RE: Action Item 2004-07-01 Solution to 168/R114

Hi folks:

Paul writes:
> Ümit wrote:
> >> Jonathan Marsh wrote:
> >> Also, how does this address Paul's concern at 
> >> http://lists.w3.org/Archives/Public/www-ws-desc/2004Jul/0043.html?
> >
> > I believe Paul's concern is addressed by requiring the mandatory 
> > extension to be declared in WSDL. The addressing mechanism engaged 
> > will provide the operation name to be present in the 
> message exchange 
> > regardless whether it is async binding or not.
> >
> > Paul? 
> 
> my concern about asynchronous routing was muddled in that for 
> a moment i saw dispatching a message received asynchronously 
> differently to those framed in a synchronous conversation.
> 
> i've now come to my senses and realise they are identical and 
> whatever proposal is used to resolve issue 168 should also 
> resolve this 'issue'.
> 
> i'm not, yet, convinced that WSDL (the language) should 
> require a mandatory dispatching mechanism, however limiting 
> the scope of 'mandatory' to within the RPCstyle does makes 
> sense to me.

Here's the thing as far as I'm concerned.  We want to make WSDL 2.0 as interoperable and useful a language as possible, to avoid the twelve different kinds of confusion we had (and still have in some cases) interpreting WSDL 1.1.  As such, WSDLs which rely on "secret sauce" which is completely out-of-band with respect to the description seem like they should be the exception rather than the rule.  Such WSDLs, as a reminder, *cannot* be implemented by people who want to build service implementations (and perhaps, but not necessarily, clients) in any kind of interoperable way.

I certainly don't want to prevent people from building WSDLs which rely on magic in order to do dispatch/operation differentiation.  But personally, I would rather see the default for all WSDL documents be that you MUST somehow indicate how this important task is to be done (for instance with a concrete feature/module/binding which satisfies an abstract OperationName feature).  Unique GEDs work, RPC style (a subset of unique GEDs) works, a module like that suggested by Umit and myself works, etc.  I'd even be fine with defining a "magic operation dispatch" feature which explicitly states that some out-of-WSDL magic is happening, but that you would have to explicitly mark in the WSDL so that people can choose to process it or not based on whether they are willing to accept the fact that they won't necessarily know what's going on.

In other words - operation dispatch is important, so WSDLs (without unique GEDs) should always have to indicate how it's happening, even if that indication explicitly says "that's my secret".  I'd rather that than the default being "that's my secret".

--Glen

Received on Thursday, 8 July 2004 11:20:00 UTC