perf: optimize gobbleParens, specificity, and regex pattern caching

[TrevorBurnham]

Summary

This CR adds a benchmark suite to parsel-js and applies three optimizations.

For context, I'm trying to get happy-dom to adopt parsel-js instead of its own regex-based CSS selector parser, which has known bugs that parsel-js fixes: capricorn86/happy-dom#2010 However, the lead maintainer is concerned about the performance hit. I'd love to see a world where JavaScript has CSS selector parser that's both fast and accurate! ⚡🎯

Changes

1. gobbleParens — slice instead of string concatenation

The original built a result string character-by-character (result += char), which creates a new string allocation on every iteration — O(n²) for long parenthesized expressions. Replaced with index tracking and a single text.slice() call at the end.

2. specificity — remove duplicate computation for selector lists

When computing specificity for a list selector (e.g. div, #foo), specificity() was called twice per child selector:

// Before — calls specificity(ast) twice
const sp = specificity(ast);
base = Math.max(base, ...specificity(ast));

// After — reuses the result
const sp = specificity(ast);
base = Math.max(base, ...sp);

3. getArgumentPatternByType — cache compiled RegExp

Each call for pseudo-class or pseudo-element types was constructing a new RegExp via string replacement. Since the token grammar is stable between calls, the compiled patterns are now cached in a Map.

Benchmark results

Measured with bench.mjs (included), which exercises tokenize, parse, stringify, specificity, walk, and end-to-end across 23 selectors of varying complexity. 5000 iterations per benchmark, 500 warmup, median reported.

Benchmark	Before	After	Change
tokenize/ALL	165 µs	154 µs	-7%
parse/ALL	232 µs	222 µs	-4%
specificity/ALL	270 µs	239 µs	-11%
specificity-ast/ALL	31.5 µs	16.3 µs	-48%
e2e/ALL	268 µs	239 µs	-11%

Optimizations evaluated but not included

• nestTokens single-pass comma detection: Merging the find() + iteration into one pass showed no measurable improvement — token arrays are small enough that the extra scan is negligible.
• tokenizeBy splice avoidance: Replacing in-place splice with a two-array swap showed ~3% on tokenize but added complexity for minimal gain.

Testing

All existing tests pass.

[netlify]

✅ Deploy Preview for parsel ready!

Name	Link
🔨 Latest commit	8cd637f
🔍 Latest deploy log	https://app.netlify.com/projects/parsel/deploys/6989efb5d0711a0008bd9be7
😎 Deploy Preview	https://deploy-preview-85--parsel.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.