Use fold in Iterator::last default implementation

[czipperz]

We already use it in all the other methods. Consistency + potential perf is a pretty nice win!

[rust-highfive]

r? @sfackler

(rust_highfive has picked a reviewer for you, use r? to override)

[czipperz]

cc @scottmcm @cuviper from #61726

[czipperz]

So nth doesn't work because Self: Sized is not necessarily fulfilled. Why does try_fold require Self: Sized? The default implementation doesn't use it and it doesn't take self by value. Strange

[cuviper]

Regarding try_fold -- methods with generic parameters need Self: Sized for trait object safety.

The last() change looks fine to me.

[czipperz]

Mmh. I didn't think about that. I'll change the title.

[scottmcm]

Looks good; thanks for the PR!

@bors r+ rollup

[bors]

📌 Commit 23450c3da83366c67bda42efb827509bba581df5 has been approved by scottmcm

[scottmcm]

@bors r-

Actually, @czipperz, can you squash the commits here please? (Or just force-push to
53ce67c.) Let's not merge in history that immediately reverts itself :)

[timvermeulen]

It shouldn't matter, but you may as well use fold here?

[llogiq]

I just did a quick benchmark and self.fold(None, move |_, x| Some(x)) was around 10% faster, besides being shorter and IMHO more readable.

[czipperz]

I'm surprised it is that much faster wow! Could you share your benchmark code / data?

[llogiq]

#![feature(test)]

extern crate test;
use test::{black_box, Bencher};

#[bench]
fn bench_for_each(b: &mut Bencher) {
    fn last<I: Iterator>(i: I) -> Option<<I as Iterator>::Item> where I: Sized {
        let mut last = None;
        i.for_each(|x| last = Some(x));
        last
    }
    let v = vec![0usize; 1024];
    b.iter(|| { last(black_box(v.iter())) })
}

#[bench]
fn bench_fold(b: &mut Bencher) {
    fn last<I: Iterator>(i: I) -> Option<<I as Iterator>::Item> where I: Sized {
        i.fold(None, move |_, x| Some(x))
    }
    let v = vec![0usize; 1024];
    b.iter(|| { last(black_box(v.iter())) })
}

Oops, I forgot running the benchmark with -C opt_level=3. The for_each version gets zero runtime with that, which kind of invalidates the benchmark, though. More research is needed.

[czipperz]

I do think fold is clearer though

[czipperz]

@scottmcm squashed it up

[scottmcm]

Thanks, @czipperz!

@bors r+ rollup

[bors]

📌 Commit 76a8bc2 has been approved by scottmcm