avalon-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Leo Sutic" <leo.su...@inspireinfrastructure.com>
Subject RE: State of the Aspect
Date Tue, 09 Mar 2004 12:23:40 GMT


> From: Stephen McConnell [mailto:mcconnell@apache.org] 
>
> Can we dig into this (its something concrete) - as it raises 
> something 
> that is worrying me.  In Merlin a component type declares a bunch of 
> stuff about itself. This includes its dependencies (deployment and 
> runtime), the logging channels it may ask for, perhaps a 
> configuration 
> schema, etc.
>
>    * the component author is expecting the container to honor
>      the criteria expressed by the type
> 
>    * is it reasonable to consider a scenario where a container
>      that initiates deployment without ensuring a type's constraints
>      have been fulfilled - I don't believe so

A container should not start up a component if it can prove (based on
metadata) that the component will likely fail to start. There's simply 
no point.

However, if no metadata is present, or none in a format that could be
loaded, the container should assume that the application developer
knows his job and should attempt to start the component.
 
>    * is it reasonable for a container to be configured such that
>      it's responsibilities are scaled back - I think so - but only
>      if the container has an overriding responsibility not to
>      attempt component deployment involving a responsibility it
>      cannot assure

A container cannot assure anything. All it can do is use metadata
(which may or may not be correct) to attempt to prove that a component
will not fail to deploy due to unsatisfied requirements.

But the container should err on the side of assuming that the developer
knows what's best. I.e. - when in doubt, deploy. This makes sense,
because even if all dependencies are satisfied, you may still get
a big hearty NullPointerException in the component and a failed 
deployment.

A container should assume that if metadata is present, it is
also correct.

> In parallel I sure that the notion of aspects 
> can play a role as well - but I'm not sure where that 
> is just yet.

Me neither. (yet)

/LS


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@avalon.apache.org
For additional commands, e-mail: dev-help@avalon.apache.org


Mime
View raw message