- From: Amelia A. Lewis <alewis@tibco.com>
- Date: Fri, 30 May 2003 12:57:06 -0400
- To: "Jonathan Marsh" <jmarsh@microsoft.com>
- Cc: www-ws-desc@w3.org
On Fri, 30 May 2003 09:12:46 -0700 "Jonathan Marsh" <jmarsh@microsoft.com> wrote: > One problem we tried to address at the FTF was distinguishing in the > WSDL between three cases not previously distinguishable: > > 1) When a set of endpoints represent "alternative" bindings for the same > interface acting on the same "thing" (e.g. SOAP and HTTP bindings as two > ways to access the interface). Perfectly feasible previously, and strengthened by the ability to do interface inheritance, without restricting services to a single interface. > 2) When a set of endpoints represent different interfaces to the same > "thing" (e.g. normal interface and a management interface). Perfectly feasible previously, and unaffected by interface inheritance, without restricting services to a single interface. > 3) When endpoints are related in some other way (e.g. your use case, I > think). > > The terminology we developed is that a "service" is an interface on a > thing, that can be accessed through various mechanisms. Thus the term > "service" and the "<service>" element have been clarified to cover case > (1). This is of course extremely useful for tooling. Why? This has been asserted repeatedly, but repeated assertion is *not* proof. If a service contains two endpoints, which are associated with two different bindings, then comparison of the interfaces associated with the bindings yields equal/not equal. If the stipulation is made that a service represents a single service (in the new, more wonderfully confusing terminology, a single "resource"), then identical interfaces are necessarily simply different bindings for the same service. The only stipulation previously lacking was that a service element is to represent a single service. Equally, if a service contains two endpoints, which are associated with two different bindings, and comparison of the interfaces associated with those bindings yields not equal, then they offer different forms of address to a single service, without introducing the term "resource" or forcing a single service to be represented in multiple service blocks. Again, all that is required is the stipulation that a single service element represents a single service. What is lacking in this model is the ability to equate two separate service elements that, in fact, represent the same service, either with the same interface or with different ones. By introducing confusion in the name of simplicity, this use case has dropped off the map. Or, as I stated when this was originally proposed, the proposal not only fails to add value, it removes it. It confuses the meaning of the term "service", it places a completely unnatural restriction on the service element, it introduces a new term "resource" with ill-defined meaning, and it fails to establish a comprehensible mechanism for expressing the relationship between related services. Had you gathered, by this point, that I find this change to be poorly motivated and worse implemented? If not, please be aware that that is my opinion. Amy! -- Amelia A. Lewis Architect, TIBCO/Extensibility, Inc. alewis@tibco.com
Received on Friday, 30 May 2003 12:56:36 UTC