Inline benchmark in rust

11/18/22

24/4/1444

---

Turns out you can do inline benchmarks in rust with a not that bad user experience.

cargo new bench-inline

In main.rs:

#![feature(test)]

extern crate test;

fn main() {
    
}

pub fn add_two(a: i32) -> i32 {
    a + 2
}

#[cfg(test)]
mod tests {
    use super::*;
    use test::Bencher;

    #[test]
    fn it_works() {
        assert_eq!(4, add_two(2));
    }

    #[bench]
    fn bench_add_two(b: &mut Bencher) {
        b.iter(|| add_two(2));
    }
}

cargo +nightly bench

Ok so far so good, we just followed cargo-bench, we have our inline benchmarks but the problem is this: we usually want to just use stable and only use cargo +nightly for bench also rust analyzer gets confused with this setup.

Turns out there is a way!

1- Add a new feature lets call it nightly

cargo.toml

[features]
nightly = []

2- change our main.rs to:

#![cfg_attr(feature = "nightly", feature(test))]

fn main() {
    
}

pub fn add_two(a: i32) -> i32 {
    a + 2
}

#[cfg(test)]
mod tests {
    #[cfg(feature = "nightly")]
    extern crate test;

    use super::*;

    #[test]
    fn it_works() {
        assert_eq!(4, add_two(2));
    }

    #[cfg(feature = "nightly")]
    #[bench]
    fn bench_add_two(b: &mut test::Bencher) {
        b.iter(|| add_two(2));
    }
}

That's it! Now rust analyzer is not confused, you can run tests with cargo test and benchmarks with cargo +nightly bench --features nightly

And for a real word example I used this idea to benchmark a PR change in irust and it was really useful irust-syntect-pr