Re: Issue 146 proposed resolution

There is another choice:
d) Get rid of actor and make all processing based on the acceptance
   of the contract specified by the QName of the block.  This removes
   this pseudo targeting that seems to be causing so many problems.
   Everything is a black box anyway - so the QName should be the
   determining factor.  We can get the semantics of actor="next"
   other ways...
:-)
-Dug


Noah_Mendelsohn@lotus.com@w3.org on 11/15/2001 12:14:20 PM

Sent by:  xml-dist-app-request@w3.org


To:   Mark Baker <distobj@acm.org>
cc:   henrikn@microsoft.com (Henrik Frystyk Nielsen), skw@hplb.hpl.hp.com,
      xml-dist-app@w3.org
Subject:  Re: Issue 146 proposed resolution



Mark Baker writes:

>> It's important that SOAP support a gateway model,
>> and in doing so, not attempt to prevent
>> "processing the message" from including
>> delegation to other processors.

Agreed.   The question in my mind is, exactly how do we model it. As I
understand Henrik, he is saying:  "there can be two nodes (and I mean
"node" in exactly the formal sense of chapter 2) in the path taken by a
given message both of which act in the role of the anonymous actor,
neither of which is considered an intermediary;  nonetheless, the first
one of these relays (again in the formal sense of "relay" per chapter 2)
the message to the second.

I still find this very problematic.  The formulations for this use case I
would prefer are your choice of:

a) From the point of view of the SOAP spec, there is a single endpoint
node, which happens to have distributed logic, about which the SOAP spec
says nothing.  You are always free to split your implementation in ways
that the spec doesn't talk about, as long as you can identify the points
in your system where conformant processing is done.

b) The first such node is the true endpoint, and from a chapter 2 point of
view the message path ends there.  If you choose to generate what SOAP
believes to be a separate but very similar looking message to the second
node, that's your business.  The second node is the endpoint for that
message.  The response, if any, to the second message can be used to
construct the resposne to the first.  This can either be done privately to
the implementations, or per the specification for some SOAP extension
feature.

c) (the one I like least, I think)  get rid of the distinction between
intermediaries and endpoints.  Just say that there are nodes that play the
role of the anonymous actor (and hence process the body) and those that
don't.

Going to some trouble to distinguish intermediaries from endpoints, and
then aving endpoints that act like intermediaries seems very strange to
me.

------------------------------------------------------------------------
Noah Mendelsohn                                    Voice: 1-617-693-4036
Lotus Development Corp.                            Fax: 1-617-693-8676
One Rogers Street
Cambridge, MA 02142
------------------------------------------------------------------------

Received on Thursday, 15 November 2001 12:41:52 UTC