AbstractList etc. functionality as interfaces with default methods? (original) (raw)

Remi Forax forax at univ-mlv.fr
Thu May 14 14:26:45 UTC 2015

javac should emit a lint warning in these cases to help API designers. I have created 8080413 [1] to track that.

Rémi [1] https://bugs.openjdk.java.net/browse/JDK-8080413

On 05/14/2015 04:04 PM, Brian Goetz wrote:

The static-instance asymmetry cancels that one out. If you have

class Foo { void m(int x) static void m(Foo f, int x) } Then Foo::x could either be an unbound mref to the instance method or a static mref. Even with a perfect target type (Foo, int) -> void, compiler will still report ambiguity. On May 14, 2015, at 9:52 AM, Vitaly Davidovich <vitalyd at gmail.com> wrote:

Ambiguous in isolation, but within context they're quite different: one takes an arg and the other is void.

sent from my phone On May 14, 2015 9:25 AM, "Remi Forax" <forax at univ-mlv.fr> wrote:

On 05/14/2015 03:05 PM, Brian Goetz wrote: Not only is there a problem with modCount, but also with equals/hashCode/toString. You can’t define these Object methods in an interface.

They could be defined as static methods to delegate to. From API consistency perspective, we have for example the following static methods on primitive wrapper classes: Right. We considered this during Lambda, but by the time we got here, we concluded that this was mostly trading one downside for another. It seemed overwhelmingly likely that people would forget to override equals/hashCode/toString in this case, and create collections that violated the contract. The other problem is that it creates ambiguous method references, if you have a class or an interface like: class A { public static int hashCode(A a) { ... } } A::hashCode is ambiguous. cheers, Rémi

More information about the core-libs-dev mailing list