combining explicit self and closures yields an LLVM failure

[nikomatsakis]

Example test:

struct Box {
    x: uint;
}

impl Box {
    fn set_many(&mut self, xs: &[uint]) {
        for xs.each |x| {
            self.x = x;
        }
    }
}

fn main() {}

I presume this is because the type is obtained by taking the type associated with the node-id of the free variable, which is a def_self that is probably generic to the impl, and not specialized to the method. The quick fix is to patch that code, perhaps the better fix is to have a distinct self-id per method? Not sure. It seems wrong to be baking in "if this is self, do something special" all over the place.

@msullivan you may have an opinion.

[msullivan]

We do have a distinct self-id per method, as it turns out, but the type of it was tracked differently than regular variables, and as a result I missed updating it when explicit self went in. I'm fixing things to track the type of self in a more normal way.