This seems to be due to classloading issues.
Using (2) creates a SystemDescriptor with a HK2Loader that uses the same
underlying ClassLoader as the AbstractBinder. This ClassLoader can see the
custom annotation type. Using (1) creates a SystemDescriptor with no
Hk2Loader that seems to be using the system classloader (I think). That
Hk2Loader can't see the annotation and so custom injection doesn't happen.
This also seems to explain the issue with knownInjectees being empty.
On Fri, Aug 9, 2013 at 12:42 AM, buko <buko_at_chiubarobot.com> wrote:
> Not sure if this is expected behavior or not but I've noticed that:
>
> (1) addActiveDescriptor(UserForm.class);
>
> In an AbstractBinder will *not* do annotation-based custom injection
> resolution. The field is left null and the known injectees is empty.
>
> But:
>
> (2)
> bind(TestForm.class).to(TestForm.class).in(PerLookup.class).qualifiedBy(new
> Annotation() {
>
> @Override
> public Class<? extends Annotation> annotationType() {
> return Bindable.class;
> }
> });
>
> Does do custom injection resolution.
>
> Is this right? I'd prefer to use (1) for binding stuff since the qualified
> syntax is quite annoying but does addActiveDescriptor somehow not
> detect/check for all annotations on it, particularly those custom injection
> resolvers might be interested in?
>