Unused results lint fails on trivial program

[milibopp]

The unused_results lint appears to be somewhat overzealous on the current nightly compiler. I managed to reduce the issue down to a simple hello world-program, so something is definitely broken.

I tried this code:

#![deny(unused_results)]

fn main() {
    println!("hello world!");
}

I expected to see the code compile successfully without any warnings.

Instead, the following error message showed up.

error: unused result
 --> src/main.rs:4:5
  |
4 |     println!("hello world!");
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^
  |
note: lint level defined here
 --> src/main.rs:1:9
  |
1 | #![deny(unused_results)]
  |         ^^^^^^^^^^^^^^
  = note: this error originates in a macro outside of the current crate

Meta

$ rustc --version
rustc 1.21.0-nightly (13d94d5fa 2017-08-10)

[Mark-Simulacrum]

Bisected to #43728, cc @zackmdavis, @eddyb

[kennytm]

Every function call, even those returning () or !, are considered unused now. 😕

#![deny(unused_results)]
fn g() {}
fn main() {
    g(); //~ ERROR unused result
}

[pengowen123]

I would like to work on this if no one else wants to.

[zackmdavis]

I see. During #43728, I put the new ununsed-for-functions code after the existing unused-must-use code but before the unused-results check, and modified the unused-must-use code to not return early for types that couldn't be results (including, notably, () and !), because I wanted to "fall through" to the unused-for-functions code. But () and ! shouldn't furthermore "fall through" and be seen as unused results.

😰 😢 😞 Sorry, I can submit a fix this afternoon. Thanks for the report, @aepsil0n.

[zackmdavis]

@pengowen123 I think I'm morally obligated to do everything in my power to minimize the number of nightlies that are contaminated with my mistake, so I should probably take this unless you're really enthusiastic about fixing it today?

[pengowen123]

@zackmdavis I should be able to fix it today. Don't worry about it being your mistake, I will work on it.

[zackmdavis]

@pengowen123 Works for me. I think returning early before the unused-results check if t.sty is ty::TyNever or the empty tuple should suffice? And we'd also want a UI test of the unused_result lint working as expected to make sure this can never happen again.

[alexcrichton]

This is being fixed in #43813

[pengowen123]

This is still broken for functions returning structs or non-empty enums. I am working on fixing this.

EDIT: Nevermind, this is intended behavior.

[briansmith]

See also #44119. It seems the unused_results lint is stricter now, including for example on functions returning boolean results like PathBuf::set_extension().