W3C home > Mailing lists > Public > public-ws-chor@w3.org > November 2004

RE: Session keys and CDL

From: Furniss, Peter <Peter.Furniss@choreology.com>
Date: Thu, 25 Nov 2004 09:58:37 -0000
Message-ID: <221369570DEDF346AE42821041345E897D3CB3@imap.choreology.com>
To: "Steve Ross-Talbot" <steve@enigmatec.net>, "WS-Choreography List" <public-ws-chor@w3.org>
I'm not sure if the pattern of deliberately quoting the orginal PO
number on the subsequent messages is covered by one of the suggested
solutions. Sort of analoguous to putting "Your reference: ...." on a
business letter.
 
There's also the possibility of the distributor-assigned identifier
being passed back via the supplier to the buyer on an acknowledgement -
which would be like telling the customer the airwaybill number of the
shipment.
 
As Perl enthusiasts say, TIMTOWTDI (there's more than one way to do it)
 
Expressing it as soap-ish practicalities, 
    a) the buyer could set up a different url for the returning delivery
notices for each purchase, which is passed round the loop
    b) the buyer could use ws-address/md type additional fields to have
a discriminated destination. again passed round the loop
    c) the parties could use a common header (ws-context style) that it
was agreed would be passed everywhere
    d) the buyer's message could have specific id field that the seller
and distributor knew must be copied on, identified as the buyers id
    e) the seller gets back to the buyer with an id that will be on the
distributor->buyer message.
 
    
Forcing CDL-based protocol design to use a particular pattern doesn't
seem right.  The question then is whether CDL makes all ways of doing it
look the same, which I think comes down to just how much of what kind of
information in the concrete is consdiered as covered by the channel
variable.  If one allows the concrete binding for a channel to include
assigning values in the messages themselves, and also allows it refer to
any sort of address, then all of a) to d) could be represented by
channel passing. (So could e), but the channel is passed the other way
round).
 
Having channel id's that mapped to anything from the tcp port number to
context-id in a header to //Buyer/POreference would seem to be similar
to where bpel correlation sets are heading on their issue 96
(engine-managed correlation sets =
http://www.choreology.com/external/WS_BPEL_issues_list.html#Issue96) .
Correlation sets are different in that they are usually concerned with
tieing together messages between the same parties, but the same
mechanisms apply.
 
Of course, saying "it's all in the binding" makes CDL more powerful (one
way to represent a lot of things that are similar in abstract) but
defers the practical utility.
 
Peter
Choreology Anti virus scan completed
Received on Thursday, 25 November 2004 09:59:21 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Saturday, 18 December 2010 01:01:11 GMT