Hi Jesper
On Wednesday 23 January 2013 12:36 AM, Jesper Pedersen wrote:
> Hi,
>
> On 01/21/2013 08:00 AM, Sivakumar Thyagarajan wrote:
>>> It is the other way around - the 1.5 MEF is deployed to EE 6 and a 1.6
>>> RAR
>>> is used calling the new method.
>>
>> This is not allowed, as such a combination would not pass the
>> compatibility tests.
>>
>
> There are other CTS tests in place that won't allow that to happen.
>
> That doesn't mean it is a valid use-case though.
A valid usecase maybe, but not a compatible/portable usecase to consider.
>>> Since MEFs are external to the application server we should have added a
>>> TimeoutMessageEndpointFactory with the method, but we didn't. Hence we
>>> should update the documentation with that fact.
>>
>> MEFs are /not/ external to the application server. This is where I think
>> I am not following the point above. One of the MEF implementation in the
>> case of a 'full profile' EE6 application server would be a MDB
>> container. If a Connector 1.6 standalone connector container chooses to
>> support Message Inflow, the container must provide a MEF with the new
>> method. So, in both of these cases though a MEF implementation is
>> external to the connector container implementation, it is still
>> delivered along with the connector implementation and hence the
>> old-on-new scenario would not occur.
>
> Note, I said external to the EE implementation - meaning the platform that
> passes the TCK.
>
> It doesn't mean that the services are deployed externally to the EE process.
>
> The use-case is a SOA provider which provides a MEF implementation and
> deploys that on an EE implementation using their vendor specific
> integration points - in this case this implementation is the "app server"
> in the context of chapter 13.
Ah, okay. Now, I understand what you meant by old-in-new. In this case,
the SoA provider's MEF implementation is deployed onto the EE
implementation in a /vendor-specific/ fashion, and a 1.7 implementation of
that application server could reject the deployment of an old MEF
implementation, because that MEF cannot service 1.7 RAs. Again all of this
is beyond the usecases envisioned and covered in the spec.
> This scenario isn't covered in the spec - explicit.
Exactly, and so we can't design for this. Would you agree?
Thanks
--Siva.