- From: Jonathan Marsh <jmarsh@microsoft.com>
- Date: Wed, 24 Jul 2002 15:31:01 -0700
- To: <www-ws-desc@w3.org>
[Forwarding to the public list as authorized in http://lists.w3.org/Archives/Public/www-ws-desc/2002Jul/0113.html - JM.] SOAP TF telcon, July 15th 2002 Present: Philippe Le Hegaret (W3C) Glen Daniels (Macromedia) Sanjiva Weerawarana (IBM) Jean-Jacques Moreau (Canon) Roberto Chinnici (Sun Microsystems) Regrets: Dietmar Gaertner (Software AG) Joyce Yang (Oracle) Barbara Zengler (Daimler Chrysler) Michael Mahan (Nokia) Scribe: Roberto Subject: MEPs in SOAP 1.2 Roberto: MEPs seem very general, they can represent choreographies, processes, right? Glen: Right. We should not try to boil the ocean though. We have to deal at least with the case of request-response on one-way protocols. If we do the latter, we may solve something a bit more general (leaving the rest to future work). Sanjiva: OK. Glen: Issue with request-response over one-way: we need a response address. Sanjiva: But SOAP doesn't have a one-way MEP even! Glen: Easy to define. The "experimental" SMTP binding defines a 1-way binding. Jean-Jacques: Even that one is request-response, and they use a correlation property. Sanjiva: Is a MEP inline in a SOAP message or it understood by sender and receiver? Glen: The latter. Sanjiva: Shouldn't there be a header (or something) specifying the MEP in use? Jean-Jacques: The Message Exchange Context lives on every node and contains the name of the MEP used. Philippe: No example of this kind in the primer. Glen: No such header today. Extensibility means that many people will define their own extension and the best one will win in the marketplace. Sanjiva: Let's take the SOAP binding for request-response. It worked in 1.1 because it was only on HTTP. Jean-Jacques: We want to support other protocols, say UDP or TCP. Sanj: Or even two different HTTP bindings. Glen: Issue: how does the backchannel work? We can: (1) stick MEP URI in WSDL; (2) specify where the backchannel URI goes. Jean-Jacques: We'd like a client to be able to find out which MEPs are supported by which binding by looking at a WSDL. Roberto: If you need to support multiple MEPs for each operation and you specify them at the binding level, you risk a combinatorial explosion. Glen: You shouldn't need that. MEPs should really be specified at the abstract level. Sanjiva: No, that ties it to SOAP. Roberto: Right. Glen: Features are in SOAP 1.2 spec because that's the first group that needed them, but they don't belong there: they should go into a Web Services Architecture spec. Sanjiva: If you don't have SOAP in the binding, they don't make sense. Glen: Nothing really ties them to SOAP. Sanj: We'd need to create our own request-response and one-way MEPs. Sanj: Any feedback for the XMLP WG? For instance, they don't define a header for the MEP. Jean-Jacques: Define a correspondence between WSDL MEPs and SOAP ones. Do we need a more general concept of MEPs? How do we tell a client that they need to use a particular MEP? Sanjiva: In response-only, should be able to specify how input arguments are encoded into the URI. Glen: Using regular expressions. Roberto: It's a template, really. Glen: I'd like to see a mapping of SOAP features to WSDL. Sanjiva: Isn't it a different problem? Glen: no, for instance it's needed for the webmethod feature (which is orthogonal to the MEP). Sanj: So a feature is just a URI and its explanation is plain English. Glen: Yes. (Sanjiva had to take another call for a few minutes.) Roberto: Is there anything in the request-response MEP that prevents it from being used for, say, asynchronous RPC? Glen: No. Conclusions (to be reported to the WG): (1) We need to be able to express MEPs in the SOAP binding. (2) The ability to describe features and MEPs at the abstract level would be a good thing, but right now we don't know how much effort it would take to do so. (3) Feedback to XMLP WG: lack of something (such as a header) to express the MEP in use. (Glen: perhaps we should say that, for each binding, there must be a way to identify unambiguously the MEP in use; of course, for bindings that support just one MEP that's a no-op). (4) Currently, there is no well-defined one-way MEP.
Received on Wednesday, 24 July 2002 18:31:31 UTC