RE: WSDL Import/Include Locations

I apologize if I have missed or misunderstood text but my understanding is
that only section 4.1 on includes defines how to behave in the face of
circular includes. Section 4.2 on imports does not provide similar language
and therefore it is unclear what the appropriate behavior of a WSDL engine
should be in the case that two imports resolve to identical content.

As for registries, it's funny you bring it up because I recently talked to a
number of people, both product managers and sales managers, about exactly
this topic. I also had occasion to talk to a number of different customers
about it, especially at two recent customer forums.

What I found was that while customers demand that a web service platform
support a registry like UDDI and while many customers do buy UDDI style
software, they almost never use it. From what I found in my un-scientific
and purely anecdotal survey is that some companies do try to use UDDI or
other registry technologies internally but quickly give up because the
registries tend to more trouble then they are worth. I also found no one who
was seriously trying to use a registry technology between companies.

I would contrast this to the use of WSDLs where, at least in rpc/encoded
cases, WSDLs were widely exchanged both within and between companies.

So, for whatever it's worth, it does appear that WSDLs, even formally
produced ones, tend to be splattered across a number of different servers.

Therefore it would be useful to be able to list a number of alternate
locations. This is essentially mirroring which we routinely see used for a
wide variety of purposes on the Internet.

	Thanks,

		Yaron



> -----Original Message-----
> From: Amelia A Lewis [mailto:alewis@tibco.com]
> Sent: Friday, January 23, 2004 8:55 AM
> To: ygoland@bea.com
> Cc: www-ws-desc@w3.org
> Subject: Re: WSDL Import/Include Locations
>
>
> The specification currently discusses the problem of circular imports
> and includes, which may address these questions (basically, it says
> that a processor *must* recognize when an import or include, even if
> nested, has been encountered before, and should not go into
> an infinite
> importation/inclusion loop).
>
> Import and include are defined to have roughly the same
> semantic as in
> W3C XML Schema, with restrictions (no chameleons).  This has the
> advantage of providing readers and implementors of the specification
> with an external model, one which is hopefully well known.  Major
> changes to those semantics ought to have benefits at least
> equal to the
> expected familiarity, since they would presumably trade off
> the benefit
> of familiarity for the benefits introduced.
>
> Given the existence of cataloging/registry/resolution APIs
> and software
> modules, I'm not certain that adding additional optional locations
> meets that standard of equal benefit.  However, I can be
> convinced, if
> it truly is the case that no one (else, that is) uses moderately
> sophisticated resolution/cataloging technologies.
>
> Amy!
> On Jan 22, 2004, at 6:09 PM, Yaron Goland wrote:
>
> >
> > Both WSDL import and include only allow for a single location to be
> > specified. Given the unreliable nature of the Internet
> would it not be
> > appropriate to allow for more than one location to be specified?
> >
> > Given the permissive semantics of include it would be tempting to
> > specify
> > multiple includes, all pointing to the same file but at different
> > locations
> > as a way to make the WSDL definition more robust in the
> face of network
> > failures. However this would needlessly waste system
> resources making
> > unnecessary file requests. If the WSDL processor knows that
> a set of
> > URIs
> > are equivalent then it need only make requests until it finds a URI
> > that
> > works.
> >
> > In the case of import the specification doesn't actually
> define what
> > happens
> > if someone writes two imports for an identical namespace.
> Although some
> > limited definition redundancy is supported by the spec the support
> > would not
> > appear to be robust enough to support importing the same WSDL
> > definition
> > twice. Therefore putting in two imports as a way to provide
> redundant
> > locations would appear illegal.
> >
> > But this begs the question - Is it illegal to specify two
> imports for
> > the
> > same namespace? If so, shouldn't this be explicitly stated
> in the spec?
> >
> > What is the required behavior if it is impossible to successfully
> > import/include an identified document? If this an
> unrecoverable error
> > that
> > requires that the WSDL be rejected for processing? If so, then
> > shouldn't the
> > spec explicitly state this?
> >
> > 	Thanks,
> >
> > 		Yaron
> >
> >
> >
>
>

Received on Friday, 23 January 2004 19:32:14 UTC