Integer to_string is slow

[jh05013]

On my machine, the following code runs in 3 seconds in stable version, release build:

// (EDIT: made it compile)
fn main() {
    for i in 0..100000000u32 {
        let s = i.to_string();
        assert!(s.len() > 0);
    }
}

whereas the C++ counterpart runs in 1.2 seconds with -O2:

#include <string>
#include <cassert>
int main() {
    for(unsigned int i=0; i<100000000; i++){
        std::string s = std::to_string(i);
        assert(s.size() > 0);
    }
}

I've found that most of the time loss comes from passing &str through a formatter instead of directly memcpying; replacing to_string with _fmt with .to_owned() at the end speeds it up to around 1.6s.

[clubby789]

FWIW, my understanding is that formatting goes through dynamic dispatch, as a deliberate choice to reduce the binary size bloat from inlining all formatting

[jfrimmel]

You can try the itoa-crate, which removes some overhead mentioned in the first comment.

[GuillaumeGomez]

To be noted that there is work in progress to improve this situation. Already two PRs were merged:

• Reduce integer Display implementation size #133247
• Small optimization for integers Display implementation #128204

Hopefully more will follow soon. :)

[tgross35]

@jh05013 I'd be curious to see what your results are rerunning on a recent nightly. Rough numbers for me, running the top post's example code on rustc 1.89.0-nightly (8da623945 2025-06-13) on an aarch64 laptop with -Copt-level=3 / -O3:

$ hyperfine ./rs-bin ./cpp-bin
Benchmark 1: ./rs-bin
  Time (mean ± σ):     237.7 ms ±   1.8 ms    [User: 235.8 ms, System: 1.3 ms]
  Range (min … max):   236.1 ms … 241.4 ms    12 runs

Benchmark 2: ./cpp-bin
  Time (mean ± σ):     542.8 ms ±  12.6 ms    [User: 540.0 ms, System: 2.1 ms]
  Range (min … max):   534.5 ms … 578.0 ms    10 runs