce it."
>
> Anyone crossing over to my camp?
>
> Sanjiva.
>
> ----- Original Message -----
> From: "David Orchard" <dorchard@bea.com>
> To: "'Roberto Chinnici'" <Roberto.Chinnici@Sun.COM>;
> "'Sanjiva Weerawarana'"
> <sanjiva@watson.ibm.com>
> Cc: "'Martin Gudgin'" <mgudgin@microsoft.com>; <www-ws-desc@w3.org>
> Sent: Saturday, September 20, 2003 2:28 AM
> Subject: RE: is the uniqueness constraint on top level components
> sufficient?
>
>
> >
> > Indeed, I agree with gudge and roberto.  Seems like it's the
> responsibility
> > of the ns owner to figure out the names, vocabularies and
> languages within
> > it's namespaces.  Using URIs for ns names at least makes it
> very clear
> what
> > the domain authority is.  Now if we had better RDDL
> support, it might be
> > easier to check what sanjiva wants.  And I think that's the
> route to go.
> >
> > For now, I think the constraints on names are reasonable.
> >
> > Cheers,
> > Dave
> >
> > > -----Original Message-----
> > > From: www-ws-desc-request@w3.org
[mailto:www-ws-desc-request@w3.org]On
> > Behalf Of Roberto Chinnici
> > Sent: Friday, September 19, 2003 11:33 AM
> > To: Sanjiva Weerawarana
> > Cc: Martin Gudgin; www-ws-desc@w3.org
> > Subject: Re: is the uniqueness constraint on top level components
> > sufficient?
> >
> >
> >
> > Sanjiva Weerawarana wrote:
> > > Hi Gudge,
> > >
> > >
> > >>I think you will find that as far as our spec is concerned there is
> > >>always EXACTLY ONE definitions container even in cases where the
> > >>contents of that container came from multiple locations.
> > >
> > >
> > > That's for included or imported stuff right? Is there anything
> > > in the spec which says that *all* stuff for the target namespace
> > > are part of EXACTLY ONE definitions container? If so then we're
> > > in business. If not its certainly possible for two documents
> > > to point to the same namespace yet not be aware of each other:
> > >
> > > <definitions targetNamespace="http://www.ibm.com/">
> > >    ... stuff for one service ...
> > > </definitions>
> > >
> > > <definitions targetNamespace="http://www.ibm.com/">
> > >    ... stuff for another service ...
> > > </definitions>
> >
> > I don't think that adding an untestable requirement of this kind to
> > the spec does any good. If somebody wants to have two WSDL documents
> > for the same target namespace, so be it. The burden to be
> > extra-careful
> > in their definitions falls on them.
> >
> > Roberto
> >
> > --
> > Roberto Chinnici
> > Java Web Services
> > Sun Microsystems, Inc.
> > roberto.chinnici@sun.com
> >
> >

Received on Friday, 19 September 2003 18:27:42 UTC