Feedback and comments on ARM proposal (original) (raw)

Mark Mahieu markmahieu at googlemail.com
Sat Mar 21 14:26:15 PDT 2009

On 21 Mar 2009, at 18:01, Joshua Bloch wrote:

I don't think I did a good job of explaining the problem. Consider this:

Ah, ok I'm with you! Thanks for the explanation.

The problem with Nope is that it contains two "magic methods" (sole parameterless method in an interface that directly extends MagicMarker). Is it sufficient to say that a class must have one and only one magic method to make it eligible for automatic resource management?

Right, clearly we'd need to be more specific about all this. I guess
I should offer something up, so here's a first stab at it (caveat:
not JLS-quality):

A variable/expression with static type T is eligible for automatic
resource management iff T has exactly one method m for which all of
the following hold: * m has no parameters * there exists one or more type S where: * S declares exactly one method, which is override-equivalent with m * S is an interface which directly extends MagicMarker * S :> T

I suppose we might also want to consider whether it should be an
error for a class to implement MagicMarker directly, or for an
interface that extends it to supply the 'wrong' number/type of
methods, etc.

Is it practical to implement this in the compiler? I dunno.

To me, it bears more than a passing resemblance to determining the
single-abstract-method in CICE, and I don't remember that aspect as
being particularly difficult to implement. Purely anecdotal though,
of course.

Josh

Mark

More information about the coin-dev mailing list