perf(parser): eat whitespace after line break

[overlookmotel]

Uses the byte_search! macro introduced in #2352 to consume whitespace after a line break.

[overlookmotel]

Current dependencies on/for this PR:

• main
  • PR refactor(parser): all pointer manipulation through SourcePosition #2350
    • PR refactor(parser): macro for ASCII identifier byte handlers #2351
      • PR perf(parser): lex identifiers as bytes not chars #2352
        • PR perf(parser): eat whitespace after line break #2353 👈
          • PR perf(parser): skip single space between tokens branchlessly #2354

This stack of pull requests is managed by Graphite.

[codspeed-hq]

CodSpeed Performance Report

Merging #2353 will improve performances by 27.35%

_{Comparing 02-08-perf_parser_eat_whitespace_after_line_break (d89b376) with 02-08-perf_parser_lex_identifiers_as_bytes_not_chars (070d3c4)}

Summary

⚡ 7 improvements
✅ 20 untouched benchmarks

Benchmarks breakdown

	Benchmark	02-08-perf_parser_lex_identifiers_as_bytes_not_chars	02-08-perf_parser_eat_whitespace_after_line_break	Change
⚡	parser[checker.ts]	369.2 ms	349.7 ms	+5.56%
⚡	lexer[RadixUIAdoptionSection.jsx]	148.7 µs	129.9 µs	+14.48%
⚡	minifier[typescript.js]	1.6 s	1.6 s	+4.08%
⚡	lexer[checker.ts]	95.6 ms	75.1 ms	+27.35%
⚡	lexer[cal.com.tsx]	43 ms	36.9 ms	+16.58%
⚡	lexer[pdf.mjs]	22.5 ms	20.3 ms	+10.92%
⚡	lexer[antd.js]	157.7 ms	146.1 ms	+7.94%

[Boshen]

I LOVE all the performance improvements.

One nitpick is the amount of macros we are introducing, it may impact compilation speed and hinder other contributors to navigate the code.

It's troublesome but I think we can expand all the code once we are done ... just like how I did it with the visitors https://github.com/oxc-project/oxc/blob/main/crates/oxc_ast/src/visit.rs ... I wrote it manually 😅

[overlookmotel]

It's been fun! By our joint efforts, over the past month the checker.ts parser benchmark has gone from 533ms to 350ms.

Maybe you can update the website where it says "2x faster than SWC" to "3x faster"!

It's funny, originally I had it all written out by hand, but it was so repetitive I found it hard to follow, so that's why I brought in the macro.

Have you noticed a difference in compilation times? I haven't noticed one, but OXC compilation falls into the "fast enough" category for me, especially after working on SWC which was so hideously slow in that department.

[Boshen]

Maybe you can update the website where it says "2x faster than SWC" to "3x faster"!

https://github.com/oxc-project/bench-javascript-parser-written-in-rust will automatically pick up the latest release via renovate bot, and I'll update the numbers from that benchmark.

[Boshen]

Have you noticed a difference in compilation times?

We'll need to look at cargo build --timings

[overlookmotel]

Would you be willing to cut a release at this point? I think the speed boost in parser justifies it, and I'm dying to see the updated comparison to SWC.