W3C home > Mailing lists > Public > xml-dist-app@w3.org > August 2004

RE: request for clarification of SOAP/HTTP binding in SOAP 1.2

From: <noah_mendelsohn@us.ibm.com>
Date: Wed, 4 Aug 2004 14:54:40 -0400
To: "David Orchard" <dorchard@bea.com>
Cc: "Anish Karmarkar" <Anish.Karmarkar@oracle.com>, "Sanjiva Weerawarana" <sanjiva@watson.ibm.com>, xml-dist-app@w3.org
Message-ID: <OF4283F938.1FE7EC11-ON85256EE6.006D8787@lotus.com>

Dave Orchard writes:

> I'd like to follow a bit further on this.

> It is my understanding that the SOAP Request-Response
> MEP does not constrain what goes in the request and the
> response, 

Right, except that each request and each response is pretty much 
restricted to a single XML Infoset (plus some stuff around the edges like 
a destination address for the request.)

> merely that when using request-response and the SOAP
> HTTP binding then the request part of the MEP goes in
> the HTTP request and the response part of the MEP goes
> in the HTTP response.

Right.  The MEP itself does not embody this constraint, but the SOAP HTTP 
binding provided with SOAP 1.2 maps the MEP in the way you describe. 
Another binding could do it differently.

> Specifically, it does not say that an application
> level request/response (ie wsdl operation) is
> constrained to use a single req-resp MEP.  It is a
> perfectly valid use of the SOAP req-resp MEP for an
> application level req/resp to use two(2) req-resp MEPs
> when bound to SOAP.

One view would be:  the MEP doesn't talk about applications, so it's 
silent on this question.

The view I'd prefer would be:  the reason the MEP exists is implicitly 
(not explicitly) to promote sharing of middleware and application models 
for a very common pattern of communication, I.e. request/response.  If you 
use the MEP in the obvious way, then the chances go way up that a wide 
variety of software will help you do your job.  If you bind what the app 
considers to be a single request response to a pair as you propose, then 
I'd suspect that a lot of WSDL-aware application development tools, 
perhaps debugging tools, and maybe others, will do a less good job of 
helping you build and test your application.  Insofar as you're promoting 
a second on-the-wire representation for that single request/response, 
you've got another convention to convince the industry to support in a 
standard manner.  Maybe that's a good tradeoff in some cases, but it seems 
like something to be avoided where possible.  As you say, I don't think it 
specifically contradicts anything in the Recommendations.

> Cheers,
> Dave

Cheers to you!

Noah Mendelsohn 
IBM Corporation
One Rogers Street
Cambridge, MA 02142
Received on Wednesday, 4 August 2004 16:07:15 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 22:01:26 UTC