AST transfer WIP

[overlookmotel]

Very rough POC for #2409. Nowhere near ready to merge.

Mostly works. JS AST for checker.ts via "raw" transfer matches AST produced via current JSON API exactly. Some errors in AST for pdf.mjs and antd.js due to some of reasons listed below.

Edit 8/3/24: Output for pdf.mjs and antd.js is now correct.

Current deficiencies:

• Incorrect output for files containing any non-ASCII characters.
• Does not handle strings/identifiers/regexps containing \ escapes.
• Statement and a few other types increase in size by 8 bytes (likely the cause of the performance regression seen in benchmarks).
• Requires size of AST to be known in advance.
• Hacky method of creating allocations.
• Memory handling may be unsound!

[overlookmotel]

• AST transfer WIP #2457 👈
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @overlookmotel and the rest of your teammates on Graphite

[codspeed-hq]

CodSpeed Performance Report

Merging #2457 will not alter performance

_{Comparing 02-20-AST_transfer_WIP (79d2808) with main (88ded0c)}

Summary

✅ 30 untouched benchmarks

🆕 5 new benchmarks

Benchmarks breakdown

	Benchmark	main	02-20-AST_transfer_WIP	Change
🆕	parser_napi[RadixUIAdoptionSection.jsx]	N/A	44.6 µs	N/A
🆕	parser_napi[antd.js]	N/A	146 ms	N/A
🆕	parser_napi[cal.com.tsx]	N/A	24.5 ms	N/A
🆕	parser_napi[checker.ts]	N/A	52.5 ms	N/A
🆕	parser_napi[pdf.mjs]	N/A	15.4 ms	N/A

[andymac4182]

Does this mean we could use oxc_parser for eslint as well as prettier?

[overlookmotel]

Maybe... Just to say this is very early stage work, the concept is not proven, and there is a long way to go yet. Some significant things would need to change within OXC to convert this from an experiment to a solid, dependable reality.

Personally, I do believe there is potential in this approach, but I think it's premature at this stage to talk a great deal about applications.

Sorry if that sounds negative. I'm just trying to give a realistic answer.

[andymac4182]

That's okay 😀 that's a perfectly reasonably answer

[overlookmotel]

I don't believe the Codspeed benchmarks are accurate. Running benchmarks locally, I'm seeing more like x3 - x5 speed-up. Will investigate.

[overlookmotel]

Yes, benchmarks on CodSpeed are completely wrong. Have raised an issue CodSpeedHQ/action#96