Unreliable warnings about unreachable code downstream of uninhabited expression

[graydon]

Code

use core::convert::Infallible;
use core::hint::unreachable_unchecked;

fn make_infallible() -> Infallible { 
    // Infallible implements From<!>
    loop { }
}

extern {
    fn side_effects();
}

pub fn has_unit_type() {
    let _a : Infallible = make_infallible();

    // This gets no unreachable-code warning
    println!("I cannot execute");

    // This gets no unreachable-code warning
    unsafe { side_effects() }

    // This gets an unreachable-code warning
    if false { }
}

pub fn has_never_type_no_warning() -> ! {
    let _ : Infallible = make_infallible();

    // This gets no unreachable-code warning (probably correctly?)
    unsafe { unreachable_unchecked() }
}

pub fn has_never_type_has_warning() -> ! {
    let _ : Infallible = make_infallible();
    // This has the same never-type, but does get a warning
    loop {}
}

Current output

warning: unreachable expression
  --> src/lib.rs:23:5
   |
14 |     let _a : Infallible = make_infallible();
   |                           ----------------- any code following this expression is unreachable
...
23 |     if false { }
   |     ^^^^^^^^^^^^ unreachable expression
   |
note: this expression has type `Infallible`, which is uninhabited
  --> src/lib.rs:14:27
   |
14 |     let _a : Infallible = make_infallible();
   |                           ^^^^^^^^^^^^^^^^^
   = note: `#[warn(unreachable_code)]` on by default

warning: unreachable expression
  --> src/lib.rs:36:5
   |
34 |     let _ : Infallible = make_infallible();
   |                          ----------------- any code following this expression is unreachable
35 |     // This has the same never-type, but does get a warning
36 |     loop {}
   |     ^^^^^^^ unreachable expression
   |
note: this expression has type `Infallible`, which is uninhabited
  --> src/lib.rs:34:26
   |
34 |     let _ : Infallible = make_infallible();
   |                          ^^^^^^^^^^^^^^^^^

Rationale and extra context

It should warn on the first (and every) unreachable expression, not just the seemingly random set of expressions it fires on here. Note that if you change any of the make_infallible() calls to panic!() it will warn at the correct spots. So I think this is specifically related to the treatment of empty types.

[graydon]

At a guess I would say this might be related to #100288 and so perhaps @camsteffen would be interested (i.e. it might be a typeck issue not a "diagnostics", it's hard for me to tell from my current level of ignorance of how this lint works)

[JonathanBrouwer]

This is a duplicate of #88393

[asquared31415]

I don't think this is a duplicate, but it's likely the other side of a similar base issue with how uninhabited types are related to reachability analysis. The issue you linked is about incorrectly emitting the unreachable code lint on code that is actually reachable, where as this is not emitting the lint on code that should have it emitted.

[JonathanBrouwer]

Ah I see indeed, skimmed over the issue a bit too quickly, sorry :P
I agree that the root cause is most likely the same though