Scope type without @Scope or @NormalScope

[koichik0818]

CDI spec (2.4.2. Defining new scope types) says

All scope types must also specify the @javax.inject.Scope or @javax.enterprise.context.NormalScope meta-annotation

But a scope type without the @Scope or @NormalScope works well.

1) declare scope type without @Scope or @NormalScope.

@Target( { TYPE, METHOD, FIELD })
@Retention(RUNTIME)
public @interface SomeScoped {}

2) register the scope.

beforeBeanDiscovery.addScope(SomeScoped.class, true, false);

3) register the context.

afterBeanDiscovery.addContext(new SomeContext());

I think that the @SomeScoped should become Error If all scope types must specify @Scope or @NormalScope.

But Weld accept the @SomeScoped if I register it (step 2).

Which is right?

[nickarls]

If this it the case, I would call it a bug. Could you file a JIRA?

[koichik0818]

Does that mean it's a Weld's bug?
If so, I have another question.

CDI spec (2.4.2. Defining new scope types) says

Custom scopes are normally defined by portable extensions, which must also provide a context object, as defined in Section 6.2, 'The Context interface', that implements the custom scope.

But, I can use a scope type without register (I must register only context).

Is this portable? Depends on implementation(Weld)? or Weld's bug?

[nickarls]

Without digging that much into the spec, I'm guessing:

They are not linked. Context.getScope() returns the scope type but I'm not sure if the container is obliged to check if it's a registered scope. Of course, if you intend to use it normally @FooScoped, you should register it.

[koichik0818]

Thanks for the reply, but I want to know the correct interpretation of CDI spec.

I think that the Weld's behavior is rational.

1) I do not have to register a scope type with @Scope or @NormalScope.
2) I must register a scope type without @Scope or @NormalScope.

However, 2) does not match CDI spec.
Then I do not have to call BBD#addScope() when using Weld.

In CDI spec, when do I have to register a scope type?
Why is there BBD#addScope()?

[pmuir]

Koichi Kobayashi wrote on Feb 10, 2010 02:21:


Thanks for the reply, but I want to know the correct interpretation of CDI spec.

I think that the Weld's behavior is rational.

1) I do not have to register a scope type with @Scope or @NormalScope.
2) I must register a scope type without @Scope or @NormalScope.

However, 2) does not match CDI spec.
Then I do not have to call BBD#addScope() when using Weld.

In CDI spec, when do I have to register a scope type?
Why is there BBD#addScope()?

Both (1) and (2) are correct and are contained in the spec. addScope() has the effect of causing the CDI implementation to see the annotation as having @Scope/@NormalScope.

[koichik0818]

Pete Muir wrote on Feb 17, 2010 12:23:



Both (1) and (2) are correct and are contained in the spec.

(2) means that I can use a scope type without @Scope or @NormalScope.

for example,

@Target( { TYPE, METHOD, FIELD })
@Retention(RUNTIME)
public @interface BadScoped {}

I can use the scope type if (and only if) I register it.

public void beforeBeanDiscovery(@Observes BeforeBeanDiscovery bbd) {
    bbd.addScope(BadScoped.class, true, false);
}

But CDI spec (2.4.2. Defining new scope types) says

All scope types must also specify the @javax.inject.Scope or @javax.enterprise.context.NormalScope meta-annotation

Therefor, @BadScoped Should become error, shouldn't it?

[pmuir]

This is not the intent of the spec, I guess it just needs some clarification - Gavin reads these forums, so will pick it up.

[gavin.king]

Pete Muir wrote on Feb 17, 2010 12:23:



Both (1) and (2) are correct and are contained in the spec. addScope() has the effect of causing the CDI implementation to see the annotation as having @Scope/@NormalScope.

Right, exactly.

[gavin.king]

There is no problem here, the whole SPI is understood to be about overriding and messing with the annotation-based metamodel. That's the point of it.

[koichik0818]

Thanks for the reply. I understood it.