[cdi-dev] [JBoss JIRA] (CDI-574) Should a disabled @Specialized disable a second bean?

[Mark Struberg (JIRA)]

    [ https://issues.jboss.org/browse/CDI-574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13133875#comment-13133875 ] 

Mark Struberg edited comment on CDI-574 at 11/25/15 11:18 AM:
--------------------------------------------------------------

Imo OWB is spec conform as well. It's all a matter of how you interpret the term BDA in the CDI spec. The inconsistency in Weld is a direct consequence of Weld interpreting the term 'BDA' as per-jar (section 12 interpretation option) and not as per-module (section 5 interpretation option). See CDI-18 and other issues for the discussion. In OWB the alternative is 'per ee module' by default. So the whole EE module (e.g. a single WAR) either has the alternative enabled or not. And thus the whole specialized bean is enabled for the whole ee module or not. That way we are a.) much faster (as we don't need to check the invoker chain each proxy invocation) and b.) more consistent. EE-modules usually are 1:1 to ClassLoaders and BeanManagers. Thus for each BeanManager a bean is clearly either enabled or not. 

It's an old story that the bda-per-jar interpretation is inconsistent within itself. How should that be solved? Will the observer method in the specialized away bean get invoked or not? What if I just add a SINGLE jar which don't have this Alternative enabled? Will the observer method then get invoked again? What about implicit bean archives? Those are almost always present, Just add a single slf4j api or whatever other util jar. In practice this will render disabling beans via specialisation useless. etc, . 



was (Author: struberg):
No, this is imo an inconsistency in Weld. It is a direct consequence of Weld interpreting the term 'BDA' as per-jar (section 12 interpretation option) and not as per-module (section 5 interpretation option). See CDI-18 and other issues for the discussion. In OWB the alternative is 'per ee module' by default. So the whole EE module (e.g. a single WAR) either has the alternative enabled or not. And thus the whole specialized bean is enabled for the whole ee module or not. That way we are a.) much faster (as we don't need to check the invoker chain each proxy invocation) and b.) more consistent. EE-modules usually are 1:1 to ClassLoaders and BeanManagers. Thus for each BeanManager a bean is clearly either enabled or not. 

It's an old story that the bda-per-jar interpretation is inconsistent within itself. How should that be solved? Will the observer method in the specialized away bean get invoked or not? What if I just add a SINGLE jar which don't have this Alternative enabled? Will the observer method then get invoked again? What about implicit bean archives? Those are almost always present, Just add a single slf4j api or whatever other util jar. In practice this will render disabling beans via specialisation useless. etc, . 


> Should a disabled @Specialized disable a second bean?
> -----------------------------------------------------
>
>                 Key: CDI-574
>                 URL: https://issues.jboss.org/browse/CDI-574
>             Project: CDI Specification Issues
>          Issue Type: Clarification
>          Components: Inheritance and Specialization
>    Affects Versions: 1.2.Final
>         Environment: n/a
>            Reporter: Emily Jiang
>
> In CDI specification Section 4.3:
> When an enabled bean, as defined in Section 5.1.2, “Enabled and disabled beans”, specializes a second bean, we can be certain that the second bean is never instantiated or called by the container. Even if the second bean defines a producer or observer method, the method will never be called.
> The spec says only an enabled bean can specialize a second bean. Can a disabled specialized bean specialize a second bean?
> Weld asserts a disabled specialized bean specializes a second bean while OWB asserts a disabled specialized bean does not specialize a second bean.
> This needs to be clarified.
> In more details:
> I have an application containing two wars.
> testDiffBDA.war
> testDiffBDA.war/WEB-INF/classes/test/diff/web/FrontEndServlet.class
> @Inject CounterProducerConsumerModified2 bean;
> beans-xml-modified2.jar
> containing one bean and an empty-ish beans.xml :
> @Inject at CounterModifiedQualifier String modifiedProducer;
> beans-xml-modified.jar.jar
> CounterModifiedQualifier  (the interface)
> CounterProducerModified (the bean implementing that interface)
> AlternativeCounterProducerModified (an alternative specialized bean)
>  beans.xml
>    <alternatives>
>         <class>com.ibm.jcdi.test.beansxml.AlternativeCounterProducerModified</class>
>    </alternatives>
> My application failed deployment with the error on Weld but worked on OpenWebBeans
> {code}
> [ERROR   ] CWWKZ0004E: An exception occurred while starting the application testDiffBDA. The exception message was: com.ibm.ws.container.service.state.StateChangeException: org.jboss.weld.exceptions.DeploymentException: WELD-001408: Unsatisfied dependencies for type String with qualifiers @CounterModifiedQualifier
>   at injection point [BackedAnnotatedField] @Inject @CounterModifiedQualifier com.ibm.jcdi.test.beansxml.CounterProducerConsumerModified2.modifiedProducer
>   at com.ibm.jcdi.test.beansxml.CounterProducerConsumerModified2.modifiedProducer(CounterProducerConsumerModified2.java:0)
> {code}
> After further investigation and talking to Martin from Weld, the error was caused due to the fact of AlternativeCounterProducerModified disabling the CounterProducerModified bean but itself is not enabled in the jar of beans-xml-modified2.jar. Therefore, no producer is active to produce a bean with the qualifier CounterModifiedQualifier.
> From Weld's perspective, any bean annotated with @Specialized disables a second bean regardless whether itself is active or not.
> My understanding is that the specialized should only take effect if itself is enabled. Otherwise, we run into the situation of where the specialized bean is not enabled but it disabled another bean. To me, it is wrong.



--
This message was sent by Atlassian JIRA
(v6.4.11#64026)