RFR: 8004912: Repeating annotations - getAnnotationsByType is not working as expected

[Peter Levart]

Hi Andreas,

I think this is correct now. By coincidence, Joe Darcy is trying to 
implement default methods in interface AnnotatedElement and I think 
those two attempts could be merged. The s.r.a.AnnotationSupport static 
methods: getDirectlyAndIndirectlyPresent & getAssociatedAnnotations 
could be implemented in terms of AnnotatedElement's 
getDeclaredAnnotation(Class) - not needing to pass annotation maps 
around and not needing JavaLangAccess. The major obstacle that you have 
already solved somehow in your implementation is keeping the right order 
of returned annotations. For example:

#1

@Ann(1)
@AnnCont({@Ann(2), @Ann(3)})

vs.

#2

@AnnCont({@Ann(2), @Ann(3)})
@Ann(1)


Is this corner case really so important that specification could not be 
"tweaked" a little to say that #2 is equivalent to #1 as far as order or 
returned annotations is concerned (first comes any directly present and 
then any in-directly present annotations). If spec. could be tweaked, 
the implementations would be much simpler. Here's a rough sketch of such 
simplified AnnotatioSupport:

http://cr.openjdk.java.net/~plevart/jdk8-tl/RepeatingAnnotations/webrev.01/


Regards, Peter


On 10/23/2013 05:36 PM, Andreas Lundblad wrote:
> On 10/22/2013 03:20 PM, Peter Levart wrote:
>> I think the problem could be solved in two ways:
>>
>> - by explicitly scanning the inheritance chain for the 1st class that 
>> has the annotations of type T present directly or indirectly (by 
>> containment).
>>
>> - by "canonicalizing" the representation of the repeating annotations 
>> in the class file attributes. By this I mean that each repeating 
>> annotation is placed inside it's container at compile-time even if it 
>> is a single annotation of a particular type. This would mean that 
>> some features like specifying different @Targets or @Retentions for 
>> repeating annotation types and their containers would be prohibited 
>> and the specification would have to be changed. But I think this way 
>> the "inheritance" aspect of repeating annotations would be consistent 
>> even if not "looking through" containers (by using the old JDK7 API)...
>>
>> The "canonicalization" could be performed at runtime though, and I 
>> think there were such attempts already in the past. What happened to 
>> them?
>>
>>
>> Regards, Peter
>>
>
> Hi Peter,
>
> A new patch is available for review here:
>
> http://cr.openjdk.java.net/~alundblad/inherited-associated/
>
> This includes a new test based on your code. The test passes after 
> applying the patch.
>
> best regards,
> Andreas Lundblad