[cdi-dev] Support DI within BV constant validators

[Bill Shannon]

I don't have enough background to evaluate the specific issue being
discussed here, but let me comment on the general issue of whether the
Java EE platform spec should be the place to define how all the component
technologies integrate with each other.

Originally, that's more or less the approach we took, but it has proven
to be somewhat limiting.  With more and more component technologies being
available independently of the platform, it's important to define how these
technologies work together even if they're not part of a Java EE product.

Thus, we've moved to an approach that's more like this...  In the spec
for technology X, we might have requirements of the form "when technology X
is included in a product that also includes technology Y, the following
additional requirements apply...."  the advantage of this approach is that
you can tell how the pieces have to fit together in the many contexts where
they can be used.  A disadvantage is that there's no longer the "master list"
of all the integration requirements.

This approach works best when the integration points between specs are
fully defined by the specs.  It works less well when the integration
points rely on product-specific interfaces.  Fully defining all the
integration points for all our technologies is a huge task that we'll
probably never complete.  And defining such integration points is often
less important than improving the application-visible APIs.  We live in
a world of compromise with limited time and resources.


Gunnar Morling wrote on 08/13/12 13:42:
> Hi Pete,
> 
> I can see your concerns, though I'm personally not really convinced of
> putting this into the BV spec either.
> 
> I like the idea of BV providing services/functionality, which then can
> be leveraged by integration technologies such as CDI/JEE or e.g.
> Spring. Personally I feel we shouldn't refer to these integrating
> technologies in the BV spec.
> 
> Wouldn't the JEE spec be the right place for defining the integrations
> between the different contained specs/technologies? Those wouldn't
> have to deal with such integration questions, could be used
> independently under SE and closely integrated under EE.
> 
> Do you know whether there are any ideas already for using CDI within
> JPA entity listeners? I think this should be rather similar: a spec
> defines so-far unmanaged objects which should make use of CDI under
> EE.
> 
> --Gunnar
> 
> 
> 2012/8/13 Pete Muir <pmuir at redhat.com>:
>> Or, we could move that requirement from the CDI spec to the BV spec, and have BV cover it all...
>>
>> If we put too much into CDI, it ends up becoming an integration dumping ground, which doesn't work so well, IMO.
>>
>> On 5 Aug 2012, at 22:11, Gunnar Morling wrote:
>>
>>> Hi Pete,
>>>
>>> In the BV 1.1 early draft we're currently saying the following [1]:
>>>
>>> ===
>>> Java EE should [...] enable Context and Dependency Injection (CDI)
>>> support to ValidatorFactory instances. In particular:
>>>
>>> 1) Let Validator and ValidatorFactory object be injectable.
>>> 2) Use a default ConstraintValidatorFactory implementation that
>>> returns CDI managed ConstraintValidator objects. The scope of these
>>> instances is dependent as the Bean Validation provider is responsible
>>> for them.
>>> 3) Provide CDI managed instances of ConstraintValidatorFactory,
>>> MessageInterpolator and TraversableResolver if customized classes are
>>> requested in the XML deployment descriptor.
>>> ===
>>>
>>> I think that's pretty much in line with what you're saying.
>>>
>>> What caught my attention was section 3.7 of the CDI spec which
>>> mentions built-in beans for Validator and ValidatorFactory to be
>>> provided by an EE container. So this covers 1) from above, but not the
>>> others.
>>>
>>> I'm wondering now whether 2) and 3) should be mentioned there as well
>>> or whether the entire section should be moved to the Java EE spec (I
>>> think that's what you're saying?).
>>>
>>> --Gunnar
>>>
>>> [1] http://beanvalidation.org/1.1/spec/#d0e6698
>>>
>>>
>>> 2012/8/2 Pete Muir <pmuir at redhat.com>:
>>>> Hi Emmanuel, Gunnar,
>>>>
>>>> Gunnar filed https://issues.jboss.org/browse/CDI-241 however I don't think we should reference this in the CDI spec. Instead, Bean Validation should declare which of it's classes should be "Java EE component classes", which then makes them eligible for injection by CDI. These types are also referenced by the Java EE spec IIRC. ccing Linda and Bill, as this relates to how we define what receives CDI injections going forward, as I'm not sure they want to continue with this method.
>>>>
>>>> Pete
>>