Inconsistent dead code warnings?

[tad-lispy]

I'm curious to understand the rules for cargo check and rust analyzer to report unused functions. It seems inconsistent to me, but probably I'm missing something. Here's what I see.

While learning Rust I copied some modules from a previous project to a new one and indeed some utility functions are not used in the new project. Here are two examples:

The Region::new function is never used except for unit tests, and I see no complaints about it being unused.

• Implementation: src/region.rs · 5b9868f9bc32163a7b4646f5c346429565dc9677 · Tad Lispy / Bevy Lyon Playground · GitLab

• Use in a test e.g. on lines 165-170

At the same time, the TouchPoints::get is also never used except for unit tests, but tools are complaining about it:

• Implementation: src/touch.rs · 5b9868f9bc32163a7b4646f5c346429565dc9677 · Tad Lispy / Bevy Lyon Playground · GitLab

• Use in a test e.g. on line 154

Here's complete output from cargo check:

$ cargo check
    Checking bevy-lyon-playground v0.1.0 (/home/tad/Projects/bevy-lyon-playground)
warning: associated function `get` is never used
  --> src/touch.rs:19:12
   |
19 |     pub fn get(&self, identifier: i64) -> Option<&TouchPoint> {
   |            ^^^
   |
   = note: `#[warn(dead_code)]` on by default

warning: `bevy-lyon-playground` (bin "bevy-lyon-playground") generated 1 warning
    Finished dev [unoptimized + debuginfo] target(s) in 0.31s

The two functions are both public and implemented for a public struct. What makes cargo complain about one of them but not the other?

[steffahn]

Reduced the example to something like

mod foo {
    pub struct Region; // no warning

    impl Region {
        pub fn new() -> Self { // no warning
            Self
        }
    }

    impl From<Region> for () {
        fn from(_: Region) {
            ()
        }
    }
}

mod bar {
    pub struct Region; // struct `Region` is never constructed

    impl Region {
        pub fn new() -> Self { // associated function `new` is never used
            Self
        }
    }
}

Rust Playground

[tad-lispy]

Indeed, adding the following to my code eliminates the warning.

impl From<TouchPoints> for () {
    fn from(_: TouchPoints) -> Self {
        ()
    }
}

That is weird, right?

[steffahn]

Hmm… so maybe, the compiler anticipates settings that don’t quite apply in this concrete situation, but in principle might be (conservatively) correctly analyzed to avoid false warnings… settings such as… imagine new was a method, i.e. with a self type, and imagine From<T> was some other trait with a fn blah(self) -> T method and no default implementations, and this was the only From<T> implementation for ().

The code could arguably call ().blah().new() from outside the crate, relying on type inference to choose Region for T in the From<T> implementation to be chosen (as it’s the only one in existence) and hence the function new was not entirely unused.

---

Some illustrative code

pub trait From<T> {
    fn from(self) -> T;
}

// this could be in a different crate, since everything mentioned is `pub`
fn usage_demo() {
    ().from().new(); // calls `new`
}

mod foo {
    use super::From;

    pub struct Region; // no warning

    impl Region {
        pub fn new(&self) -> Self { // no warning
            Self
        }
    }

    impl From<Region> for () {
        fn from(self) -> Region {
            Region
        }
    }
}

[steffahn]

I’m not sure if this is or isn’t a “best practice”… but since this seems to be a binary crate, to get more reliable unused warnings, maybe simply change all usage of pub into pub(crate) instead?