Support qualifiers with members

[SentryMan]

Can now use a qualifier's members to help distinguish beans:

So given:

@Qualifier
@Retention(RetentionPolicy.RUNTIME)
public @interface TempQualifier {
  Scale[] value();

  String someOtherString();

  int defaultVal() default 0;

  NestedAnnotation[] inject() default {@NestedAnnotation()};

  enum Scale {
    CELSIUS,
    FAHRENHEIT,
  }

  @interface NestedAnnotation {

    Inject[] inject() default {};
  }
}

@Singleton
class Meters {

  Thermometer imperial;
  Thermometer metric;

  public Meters(
      @TempQualifier(value = Scale.FAHRENHEIT, someOtherString = "far") Thermometer imperial,
      @TempQualifier(value = Scale.CELSIUS, someOtherString = "celsi") Thermometer metric) {
    this.imperial = imperial;
    this.metric = metric;
  }
}

this will generate

public final class Meters$DI  {

  /**
   * Create and register Meters.
   */
  public static void build(Builder builder) {
    if (builder.isAddBeanFor(Meters.class)) {
      var bean = new Meters(builder.get(Thermometer.class,"@tempqualifier(defaultval=0, inject={@nestedannotation(inject={})}, someotherstring=\"far\", value={fahrenheit})"), builder.get(Thermometer.class,"@tempqualifier(defaultval=0, inject={@nestedannotation(inject={})}, someotherstring=\"celsi\", value={celsius})"));
      builder.register(bean);
    }
  }
}

[SentryMan]

Curious, it seems only JDK 11 has some weird issue that prevents this from working.

[SentryMan]

It seems that in JDK 11, the qualifier code generates differently, it'll wire, but the ability to use test mocking is impaired.

[SentryMan]

the problem here seems to be that there's a discrepancy with how the annotation are rendered at runtime with Annotation#toString(), might have to only support a single value member at this time.

[SentryMan]

I managed to get it working with regex, with the only limitation being that qualifiers with annotation members are only supported if the nested annotation type has only a single member.

[rob-bygrave]

Cool. So we probably should add a new test for the case where someone uses @Named("@Foo") [to be explicit that the @ is not trimmed - anyone using that convention isn't broken by this change].

I'm also thinking there could be a test for the "simple case" where the qualifier only has 1 attribute? Like what was the motivation for this feature / I'm not expecting anyone to use nested attributes so TempQualifier to me looks like it would not match the common case (it's more complex than the common case which is good for testing but to me looks like we don't actually test the common/expected case).

[SentryMan]

ehh, I reverted it, we could always do it later

[SentryMan]

Like what was the motivation for this feature / I'm not expecting anyone to use nested attributes

Was looking at Weld and got inspired. In any case, it helps cut down on annotation overload so I thought it was a good idea.