Performance & Benchmarks

[WhiteBlackGoose]

Problem

Yoakke, unfortunately, turns out to have significant performance issues (compared to Antlr, for example). They are:

1. Inefficient ParseResult and related types. Making it efficient may boost it to 10x, but it's still not enough
2. A lot of nested interfaces. Every call on token stream is 5+ nested calls.

However, the parsing and lexing algos look to be fine. At least, I didn't get past 2., so cannot say for sure.

[WhiteBlackGoose]

Initial results:

Method	Mean	Error	StdDev
Parse	49.920 us	1.1544 us	3.2936 us
Lex	3.713 us	0.0708 us	0.1494 us

[WhiteBlackGoose]

BDN otherwise fails to run

[WhiteBlackGoose]

Parser - [2022-06-14 20-26-04].zip

[WhiteBlackGoose]

Replaced ParseError's binary OR with return first:

Method	Mean	Error	StdDev
Parse	15.655 us	0.3817 us	1.0641 us
Lex	2.825 us	0.0562 us	0.0998 us

[WhiteBlackGoose]

Removed creating a dictionary on ParseError

Method	Mean	Error	StdDev
Parse	7.425 us	0.1222 us	0.1143 us
Lex	2.752 us	0.0516 us	0.0552 us

[WhiteBlackGoose]

Parser - [2022-06-14 23-12-46].zip

[WhiteBlackGoose]

Warning: this is wrong result, because it dysfunctions

Replaced public static ParseResult<T> operator |(ParseResult<T> first, ParseResult<T> second) with return first.IsOk ? first.Ok : first.Error;

Method	Mean	Error	StdDev
Parse	919.3 ns	18.34 ns	25.71 ns
Lex	3,779.2 ns	73.81 ns	121.27 ns

But it looks wrong, so gotta make it functionally correct

[WhiteBlackGoose]

For AngouriMath these optimizations have these impacts

Method	Mean	Error	StdDev	Median
ParseEasy OOB	11.61 ms	0.440 ms	1.298 ms	11.10 ms
ParseEasy Optim	1.143 ms	0.0510 ms	0.1371 ms	1.131 ms
ParseEasy Antlr	0.039 ms
ParseHard OOB	155.05 ms	9.570 ms	28.217 ms	146.27 ms
ParseHard Optim	69.306 ms	2.8647 ms	7.8420 ms	66.660 ms
ParseHard Antlr	4.4 ms

[WhiteBlackGoose]

Warning: no merge! The Result type is intentionally broken!

[kant2002]

@WhiteBlackGoose i'm new maintainer of the Yoakke, so if you still (after such long time) want to continue exploration of Yoakke performance, I'm here for you.

[kant2002]

Dear @WhiteBlackGoose thanks for pushing question of the Yoakke performance. I create separate benchmark project in #173 so if you want, we can continue from that place. You are always welcome in this project. I will close this PR, but I think main findings are rightly point on the places which should be improved.

[WhiteBlackGoose]

hi @kant2002 , sorry, missed your message, congrats on becoming a new maintainer of the project! Yeah I think it's currently not on my radar to contribute here.