Fwd: Exchange type issue

Woops forgot to send this to the list

Begin forwarded message:

> From: Steve Ross-Talbot <steve@pi4tech.com>
> Date: 5 December 2006 16:48:50 GMT
> To: "Monica J. Martin" <Monica.Martin@Sun.COM>
> Subject: Re: Exchange type issue
>
> I understand the "depending on which you look at it argument". Alas  
> in WS-CDL we have directionality of channels. So if I have two  
> roles, lets say roleA and roleB. If we have a channel called A2B in  
> which roleB is the toRole, then a notification that would go from  
> roleB back to roleA that is not correlated with any request would  
> either have to be done as a respond or a request that would need an  
> additonal channel called B2A.
>
> If we take the first case of it being modelled as a respond only.  
> This is confusing to designers/architects because it is not a  
> response to anything. The intention is to make it a notification  
> from B back to A with not notion of anything that needed to preceed  
> it. This is what many protocols require.
>
> If we take the second case of needing another channel. Clearly we  
> are making the tail wag the dog because we not require an  
> additional channel to be added for any notification.
>
> In the first case we have confusion and the second explosion.  
> Neither seems elegant. As an architect I would have though simply  
> having an attribute on the message exchange that says this is a  
> "notification" would be better cleaner and less confusing than any  
> other solution.
>
> And alas a notify is semantically  different to a request or a  
> response. If you step back and think about it the idea of a one-way- 
> request makes little sense. What does it mean in reality. The term  
> notification seems to capture what Web Service message exchanges  
> have singly failed to do thus far.
>
> Sorry but I find this very frustrating having spend the last 15  
> years building async solutions that seem to do this all of the time.
>
> Cheers
>
> Steve T
>
> On 5 Dec 2006, at 16:16, Monica J. Martin wrote:
>
>>
>> I appreciate all the comments made and understand the proposal  
>> (and have discussed with Charlton). I agree with Charlton, Abbie  
>> and Martin.
>>
>> One important point, a notification can be a request or a  
>> response, given how you look at it (which indicated to Charlton).  
>> I think the flag is appropriate because the notification could be  
>> an identifier on either (and as such we may consider that).  
>> Semantically the notify is no different than either (as I said  
>> given how you look at it - as a one-way request or a request- 
>> respond). Thanks.
>>
>>
>>
>>
>