This requires a lot of fix-up in the implementation classes because there is no binding between the types of the key and value list, although they are in fact by definition identical. This is the contract implemented by ClassToInstanceMultimap#get.
I think this would work a lot better if:
The abstract class took a second, type bound, parameter B,
Reimplement everything in terms of a ClassToInstanceMultimap<B<,
In the case of StatusMetadataLoggingStage, a type bound of StatusMetadata,
In the other subclasses, a type bound of ItemMetadata
This does involve an API change, but I doubt there are any other subclasses. It might be safest to turn that into an impl class, though.
The only deployment-level issue this might cause would be if someone had (wrongly) configured StatusMetadataLoggingStage with something that wasn't a StatusMetadata. That wouldn't have done anything, but would now probably result in a class cast or perhaps a compilation error.
The following four classes have a very complicated API and implementation:
AbstractItemMetadataSelectionStage
StatusMetadataLoggingStage
ItemMetadataFilterStage
ItemMetadataTerminationStage
The basic implementation generates and uses as part of the implementation API the following:
HashMap<Class<? extends ItemMetadata>, List<? extends ItemMetadata>>This requires a lot of fix-up in the implementation classes because there is no binding between the types of the key and value list, although they are in fact by definition identical. This is the contract implemented by
ClassToInstanceMultimap#get.I think this would work a lot better if:
The abstract class took a second, type bound, parameter
B,Reimplement everything in terms of a
ClassToInstanceMultimap<B<,In the case of
StatusMetadataLoggingStage, a type bound ofStatusMetadata,In the other subclasses, a type bound of
ItemMetadataThis does involve an API change, but I doubt there are any other subclasses. It might be safest to turn that into an impl class, though.
The only deployment-level issue this might cause would be if someone had (wrongly) configured
StatusMetadataLoggingStagewith something that wasn't aStatusMetadata. That wouldn't have done anything, but would now probably result in a class cast or perhaps a compilation error.