On Tue, Sep 19, 2017 at 12:09 pm, Bill Shannon wrote:

The CDI expert group didn't raise this at the platform level, and they didn't even do this consistently for their annotations

I'm not sure if there's any inconsistency, other than creating literals for the JSR 330 spec and including them in 299. Or is that what you're referring to?

Some of the annotations have literals and some do not.  There may be some rationale for why only some of the annotations need literals, but it wasn't explained.

, so I'm not clear on what their intent is with these annotation instances.  How would applications use them?

They're mostly used for the various builders that CDI has for dynamically adding annotations to beans. CDI already has a helper class for that that makes it somewhat easier (https://docs.jboss.org/cdi/api/2.0/javax/enterprise/util/AnnotationLiteral.html), but even with the helper class it's still a bit verbose.

For a practical example see: http://arjan-tijms.omnifaces.org/2017/08/dynamically-adding-interceptor-to-build.html

It's also used for lookups, as in this code:

myGreetings.select(NamedLiteral.of("northern")).get();

For the full context see: https://github.com/javaee-samples/javaee8-samples/blob/master/cdi/qualified-lookup/src/test/java/org/javaee8/cdi/qualified/lookup/QualifiedLookupTest.java#L41

In that latter example a bean is selected that has the @Named("northern") annotation applied to it.

Hopefully this is very much an advanced case and not something that typical applications will need to do.

If there's a reason applications need to be able to create annotation instances, wouldn't it be better if it worked the same whether or not the annotation has members? 

Perhaps, but without members the annotation instance is a totally static singleton, like enum values basically. With members this is obviously not the case.

If an annotation evolves to have members, would you still want a static instance?  Wouldn't it be simpler if there was always just one way to get an instance?

Maybe this really belongs in the JDK...

Possibly indeed, but not sure how feasible it is to get that in.

For now Java EE support throughout all applicable specs (basically the specs that already leverage CDI) isn't that hard at all. Had I learned about this annotation literals a tiny bit earlier I could have added them trivially to JSF and Java EE security (if the spec lead and EG would have agreed, of course).

Depending on each spec to do it means that some won't.

This is not the first thing about annotations that we wished every spec would handle.  I really wish we could push more of this into the JDK, or find a way to centralize all of this in a place that would apply to all Java EE uses of annotations automatically.

Lacking that, a "design pattern" as you suggest might be the best approach.  Another item for the future Eclipse project to address!  :-)