'static closures/FnDefs/futures with non-'static return type are unsound

[steffahn]

Argument-free closures can return non-'static types but be 'static themselves:

fn foo<'a>() {
    let closure = || -> &'a str { "" };
    assert_static(closure);
}

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

However, associated types of 'static types must be 'static, too. So the fact that the above compiles results in unsoundness, as demonstrated below:

use std::fmt;
trait Trait {
    type Associated;
}

impl<R, F: Fn() -> R> Trait for F {
    type Associated = R;
}

fn static_transfers_to_associated<T: Trait + 'static>(
    _: &T,
    x: T::Associated,
) -> Box<dyn fmt::Display /* + 'static */>
where
    T::Associated: fmt::Display,
{
    Box::new(x) // T::Associated: 'static follows from T: 'static
}

fn make_static_displayable<'a>(s: &'a str) -> Box<dyn fmt::Display> {
    let f = || -> &'a str { "" };
    // problem is: the closure type of `f` is 'static
    static_transfers_to_associated(&f, s)
}

fn main() {
    let d;
    {
        let x = "Hello World".to_string();
        d = make_static_displayable(&x);
    }
    println!("{}", d);
}

(playground)

����IV�

@rustbot modify labels: T-compiler, T-lang, A-closures, A-lifetimes, A-traits, A-typesystem, A-associated-items
and someone please add “I-unsound 💥”.

Found this bug while experimenting with variations of #84305.

Without the dyn keyword, this seems to compile “successfully” since Rust 1.4.0. Godbolt link.

[steffahn]

This also works with functions

fn make_static_displayable<'a>(s: &'a str) -> Box<dyn fmt::Display> {
    //let f = || -> &'a str { "" };
    fn f<'a>() -> &'a str { "" }
    // problem is: the type of `f::<'a>` is 'static
    static_transfers_to_associated(&f::<'a>, s)
}

but only since Rust 1.20. I don’t know if this would be better put as a separate issue since that’s probably a regression then.

1.19 gives

error[E0495]: cannot infer an appropriate lifetime due to conflicting requirements
  --> <source>:24:36
   |
24 |     static_transfers_to_associated(&f::<'a>, s)
   |                                    ^^^^^^^^
   |
note: first, the lifetime cannot outlive the lifetime 'a as defined on the function body at 20:1...
  --> <source>:20:1
   |
20 | / fn make_static_displayable<'a>(s: &'a str) -> Box<fmt::Display> {
21 | |     //let f = || -> &'a str { "" };
22 | |     fn f<'a>() -> &'a str { "" }
23 | |     // problem is: the type of `f::<'a>` is 'static
24 | |     static_transfers_to_associated(&f::<'a>, s)
25 | | }
   | |_^
note: ...so that reference does not outlive borrowed content
  --> <source>:24:46
   |
24 |     static_transfers_to_associated(&f::<'a>, s)
   |                                              ^
   = note: but, the lifetime must be valid for the static lifetime...
note: ...so that the type `fn() -> &str {make_static_displayable::f::<'a>}` will meet its required lifetime bounds
  --> <source>:24:5
   |
24 |     static_transfers_to_associated(&f::<'a>, s)
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

error: aborting due to previous error(s)

Compiler returned: 101

(compiler explorer)

[steffahn]

The version with functions regressed (started compiling without error) at nightly-2017-06-29.

> rustc +nightly-2017-06-28 --version
rustc 1.20.0-nightly (f590a44ce 2017-06-27)
> rustc +nightly-2017-06-29 --version
rustc 1.20.0-nightly (69c65d296 2017-06-28)

This includes

4079e6128f6cff3270ed2b0ebdc62fb6b1f4b5d7 - Auto merge of #42417 - eddyb:separate-fn-sig, r=nikomatsakis
88c3242ef26956266970d62571a1d7d079b5799a - Auto merge of #42431 - nagisa:core-float-2, r=alexcrichton
5bc89416388f0a852f6e48542bf2fc021682fcb5 - Auto merge of #42709 - stepancheg:discriminant-hash, r=jseyfried
6b52a1162eff2469c0da88300577863e5cc0c78f - Auto merge of #42931 - arielb1:statement-visitor, r=eddyb
c16930762a4a7762b26579f7d44273d3bab4c11e - Auto merge of #42745 - sfackler:1.19-stabilization, r=alexcrichton
47faf1d51952ecd9d4c8a7325332fba34fbe00bd - Auto merge of #42819 - scottmcm:swap-nonoverlapping, r=sfackler
69c65d29615c391c958ebf75dd65258ec23e175c - Auto merge of #42850 - estebank:unwanted-return-rotj, r=nikomatsakis

Looking at those, almost certainly #42417 is the relevant PR.

[apiraino]

Assigning priority as discussed as part of the Prioritization Working Group procedure and removing I-prioritize.

@rustbot label -I-prioritize +P-high