perf: Arena based ast

[CPunisher]

How to do it

The process of refactoring is very mechanical and labor intensive:

1. Declare lifetimes for all ast structs and other related structs and traits.
2. Create all ast nodes using bumpalo allocator.
3. Fix lots of compilation errors.

Effectiveness

I have almost finished swc_ecma_ast, swc_ecma_parser and swc_ecma_visit crates. You can run the benchmark with cargo bench -p swc_ecma_parser --bench arena_compare:

	Parser	VisitMut
Swc baseline	12.963ms	424.98us
Swc arena	11.881ms	374.05us
Oxc	4.3084ms	372.51us

Existing Barriers

When I was refactoring the parser, I found the following steps to be very mechanical and cumbersome. I think it will greatly reduce the developer enthusiasm for refactoring codebase and downstream applications.

1. Unavailable: Ast conversion by bridging Into.
2. Unavailable: Pattern match with struct destruction
3. Unavailable: AstNode::default()
4. Unavailable: Ast deserialization.
5. Wrap many ast constructions in Box::new_in(.., self.alloc)

[codspeed-hq]

CodSpeed Performance Report

Merging #9805 will degrade performances by 5.97%

_{Comparing CPunisher:perf/arena (fad24ea) with main (c81be2e)}

Summary

❌ 1 regressions
✅ 193 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	CPunisher:perf/arena	Change
❌	es/visitor/base-perf/boxing_boxed_clone	2.3 µs	2.4 µs	-5.97%