perf(es/minifier): Speed up merge_sequences_in_exprs by caching computation

[CPunisher]

Problem

In https://github.com/chenjiahan/rspack-swc-minify-test:
Rspack + SWC minify: 15.89 s
Rspack + esbuild minify: 1.74 s

This is caused by merge_sequences_in_exprs which tries to merge every pair of exprs with time complexity of O(n^2). In the given example, a vec of 5003 exprs (react component declarations and other declarations) are passed to this function.

From the profiling result below we can see visit_children_with takes a lot of times, which is called by IdentUsageFinder and idents_used_by to check whether an ident is used in an ast node. However, in the O(n^2) loops, this part of the result is heavily recalculated.

Example:

// input.js
export function source() {
    let c = 0, a = 1;
    c += a, a += 5;
    let b = c;
    console.log(a, b, c);
}

// esbuild/terser output
export function source(){let o=0,e=1;o+=e,e+=5,console.log(e,o,o)}

// swc output
export function source(){console.log(6,1,1)}

Solution

To be honest, I don't come up with a better algorithm to reduce time complexity or pruning. But caching part of the computation does work very well in this bad case. Anyway, I'm proposing this pr first.

Result

Before: 10128ms

After: 1013ms(~10x)

[changeset-bot]

🦋 Changeset detected

Latest commit: 75045a4

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[kdy1]

CI failed, and I verified locally that CI failure does not happen on the main branch.

I love the idea behind this PR. I think it's fine to simply update test cases of the swc crate in this case, considering the performance difference and the amount of difference in outputs. Can you update or fix them? You can do either of them.

[codspeed-hq]

CodSpeed Performance Report

Merging #9843 will degrade performances by 3.33%

_{Comparing CPunisher:perf/merge_sequences_in_exprs (9ddccb8) with main (66bf0e5)}

Summary

❌ 1 regressions
✅ 205 untouched benchmarks

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

Benchmarks breakdown

	Benchmark	main	CPunisher:perf/merge_sequences_in_exprs	Change
❌	common/allocator/alloc/cached-no-scope/1000000	108.6 ms	112.4 ms	-3.33%

[kdy1]

Thank you!