Uses glidesort for faster sorting

[sts10]

Uses glidesort v0.1.2 for sorting words alphabetically. Though I'm not sure this will make an appreciable performance difference until we're sorting millions of elements?

[sts10]

As an informal benchmark, I ran Tidy with no options on the 4 EFF fandom word lists, to combine (and sort) these four lists into one. I used hyperfine, with settings: 15 warm-ups runs, 50 runs.

Without glidesort:

  Time (mean ± σ):     129.8 ms ±  10.2 ms    [User: 104.3 ms, System: 25.2 ms]
  Range (min … max):   116.9 ms … 158.0 ms    50 runs

With glidesort:

  Time (mean ± σ):     122.3 ms ±   4.7 ms    [User: 98.8 ms, System: 23.3 ms]
  Range (min … max):   115.2 ms … 133.4 ms    50 runs

So it's a little quicker!

But then, as a second benchmark, I ran Tidy on a randomly shuffled 40,000-word list; my idea being in this benchmark, Tidy wouldn't be de-duplicating words, just purely sorting them into alphabetical order.

Without glidesort:

  Time (mean ± σ):     304.1 ms ±  21.6 ms    [User: 263.4 ms, System: 40.2 ms]
  Range (min … max):   279.4 ms … 393.1 ms    150 runs

With glidesort:

  Time (mean ± σ):     326.4 ms ±  16.3 ms    [User: 285.0 ms, System: 40.7 ms]
  Range (min … max):   300.0 ms … 396.0 ms    150 runs

Which is interesting (and a little surprising/frustrating)!

I also ran a third benchmark, giving each version a 40k word list with one single word out of alphabetical order. Without glidesort was about 6 ms quicker.

Not sure whether to merge this at this time...

[sts10]

I've now created a separate Rust project to more thoroughly benchmark sort_by/sort_unstable_by/glidesort::sort_by options: https://github.com/sts10/sort_bench

In those tests, on my machine, glidesort beats the other two options by a bit.

[sts10]

Closing this for now.