function type params are not checked for well-formedness

[aliemjay]

This unsound program shouldn't compile, but it does: https://play.rust-lang.org/?version=stable&mode=debug&edition=2015&gist=2e0af084bc8b2ec9f440a2c5dab7992f

use std::fmt::Display;

trait Displayable {
    fn display(self) -> Box<dyn Display>;
}

impl<T: Display> Displayable for (T, Option<&'static T>) {
    fn display(self) -> Box<dyn Display> {
        Box::new(self.0)
    }
}

fn extend_lt<T, U>(val: T) -> Box<dyn Display>
where
    (T, Option<U>): Displayable,
{
    Displayable::display((val, None))
}

fn main() {
    // The type parameter `U = &'static &'temporary str` is ill-formed
    // because it does not enforce `'temporary: 'static` ...
    let val = extend_lt(&String::from("blah blah blah"));
    println!("{}", val);
}

I was shocked discovering that. I thought WellFormed(U) is part of the caller bounds of extend_lt.

@rustbot label C-bug T-types I-unsound A-lifetimes

[aliemjay]

So this is a MIR borrowck bug because it regressed in v1.36 when it was enabled on 2015 edition.
@rustbot label regression-from-stable-to-stable

Remarkably adding type annotations to the call site

-    let val = extend_lt(&String::from("blah blah blah"));
+    let val = extend_lt::<_, &_>(&String::from("blah blah blah"));

makes it not compile because user annotations are always checked for well-formedness.

[oli-obk]

cc @lcnr

[aliemjay]

A case that regressed with full NLL in v1.63 - it was rejected by the HIR borrowck:

// ... same as above
fn test<'a>() {
    let _: fn(&'a str) -> _ = extend_lt;
}

[aliemjay]

I'd argue that this should be rejected as well:

trait Trait<A>: Sized {
    fn call(self) {}
}

impl<T> Trait<&'static T> for T {}

fn main() {
    Trait::call(&String::new());
}

This is necessary to fix #98852 in a sound way. At this point we should basically WF-check all fn item substs.

@rustbot claim

[jackh726]

This is a soundness bug that regressed fairly recently for some repros, so I think it's P-high.

https://rust-lang.zulipchat.com/#narrow/stream/245100-t-compiler.2Fwg-prioritization.2Falerts/topic/.23104005.20function.20type.20params.20are.20not.20checked.20for.20well-forme.E2.80.A6/near/308263767

[DianQK]

Hi, @aliemjay.
Are you still working on this issue? I would like to get started with MIR with this issue.

[aliemjay]

Hi, @aliemjay.
Are you still working on this issue? I would like to get started with MIR with this issue.

Hi, You can claim this. There is #104098, which I was too busy to finish. You can take it over to rebase, address review comments and push for a crater run.

[DianQK]

Hi, @aliemjay.
Are you still working on this issue? I would like to get started with MIR with this issue.

Hi, You can claim this. There is #104098, which I was too busy to finish. You can take it over to rebase, address review comments and push for a crater run.

Thanks, I'll give it a try. I added a corresponding question at https://rust-lang.zulipchat.com/#narrow/stream/182449-t-compiler.2Fhelp/topic/lifetime-issue-104005 as well.