[JSR308] Strawman proposal for repeated annotations

[Rémi Forax]

Joseph D. Darcy a écrit :
> Hello.
>
> One of the design choices of annotations as introduced in JSR 175 is 
> that repeated annotations are not allowed; that is, there can be at most 
> one application of a particular annotation to a given element.
>
> Various Java EE users of annotations wanted the ability of have multiple 
> annotations; the workaround using JSR 175 is to have a second container 
> annotation with an array-valued member, something like
>
> @interface MyAnnotation {
>     int value();
> }
>
> @interface MyRepeatedAnnotationContainer {
>     MyAnnotation [] value();
> }
>
> Addressing this limitation is one of the items included in this JSR's 
> scope.  Below is my strawman proposal to alleviate this issue.
>
> The proposal is to support syntactic sugar for the existing pattern 
> along with enhanced API support for retrieving the repeated values.  The 
> assumption of at most one value per annotation type is throughly baked 
> into the annotations design and the design of the APIs reading 
> annotations.  It is impractical to change that decision now; attempting 
> to do so would expose the numerous corner cases and interactions the JSR 
> 175 design avoided.
>
> First, a new meta-annotation is introduced to allow an annotation type 
> to declare what its containing type should be:
>
> @interface Container {
>   Class<? extends Annotation> value();
> }
>
> so that the declaration
>
> @Container(MyRepeatedAnnotationContainer.class)
> @interface MyAnnotation()
>
> would allow something like
>
> @MyAnnotatio(1) @MyAnnotation(2) @MyAnnotation(3)
> public class Foo();
>
> to be treated as
>
> @MyRepeatedAnnotationContainer(
>   {@MyAnnotation(1),
>    @MyAnnotation(2),
>    @MyAnnotation(3)})
> public class Foo();
>
> (I'm not sure what the exact syntactic sugar should be; it might be 
> beneficial to wrap the repeated annotations in "@{", "}" to emphasize 
> the grouping or to restrict the repeated annotations to be sequentially 
> applied.)
>
> Declaring the container in the annotation type provides the compiler and 
> the runtime the information needed to construct or lookup the containing 
> type.  There could be a new method on AnnotatedElement like
>
> <T extends Annotation> T[] getRepeatedAnnotation(Class<T> annotationClass)
>
> or
>
> <T extends Annotation> List<T> getRepeatedAnnotation(Class<T> 
> annotationClass)
>
> This method could potentially handle some of the tricky cases like 
> returning the non-contained annotation if there is one, otherwise return 
> the contained annotation, etc.
>
> The intention is to allow the @Container meta-annotation to be 
> retrofitted to existing wrapper annotations.
>
> Comments?
>
>   
i fully agree perhaps because i've proposed something similar in 
february :)
https://lists.csail.mit.edu/pipermail/jsr308/2007-February/000080.html 
(the last part of the message)

> -Joe
>   
Rémi