- From: Jean-Jacques Dubray <jeanjadu@Attachmate.com>
- Date: Sat, 17 Apr 2004 06:24:36 -0700
- To: "Nickolas Kavantzas" <nickolas.kavantzas@oracle.com>
- Cc: "Steve Ross-Talbot" <steve@enigmatec.net>, "WS-Choreography List" <public-ws-chor@w3.org>
Nick: Actually, I am not really back, just trying to understand your work informally. >It would be nicer though to >have you in the group where you could help us with CDL ;-> Hum...is that an invitation? Sounds like you have a lot of brain power already. I don't know if I am capable of adding anything to your work. >Anyhow, you are raising some very interesting questions and as Steve has >said earlier we are going to log issues (if they don't already exist) based >on your feedback and all other feedback we get from the public list. Actually, I was expecting a more substantive response, I guess we did not have an agreed upon choreography in place and it sounds like you took an unforeseen exception path in somewhat basic unit of work. I would hate to have this choreography park. It would be really great if you have some time and could compensate with an answer to these questions, as I really did not mean to raise any issue. At least I cannot see where I raised an issue or did I? JJ- > Interactions are bound to WSDL operations, as I understand the current > draft. This implies a few things and some questions: > > WS-CDL must be associated to some abstract WSDL service definitions > (port types but not ports)? Concrete WSDL defined by each participant > must built on this abstract WSDL? > > Why specify a message type in the interaction? Why not in the associated > WSDL? Could they be different? How do you guaranty that they cannot be > different if specified in two different places? > > As I understand it, the only operations that will be referenced are > One-Way and Request/Response, no Notifcation and Sollicite-Response will > be needed/allowed? I understand why you do that, but this is a big > departure from WSDL itself? > > A comment with respect to state alignment and work unit. It looks like > the granularity at which isAligned() can be used is too big and that > could defeat the purpose of work units along with state alignment which > are otherwise very useful concepts. BPSS offers alignment at the message > level (it is not because you received a message that this message is > understood as valid and can be processed by the receiver). If no > protocol is available to notify the sender of potential standard errors > for each message, then these error notifications must be implemented in > the choreography itself. This is not a satisfying solution, because you > end up acking the ack. As I learned it, it is only through the use of > widely agreed upon and structurally constant signals that you can really > guarantee state alignment otherwise, I am under the impression that it > will remain a wish rather than a fact. > > JJ-
Received on Saturday, 17 April 2004 09:27:19 UTC