Optimize memory allocations

[unixzii]

Remove the redundant pattern joining, which can reduce memory allocations during the iteration.

[unixzii]

Hi @tgross35, sorry for pinging. This PR has been open for a while and was out of sync. I've resolved the conflicts and simplified the changes. Would you mind giving it a review when you have time?

[tgross35]

Yeah, this repo isn't really actively maintained - and I don't know it any better than anybody else. But this is a small and straightforward change, so I think it can merge with a few small tweaks.

[tgross35]

Bumping MSRV in #172 so that part passes

[unixzii]

Bumping MSRV in #172 so that part passes

Great, so let's use the new syntax then.

[tgross35]

Thank you!

Post any benchmarks if you happen to have them (just curious if the effect is actually noticeable - the change is good either way)

[unixzii]

Thank you!

Post any benchmarks if you happen to have them (just curious if the effect is actually noticeable - the change is good either way)

Sure, I ran a simple benchmark with the following code:

#[bench]
fn bench_long_pattern(b: &mut rust_test::Bencher) {
    b.iter(|| {
        glob("/*/some_very_very_long_path").unwrap().next();
    })
}

The result shows there is roughly an 8.8% speedup. In addition, the variance dropped significantly, indicating much more consistent performance (malloc does have unpredictable overhead I guess).

before:

test test::bench_long_pattern ... bench:  23,131,008.40 ns/iter (+/- 8,300,816.31)

after:

test test::bench_long_pattern ... bench:  21,097,649.90 ns/iter (+/- 2,002,235.87)