Update benchmark

[jqnatividad]

It's been three years since the benchmark was run and the time crate recently removed its itoa dependency claiming that stdlib improvements no longer require it.

time-rs/time@26cf0ac

[dtolnay]

Looks about the same to me. #60

[dtolnay]

I guess they just ate the 3× performance regression for time formatting. 🙁
Before time-rs/time@26cf0ac: 28ns, After: 78ns

#![feature(test)]

extern crate test;

use time::format_description;
use time::macros::time;

#[bench]
fn bench(b: &mut test::Bencher) {
    let format = format_description::parse("[hour]:[minute]").unwrap();
    let time = time!(12:59);
    let mut buffer = Vec::with_capacity(40);
    b.iter(|| {
        buffer.clear();
        time.format_into(&mut buffer, &format).unwrap();
        test::black_box(&buffer);
    });
}

[jhpratt]

Hm, that certainly wasn't intentional. I was under the impression that rust-lang/rust#136264 effectively built itoa into std, but that is apparently not the case. My own benchmark using criterion and some finer tuning isn't as bad (it's 93% slower instead of 178% slower), but it's still far more than expected given how I was tracking things.

I'll be rolling back the change, but I'm curious what the difference is between itoa and std that still remains to be uplifted.