Improve parsing performance

[unlessgames]

Fixes #23

I was too lazy to benchmark and measure precisely and the pain points were fairly obvious after some quick testing. While the parser got a bit more ugly, but it's also way faster.

Added a stress test line here, this was hanging for a long-time before, now it doesn't.

I'm a bit bummed by the fact that utilizing the pest grammar to do the heavy-lifting leads to this much performance penalty. Maybe in the future I'll rewrite the entire parser using chumsky or combine...

[unlessgames]

I removed some is_err from bindings/cycle.rs that were checking empty cycles, an empty cycle will just lead to a single rest, I think this is fair as we already parse [] as a rest, which I think is better than crashing for that.

[emuell]

That's a huge improvement. Unfortunately the #[bench] feature still is unstable, so I've created some quick and dirty benches with https://github.com/bluss/bencher

master

running 2 tests
test nested_groups ... bench: 2,577,790 ns/iter (+/- 205,782)
test parser ... bench: 46,256,770 ns/iter (+/- 5,336,864)

perf/cycle-nest

running 2 tests
test nested_groups ... bench: 21,611 ns/iter (+/- 1,390)
test parser ... bench: 120,066 ns/iter (+/- 16,314)

See See https://github.com/emuell/afseq/tree/perf/benches

Would not hurt to have a few other benches in there, as all the stuff is very time critical.
Feel free to change or add some tests in that branch, if you want.

The benches can be run with cargo bench

---

I'm a bit bummed by the fact that utilizing the pest grammar to do the heavy-lifting leads to this much performance penalty. Maybe in the future I'll rewrite the entire parser using chumsky or combine...

It's also very surprising to me. As long as there are workarounds like this, this is manageable. The other Rust PEG parsers that I reviewed back then are definitely more difficult to use. I guess you can't have everything here...

[unlessgames]

This benchmarking looks neat, it's good to see how much speed we gained. It could be nice to have something like it in a github action that would run a comparison benchmark for master and the branch to be merged and log it along with test results.

The other Rust PEG parsers that I reviewed back then are definitely more difficult to use.

Chumsky seems alright (getting to v1.0 now). It's not as clean as the grammar file, but the parser could skip the entire intermediate "rule pairs" form and just generate the Step struct right away, this reduces complexity in a way (at the cost of readability) and probably gains even more speed. But for now, this is clearly good enough.

[emuell]

But for now, this is clearly good enough.

Yup.