[cdi-dev] @PostConstruct and @Inject methods in superclasses

[Stuart Douglas]

On 02/10/2011, at 2:16 PM, Pete Muir wrote:

> There is no need to refer to the interceptors spec to understand this. But the discussions about the interceptors spec is interesting :-)
> 
>>> But what about my example 4, the shared @Inject method + @PostConstruct case?
>>> Do we need to invoke all superclass injection first? thus do all @Inject fields + @Inject methods + @PostConstruct first, then the one from the Horse class?
>>> Could we also specify that all @Inject methods needs to get called before @PostConstruct?
> 
> JSR-330 says that injected fields happen before methods, and always superclass-first i.e.
> 
> class X extends Y { ... }
> 
> any injected field in Y, followed by any injected method in Y followed by any injected field in X followed by any injected method in X.
> 
> CDI then covers the @PostConstruct in 5.5.2 of the CDI spec.
> 
> The container must ensure that:
> •	Initializer methods declared by a class X in the type hierarchy of the bean are called after all injected fields declared by X or by superclasses of X have been initialized, and after all Java EE component environment resource dependencies declared by X or by superclasses of X have been injected.
> •	Any @PostConstruct callback declared by a class X in the type hierarchy of the bean is called after all initializer meth- ods declared by X or by superclasses of X have been called, after all injected fields declared by X or by superclasses of X have been initialized, and after all Java EE component environment resource dependencies declared by X or by su- perclasses of X have been injected.
> 
> Thus, 
> 
> You can be assured that the @PostConstruct for Y happens after injection, and the same for X. However you can't be certain that the @PostConstruct for Y happens before the injection for X.
> 
> Do we need to clarify this? I'm not sure we do? If we do, why?
> 
> On 1 Oct 2011, at 06:21, Stuart Douglas wrote:
> 
>> 
>> On 01/10/2011, at 7:47 PM, Mark Struberg wrote:
>> 
>>> Oki found it there:
>>>> most general superclass first.
>>> of course the interceptors_spec_1_1 only defines this for 'Lifycycle interceptors' 
>> 
>> The spec is not very clear on this point, but @PostConstruct methods on a bean are considered lifecycle interceptors.
> 
> Correct. I think this would need clarifying in the interceptors spec - Mark could you file an EJB spec issue for this?
> 
>> There is actually a fair bit of deviation between CDI and the interceptors and Managed bean spec on some points: 
>> 
>> - CDI beans don't support interceptor methods on the beans themselves (at least Weld does not, the interceptors spec allows you to have an @AroundInvoke method on the managed bean itself).
> 
> Wow, I totally missed that before. I think this is just a bug in Weld. Could you create an issue?

Actually it looks like I was wrong, weld does support this, for some reason I did not think it did. I really should have double checked before sending.

Stuart

> 
>> - @Resource @Psersitence* @EJB etc annotation of CDI managed beans do not create JNDI bindings like they are supposed to according to the EE platform spec.
> 
> This is partly a Weld bug, partly a Java EE integration issue, which never got resolved properly. The CDI spec doesn't define this, where is it in Java EE? If it's there, then it's an impl bug, not a spec clarification I think.
> 
>> 
>> As far as I know no one actually really uses these 'features' anyway, but it is a deviation from the spec that we should probably clarify.
>> 
>> Stuart
>> 
>