Static methods in traits

[Monnoroch]

Is this a bug or a feature?

trait MyTrait {
    fn act();
}

fn act<T:MyTrait>(val: &T) {
    T::act();
}

Gives:

main.rs:19:6: 19:12 error: unresolved name
main.rs:19      T::act();
                ^~~~~~
main.rs:19:6: 19:12 error: use of undeclared module `T`
main.rs:19      T::act();
                ^~~~~~
main.rs:19:6: 19:12 error: unresolved name `T::act`. Did you mean `val`?
main.rs:19      T::act();

[huonw]

This is a "feature" (well, a bug too, but it's well covered elsewhere, e.g. #6894). Static methods need to be called like MyTrait::act(). However, this doesn't allow the Self type to be infered correctly. The standard work around is using a dummy parameter that mentions Self. e.g.

trait MyTrait {
    fn act(_dummy: Option<Self>);
}

fn act<T: MyTrait>(val: &T) {
    MyTrait::act(None::<T>); // call T's act method
}

This precise behaviour is covered by "uniform function call syntax", which is being discussed in RFCs like rust-lang/rfcs#4.

(Thanks for filing.)

[Monnoroch]

Thanks for the answer!
But will it be fixed sometime? I mean, this workaround is really messy. It might be that i don't even own the MyTrait code, so i can't change it.

But my act actually returned Self, so i didn't even use this hack.

[huonw]

Yes, it will be fixed; the RFC I linked above is proposing a fix, but there is still some discussion happening in this area.

If you don't own MyTrait and the static methods don't have a workaround like this, then no-one can use them, not even the original library, i.e. the library is fundamentally broken.