Indirect inference for closure arguments fails depending on the contents of the closure

[elinorbgr]

Minimal example I could create:

trait Callback<Args> {
    fn call(&mut self, args: Args);
}

impl<F, Args> Callback<Args> for F where F: FnMut(Args) {
    fn call(&mut self, args: Args) {
        self(args)
    }
}

#[derive(Debug)]
struct Wrapper<T>(T);

#[derive(Debug)]
enum Foo {
    Bar,
    Baz,
    Baq
}

fn use_cb<C: Callback<Wrapper<Foo>>>(mut c: C) {
    c.call(Wrapper(Foo::Baz))
}

fn main() {
    // this one works
    use_cb(|variant| {
        println!("the variant is: {:?}", variant);
    });
    // this one fails to infer
    use_cb(|variant| {
        println!("the variant is: {:?}", variant.0);
    });
    // with these annotations it works
    use_cb(|variant: Wrapper<_>| {
        println!("the variant is: {:?}", variant.0);
    });
}

note that the sole difference between the first and second closure is that the first prints variant while the other tries to access its inner field.

It reproduces on both stable and nightly.