- From: Liu, Kevin <kevin.liu@sap.com>
- Date: Wed, 16 Feb 2005 17:54:53 +0100
- To: "David Hull" <dmh@tibco.com>, <public-ws-async-tf@w3.org>
>From: public-ws-async-tf-request@w3.org ... >>* Can we achieve this case now with the current specs? With how much >>"squinting"? >> >> >As I understand it, WSDL cannot properly advertise such behavior. The >HTTP binding is hard-wired to send replies and the faults on the >back-channel. > <kevin>It depends how the wsdl interface/portType is defined. - If the interface is defined as request-response, then wsdl binding can not support such case as I explained in the expenseReport use case. - If we model this use case using multiple interfaces (or operations if same binding type is in use), then it's do-able. One interface can be defined for the sync request-response, one interface for the in-only one way, another interface in the client side for a async call back. Client makes its choice by choosing the right interface to use, not by changing the binding on the fly. >>* What is the minimal change that would be necessary to what >spec(s) in >>order to achieve this case? >> >> >If WSDL bindings supported a "dynamic" indicating that a particular >endpoint or endpoints in a MEP are dynamically specified, e.g., via >WS-A. In practice, this would mean the reply and fault addresses for >request/reply and robust out-only, and possibly analogously for >solicit/response (or whatever we call it these days) and >robust in-only. > >>* What would be the "ideal" solution if we could change >anything to get >>this case covered? >> >> >I think the above would work fine. > >>Could you take a swipe at integrating those? >> >>--Glen >> >> >> >>>-----Original Message----- >>>From: public-ws-async-tf-request@w3.org >>>[mailto:public-ws-async-tf-request@w3.org] On Behalf Of David Hull >>>Sent: Friday, February 11, 2005 4:06 PM >>>To: public-ws-async-tf@w3.org >>>Subject: Use case 6 >>> >>>Actors: >>> >>> >>>* A server implementing a request/reply operation >>>* A client invoking that operation >>>* Optionally, a third party receiving the reply. >>> >>>Subcase 1: >>> >>> >>> At the application level, the client simply sees a >>>request/reply operation. The client-side infrastructure >>>would like to be able to decide, as dynamically as possible, >>>whether to invoke the operation synchronously or >>>asynchronously. In the first case, the anonymous reply-to >>>will be fine (assuming the usual SOAP/HTTP binding). In the >>>second case, the reply-to will have to be a callback address. >>> >>> >>>Subcase 2: >>> >>> >>> The party sending the request to the server is not the >>>same as the party receiving the reply. For example, the >>>actual client sends a request to some broker, who forwards it >>>to the proper server. There is no reason for the server not >>>to send the reply directly to the actual client (as opposed >>>to forwarding it through the broker). In other words, the >>>broker's value add is finding the server, not forwarding per >>>se. Note that if the client supplied an anonymous reply-to >>>(as with current proxies), the broker will have to forward >>>the reply anyway.
Received on Wednesday, 16 February 2005 16:55:40 UTC