Updated State of the Lambda

Stephen Colebourne scolebourne at joda.org
Mon Dec 12 05:01:18 PST 2011


On 12 December 2011 11:46, Maurizio Cimadamore
<maurizio.cimadamore at oracle.com> wrote:
> On 12/12/11 11:21, Stephen Colebourne wrote:
>>
>> At this point, I find the "default" keyword in default method
>> interfaces to be noise, and adding no value. Method bodies in
>> interfaces is not a complex concept, nor an unreadable one. The
>> default keyword makes it more complex and noisy, not less. Removing
>> "default" would also mean removing "default none", which again makes
>> perfect sense. Redclaring a method in a subinterface for documentation
>> purposes is rare, and for classes requires a call to super, so the
>> same should apply for interfaces.
>>
> Note that the presence of 'default none' is not accidental, nor driven by a
> syntax decision. An extension method whose body is 'default none' is used to
> explicitly get rid of any defender that might have been inherited up to that
> point. It is the equivalent of re-abstraction of concrete methods, but
> applied to extension methods. We feel like we need an explicit
> re-abstraction mechanism, as it is quite common in Java code to repeat an
> interface method declaration is a sub-interface just for documentation
> purposes. If the redeclaration was interpreted by the compiler as a
> re-abstraction we would be introducing subtle semantics change associated
> with such re-declaration of interface methods and the only way out for the
> developer would have been to re-specify the default implementation for the
> extension method (which would be the most common use case, as opposed to
> re-abstraction, which is quite rare).

Yes, there are complications, but I don't think "default none" is the
right solution.

The key problem (which I think is more in the JDK than in other code
I've seen) is documentation, but that documentation problem also
applies to classes. A better solution would be to provide a way to add
documentation to a sub interface/class without creating a new
signature in the bytecode.

Another solution would be to use the abstract keyword, perhaps at the
end in place of "default none".

I'm simply saying that "default" in general and "default none" in
particular don't have any appeal at this point.

Stephen


More information about the lambda-dev mailing list