Implement size_hint function for SeqVisitor and MapVisitor

[knsd]

So it's possible to preallocate things in Deserialize implementations.

[coveralls]

Coverage increased (+0.02%) to 73.927% when pulling 31e7727 on knsd:master into 2fe3ed9 on 3Hren:master.

[3Hren]

Hi!

Looks good!

This is some kind of new feature of serde I suppose? Can you make a simple benchmark to see performance benefit before and after this PR?

[knsd]

Hi!

Looks like it a quite old feature from serde 0.2.0 :)

Of course, for example for this benchmark:

#![feature(test)]

extern crate test;

extern crate serde;
extern crate rmp_serde;

use std::collections::{HashMap};
use serde::{Serialize, Deserialize};

use test::{Bencher};

#[bench]
fn bench_map_1000(bencher: &mut Bencher) {
    bench_map(bencher, 1000)
}

#[bench]
fn bench_map_5000(bencher: &mut Bencher) {
    bench_map(bencher, 5000)
}

#[bench]
fn bench_map_10000(bencher: &mut Bencher) {
    bench_map(bencher, 10000)
}

fn bench_map(bencher: &mut Bencher, size: usize) {
    let mut map: HashMap<usize, usize> = HashMap::new();
    for item in 0..size {
        map.insert(item, item);
    }

    let mut buf = Vec::new();
    {
        let mut serializer = rmp_serde::Serializer::new(&mut buf);
        map.serialize(&mut serializer).unwrap();
    }
    bencher.iter(|| {
        <HashMap<usize, usize>>::deserialize(&mut rmp_serde::Deserializer::new(&buf[..])).unwrap();
    })
}

Without size hint:

test bench_map_1000  ... bench:     112,692 ns/iter (+/- 3,586)
test bench_map_5000  ... bench:   3,679,561 ns/iter (+/- 775,236)
test bench_map_10000 ... bench:  13,976,963 ns/iter (+/- 1,625,630)

And with size hint:

test bench_map_1000  ... bench:      58,376 ns/iter (+/- 9,696)
test bench_map_5000  ... bench:     297,913 ns/iter (+/- 21,654)
test bench_map_10000 ... bench:     602,640 ns/iter (+/- 119,802)

Should I include this benchmark into the repository?

[3Hren]

Great results!

Should I include this benchmark into the repository?

I think it's not necessary.

Thanks!