Avoid compiling ignore patterns twice

[webpro]

The ignore patterns were compiled both via picomatch's ignore option and as a standalone matcher. This PR makes it compile once, and apply manually in filter only.

Coming from webpro-nl/knip#1462 (comment). The change in this PR would make globbing go from >700ms to <320ms in that particular setup, because there are lots of .gitignore files resulting in lots of ignore patterns passed in to tinyglobby#globSync

Here's a script to benchmark the difference:

import { globSync } from './dist/index.mjs';
const ignore = [
  '**/node_modules/**',
  '**/.git/**',
  '**/dist/**',
  '**/build/**',
  '**/coverage/**',
  ...Array.from({ length: 35 }, (_, i) => `**/ignored${i}/**`)
];
const patterns = ['**/*.ts', '**/*.js'];
const cwd = process.cwd();
globSync(patterns, { cwd, ignore }); // Warm up
const iterations = 500;
const start = performance.now();
for (let i = 0; i < iterations; i++) {
  globSync(patterns, { cwd, ignore, expandDirectories: false });
}
const elapsed = performance.now() - start;
console.log(`Total: ${elapsed.toFixed(0)}ms`);
console.log(`Avg: ${(elapsed / iterations).toFixed(2)}ms`);

This should show >40% improvement in time spent compared with main.

Let me know if you need anything else or how I can help with improve/benchmark this. Figured tests passing is enough for general QA?

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/tinyglobby@190

commit: 4e8031d

[SuperchupuDev]

i think i didn't do that initially because of some edge cases around the behavior of the negation of globstars? i'll test the pr soon (hopefully). but it's a good sign that the tests are passing 👍

[wojtekmaj]

@SuperchupuDev Is there anything I could help you with to land this sooner? I'm super excited for tinyglobby to get faster and consequently knip replacing fast-glob with it, shaving off 15 deps. If you could recall the edge cases, maybe I could write some extra tests for example?

[SuperchupuDev]

hi!! i've just finished some important university exams and have enough free time now to look at tinyglobby development, sorry for the wait. i'd say don't worry about it, i think i can test it myself in a day or two. the edge case i'm looking for isn't complicated i just can't remember right now exactly what it was

[SuperchupuDev]

okay i tested a bunch and couldn't find any pattern that produces a different result than 0.2.15, so i'm merging this. will probably test some more later just in case. thank you!!