RFR: implementation for JEP 334: JVM Constants API

jbvernee jbvernee at xs4all.nl
Sun May 27 09:57:46 UTC 2018


1. After looking around some more, I also noticed that you can make the 
return type of `FieldTypeDescriptor::arrayType` in `Class` more 

public Class<T[]> arrayType() {
     return (Class<T[]>) Array.newInstance(this, 0).getClass();

I don't think I'm wrong in assuming that this always returns 
`Class<T[]>`? I imagine that could be very useful.


2. It's unfortunate that the return type of `describeConstable` in 
`EnumDesc` and `VarhandleDesc` can't be simplified any more currently. 
The cause seems to be that `DynamicConstantDesc` implements `Constable` 
on it's own, and does so with a less specific type parameter than is 
needed for the simplification. I can think of 2 ways to work around 
this. The first is to not let `DynamicConstantDesc` inherit from 
`Constantable` itself, but leave that up to the subclass, so it can use 
a more specific type parameter. For `EnumDesc` for instance that would 
be `Constable<EnumDesc<E>>`, then the return type of `describeConstable` 
could be `Optional<ConstantDesc<EnumDesc<E>>>`. I'm not sure how 
important it is for `DynamicConstantcDesc` to be a sub type of 
`Constable` though.

The other solution, which does keep it as a sub type, would be to add an 
additional type parameter to `DynamicConstantDesc` for the type with 
which to implement `Constable` i.e.:

public abstract class DynamicConstantDesc<T, C extends ConstantDesc<? 
super T>>
         implements ConstantDesc<T>, Constable<C> {

Then `EnumDesc` would extend `DynamicConstantDesc<E, EnumDesc<E>>`, and 
return `Optional<ConstantDesc<EnumDesc<E>>>` (similarly for 
`VarhandleDesc`). Basically `C` is the type implementing 
`DynamicConstantDesc` here. The default implementation of 
`describeConstable` in `DynamicConstantDesc` would have to be removed 
(making it abstract), or change it's implementation to return 
`Optional.empty()`, since there is no easy way (afaik) to return 
something that satisfies the new return type. So that pretty much forces 
subclasses to implement `describeConstable` themselves. Right now, both 
subclasses of `DynamicConstantDesc` override it any ways, and with the 
help of the factory methods `DynamicConstantDesc::of` an implementation 
looks trivial, so that shouldn't be too much of a problem for future 
subclasses (?).

It's the question if doing this is worth it for the sake of simpler 
return types for the subclasses. I think it is, since the abstract base 
is just a template to build classes on top of, but the subclasses and 
interfaces are what you end up using mostly (how often do you see 
`AbstractList` in code?) and you want their types to be specific and not 
some erased version.

(I hope I didn't make any mistakes, I can't build the JDK right now, so 
it's mostly been copy-and-pasting code into my own IDE)

Best regards,
Jorn Vernee

More information about the amber-dev mailing list