- From: David Hull <dmh@tibco.com>
- Date: Tue, 31 Oct 2006 01:35:51 -0500
- To: "public-ws-addressing@w3.org" <public-ws-addressing@w3.org>
This is an attempt to pull some thoughts together after looking through the WS-RX spec [1], with particular attention to section 3.10, and reading Dug's example flow [2] and Paco and Anish's proposal for CR33 [3]. * Section 3.10 of WS-RX uses the phrase "transport-specific back-channel" repeatedly but does not define it (the definition is clearly left to the transport). * The MakeConnection message seems to be essentially a poll for a pending message (it appears that no more than one can be returned at a time). If so, perhaps the name should reflect this better. My first impression of "MakeConnection" is that it must be similar to CreateSequence. * The MakeConnection message is described as one-way. A comment in the WSDL particularly says "In SOAP terms the message [coming back] is not a response, so there is no WSDL output message." I'm not sure that the implication follows, and I'm of two minds about the assertion that the message not being a response. On the one hand, if I POST a SOAP message and a SOAP message comes back in the HTTP response, the interaction is indistinguishable on the wire from an instance of the SOAP request-response MEP. On the other hand, I'm all for allowing such an exchange also to be described as two one-way messages (after all, the still-ante-natal one-way is the building block from which all other interactions may be built :-) * If the back-channel is transport specific, then it's implicit in the transport binding. Any marker for it in this sense is thus redundant. * Since WS-RX appears quite SOAPy, I would be more comfortable seeing much of this recast, in line with Anish's suggestion and the WSA description of anonymous, in terms of the SOAP request-optional-response MEP. It looks like RX leaves open the possibility of other forms of back-channel, but this would cover the common case of HTTP and would pick up anything else that supports R-O-R for free. * Paco & Anish's proposal also uses back-channel without defining it. * The SOAP 1.2 Adjuncts [4], which include the SOAP/HTTP binding, do not mention "back-channel" (though they do mention Backus-Naur form). * RFC 2616 [5] neither uses nor defines "back-channel" (though it does mention back links). * From a purely WSA point of view, the most important thing to mark is whether an endpoint supports non-anonymous replies. If it prohibits anonymous, that will generally be because the request is coming in on a one-way transport, in which case the "I don't' support anonymous" marker is, strictly speaking, redundant. * I believe I would be comfortable with a specification of a "back-channel available" marker that defined "back-channel" clearly in terms of SOAP MEPs. I can't live with the current "everyone knows" definition. We're writing specs here. That means we specify things. * OTOH, I'm 80+% sure that CR33 can also be closed with no action. [1] http://www.oasis-open.org/committees/download.php/20910/wsrm-1.1-spec-wd-16.pdf (you may need to be an OASIS member) [2] http://lists.w3.org/Archives/Public/public-ws-addressing/2006Oct/0036.html [3] http://lists.w3.org/Archives/Public/public-ws-addressing/2006Oct/0094.html [4] http://www.w3.org/TR/2003/REC-soap12-part2-20030624/ [5] http://www.ietf.org/rfc/rfc2616.txt
Received on Tuesday, 31 October 2006 06:36:12 UTC