Possible Linux(glibc) memory usage improvement

So I was watching this video and wondered if Rust-Analyzer could take this and use it to improve it's memory usage.

It tells GLIBC to not pool large allocations in arenas and can prevent a pile-up of wasted memory space.

I did take the code from cosmic-bg and just added the call to mallopt(M_MMAP_THRESHOLD, 65536); and saw a memory usage decrease. Only like a 100mb however, I think this might want to be explored.

The Cloudflare article mentions calling malloc_trim from time to time as this goes through and removes unused arena space. Which I think might be a good idea to call after project loads?

Code

#[cfg(target_env = "gnu")]
mod malloc {
    use std::os::raw::c_int;
    const M_MMAP_THRESHOLD: c_int = -3;

    extern "C" {
        fn mallopt(param: c_int, value: c_int) -> c_int;
    }
	/// tells GLIBC to not pool large allocations in arenas  
    pub(crate) fn limit_mmap_threshold() {
        unsafe {
            mallopt(M_MMAP_THRESHOLD, 65536);
        }
    }
}
fn main() {
	#[cfg(target_env = "gnu")]
    malloc::limit_mmap_threshold();
    // The rest of the main function
}

The reason this is an issue and not a pull request is I think maybe more experimentation might be needed first and also want to give the developers with more knowledge a chance to give any thoughts on this subject.

Links

• Cloudlfare Article
• Cosmic Desktop
• Cosmic Desktop Dev Post

Start from rust-analyzer's main entry point and review how Linux glibc allocations behave during project loads. Experiment with the proposed mallopt and malloc_trim approaches, measuring memory use and checking their platform and safety implications; done would require evidence of a reliable improvement and agreement on the approach.