[weld-dev] AfterBeanDiscovery and observer resolver cache

[Martin Kouba]

Hi Christian,

actually invoking any container lifecycle event method after the
container initialization finished should have no effect. ABD event
reference can escape but it does not mean you can invoke ABD.addBean()
after ADV is fired.

The spec wording is not very explicit here:
"During the application initialization process, the container fires a
series of events, allowing portable extensions to integrate with
the container initialization process defined in Section 12.2."

Maybe we should file a new spec issue to clarify that such invocations
should result in IllegalStateException...

Martin


Dne 4.3.2014 17:42, Christian Sadilek napsal(a):
> Hi Jozef,
> 
> I think clearing the cache at the end of the Weld bootstrap process is not enough to solve that particular problem since a CDI extension can hold on to the ABD reference and invoke addObserverMethod later (multiple times) which causes the same problem I described below. There's no indication to the caller of addObserverMethod that it's in fact too late to call that method.
> 
> Since an ABD event reference can always escape (can be used outside the method that observes it) it seems this issue should be resolved (although it admittedly is an edge case).
> 
> Cheers,
> Christian
> 
> On 2014-03-04, at 11:29 AM, Jozef Hartinger <jharting at redhat.com> wrote:
> 
>> Hi Christian,
>>
>> this sounds like a bug. All the resolution caches should be cleared at the very end of Weld's bootstrap sequence (after ABD observers are called). (see https://github.com/weld/core/blob/master/impl/src/main/java/org/jboss/weld/bootstrap/WeldStartup.java#L415)
>>
>> Jozef
>>
>> On 03/04/2014 04:36 PM, Christian Sadilek wrote:
>>> Hi everyone,
>>>
>>> CDI extensions can observe the AfterBeanDiscovery event to register observer methods (addObserverMethod). However, when an event is first fired, the observers for that event are resolved and then cached (in TypeSafeResolver). All future calls to addObserverMethod for an already fired event with corresponding qualifiers will have no effect because the observer result is read from cache and not recomputed.
>>>
>>> >From an API perspective that's unfortunate because addObserverMethod will only work until an event (with corresponding qualifiers) is fired and there is no indication to the caller of that method that it didn't have any effect when invoked after that.
>>>
>>> Possible solutions:
>>>
>>> - Provide some public API to clear/recompute that part the observer cache. Maybe that exists? I couldn't find it which is why I am using the private API and Reflection :(. Also let AfterBeanDiscovery.addObserverMethod fail in that case with the advice to reset the cache.
>>>
>>> - Recompute the corresponding part of the cache when addObserverMethod is called (seems preferable).
>>>
>>> OpenWebBeans doesn't have this issue as their NotificationManager will simply add the new ObserverMethod to a ConcurrentHashMap that is also accessed when an event is fired.
>>>
>>> What do you think? Can this already be done or is there another solution?
>>>
>>> Cheers,
>>> Christian
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> weld-dev mailing list
>>> weld-dev at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/weld-dev
>>
> 
> 
> _______________________________________________
> weld-dev mailing list
> weld-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/weld-dev
>