NLL promotes arbitrary code to be evaluated by miri if the result is unused

[oli-obk]

This takes a while to compile. I presume because loop {} needs to hit the loop checker in miri. Changing the _ to _foo or foo will immediately bail out due to &(loop{}, 1).1 not living for the 'static lifetime.

#![feature(nll)]

fn main() {

    let _: &'static usize = &(loop {}, 1).1;
}

(Playground)

Errors:

   Compiling playground v0.0.1 (file:///playground)
warning: unreachable expression
 --> src/main.rs:4:40
  |
4 |     let _: &'static usize = &(loop {}, 1).1;
  |                                        ^
  |
  = note: #[warn(unreachable_code)] on by default

    Finished dev [unoptimized + debuginfo] target(s) in 1.09s
     Running `target/debug/playground`
/root/entrypoint.sh: line 8:     7 Killed                  timeout --signal=KILL ${timeout} "$@"

[matthewjasper]

The following code times out with either borrow checker:

fn main() {

    let _ = &(loop {}, 1).1;
}

[nikomatsakis]

The reason that _ vs _foo matters is because of #47184

[nikomatsakis]

I don't think this is an NLL issue in particular?

[pnkfelix]

Based on @matthewjasper 's identification of a test that reproduces the issue on either borrow-checker, I am going to remove the A-NLL label.

[oli-obk]

That example is not problematic, because nothing is promoted at all.

I now don't even think my example is problematic anymore, because it seems to be perfectly alright to promote anything if we diverge before we give the user access to the value.

I guess this just needs a test now (just a ui test with a //compile-pass comment, not a run-pass test, as that would hang forever)

[saleemjaffer]

@oli-obk Can I work on this? I'm a newbie to rust.

[oli-obk]

Sure!

You can take the example from the issue message and create a new file in src/test/ui/consts/ and add the //compile-pass comment at the top.

Then, when running ./x.py test --stage 1 src/test/ui --test-args const --bless it should succeed to run the tests.