- From: David Hull <dmh@tibco.com>
- Date: Thu, 14 Jul 2005 16:34:43 -0400
- To: public-ws-async-tf@w3.org
- Message-id: <42D6CC63.3060802@tibco.com>
I've never been completely comfortable with the SOAP request-response MEPs distributed state machine. In the "no SOAP MEPs" world, we can get rid of it, in the spirit of "the advertising party sets the contract." An endpoint advertising an in-out operation is advertising that it will accept messages, ponder them duly, and send responses as per the rules in section 3 (or something like them). A client is free to do what it wants with that information. Consider two scenarios: 1. The client sends out a request with a callback endpoint that it owns, spawning a listener to catch the response. The client is going through essentially the same state transitions as with a synchronous request, to essentially the same end. 2. A third party is forwarding this request onward to the actual server, maybe with some intermediate processing. It leaves the [reply endpoint] pointing at the original callback endpoint, fires off the amended request, and gets out of the way. The state transitions are split between the third party and the callback endpoint. The point is that the actual server generally can't tell these cases apart, nor should it. It does the same things in either case, and we don't need to capture what the client, third party and so forth do around this. A single non-distributed state machine should be enough.
Received on Thursday, 14 July 2005 20:34:48 UTC