perf: pre-allocate arrays and maps

[nik-rev]

No description provided.

[coveralls]

Pull Request Test Coverage Report for Build 18668224799

Details

• 7 of 7 (100.0%) changed or added relevant lines in 4 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 64.846%

---

Totals
Change from base Build 18169995909:	0.0%
Covered Lines:	950
Relevant Lines:	1465

---

💛 - Coveralls

[epage]

How much of a difference does this make in practice?

[nik-rev]

For 1 million TOML keys, it took 8.82 seconds and now it takes 8.25 seconds to call Config::build(). 6.5% increase.

Feel free to close the PR if you think it's not worth it. I needed to understand a region of code in this crate and found a possible, albeit minor, improvement

benchmark

#[test]
#[cfg(feature = "toml")]
fn benchmark() {
    use std::time::Instant;
    use config::{Config, File, FileFormat};

    let config = (0..1_000_000)
        .map(|i| format!("key{i} = {i}\n"))
        .collect::<String>();

    let c = Config::builder().add_source(File::from_str(&config, FileFormat::Toml));
    let before = Instant::now();
    let c = c.build();
    println!("{:?}", before.elapsed());
}

Outputs (with this PR):

8.200018192s
8.270070666s
8.27713242s

Before this PR:

8.779736576s
8.760608855s
8.913664742s

[epage]

I think I'll hold off. If we want to improve this, likely the best way is to deserialize directly to our own type