[cdi-dev] Thoughts about CDI-527

[Antoine Sabot-Durand]

Hi guys,

Re-reading materials and all our discussion on CDI-527, I think there's an
option we didn't explore

Instead of introducing a new annotation, we could loose rules on proxied
bean creation for class having final methods.
Today if you declare a class requiring a proxy (because of scope or AOP)
you'll get an exception, so nobody declares a bean with such a feature
since the app wouldn't deploy.
It's the same for producer: you won't declare an @ApplicationScoped
producer for a type having final method since your app won't deploy.

So in today existing code, final method are in non cdi code or in cdi code
that doesn't requires proxy (@Dependent without AOP)

If we decide to loose the rules in "3.15 Unproxyable bean types" [1] it
won't have impact on existing code (unless I missed something), since you
can't have proxied bean with final method without deployment error.
It would only allow new proxied bean with final method in it.

Change should also be made in "7.2. Container invocations and interception"
[2] to explicitly exclude final methods from the list of business method.

This change could be seen as not backward compatible, but since current
code cannot have proxied bean with final class it doesn't change existing
behavior but allow new usage.
It also allow code with producer for class of the JDK introducing final in
their code to continue working seamlessly in CDI 2.0.
Finally that prevent us to introduce specific annotation or configuration
looking like a hack.

In the past we already introduce behavior changes when we changed the list
of bean defining annotation in CDI 1.2 or when we loose usage of certain
beanManager method in AfterBeanDiscovery event (I see this proposal very
like this last example).

Wdyt ?

Antoine


[1] http://docs.jboss.org/cdi/spec/1.2/cdi-spec.html#unproxyable
[2] http://docs.jboss.org/cdi/spec/1.2/cdi-spec.html#biz_method
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/cdi-dev/attachments/20160216/275ea3ae/attachment-0001.html