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

[workingjubilee]

your Rust does not compile. did you mean

fn main() {
    for i in 0..100000000u32 {
        let s = i.to_string();
        assert!(s.len() > 0);
    }
}

[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. :)