Can't fulfill 'static bound with &'static pointer.

[bblum]

This should be legal but isn't:

fn foo<T: 'static>(_: T) {}

fn bar<T>(x: &'static T) {
    foo(x);
}
fn main() {}

@nikomatsakis ?

nominating feature-complete

[nikomatsakis]

no bug. The type parameter T could contain borrowed pointers, as the error message correctly pointed out. You want:

fn foo<T: 'static>(_: T) {}

fn bar<T: 'static>(x: &'static T) {
    foo(x);
}
fn main() {}

[bblum]

Not to mean this should change, but it should also be impossible to (safely) construct a &'static &'r T, right?

[nikomatsakis]

That is true.

[nikomatsakis]

I decided you're right, we ought to try &'static T as possessing the static bound for all T, since there are already other parts of the type system that rely on the fact that for all &'foo T, T cannot have region pointers that have shorter lifetime than 'foo

[bluss]

Here is a testcase for putting &'static fn() closures into trait objects. Of course sendable closures are also interesting (do we have to use extern fn for that?).

/*
error: cannot pack type `~Test`, which does not fulfill `'static`, as a trait bounded by 'static
let trait_obj = (~t) as ~EvalOnce:'static;
                ^~~~~~~~~~~~~~~~~~~~~~~~~
*/

struct Test {
    f: &'static fn:'static(),
}

trait EvalOnce {
    fn eval(~self);
}

impl EvalOnce for Test {
    fn eval(~self) {
        (self.f)();
    }
}

fn test_me() {
    let t: Test  = Test { f: || { } };
    let trait_obj = (~t) as ~EvalOnce:'static;
}

[msullivan]

Seconding the nomination for feature complete, I suppose?

[catamorphism]

accepted for production-ready

[pnkfelix]

Could be added backwards-compatibly in the future. Assigning P-low, not 1.0 blocker.

[pnkfelix]

@nikomatsakis the natural generalization here, assuming we finish #5121 by allowing lifetime kind bounds other than 'static, is that fn foo<T:'b>(x: &'a T) { ... } would imply that actually T: 'a + 'b, right?