Slower than `node-glob` when ignoring a directory with `dir/**`

[mrmlnc]

Slower than `node-glob` when ignoring a directory with `dir/**`

[mrmlnc]

test
├── a
│   ├── a.css
│   ├── a.js
│   └── a.txt
└── b
    ├── b.css
    ├── b.js
    └── b.txt

const globby = require('globby');

const entries = globby.sync(['**/*', '!a/**', 'a/*.{js,css}'], { cwd: 'test' });

console.dir(entries, { colors: true });

In this case, globby respects the order of patterns and results will be:

[ 'b/b.css', 'b/b.js', 'b/b.txt', 'a/a.css', 'a/a.js' ]

So in this case we need to read the a directory.

But the fast-glob package does not respect the order of patterns. First, all the negative patterns are applied, and only then the positive patterns. Results will be:

[ 'b/b.css', 'b/b.js', 'b/b.txt' ]

In this case, we can safely ignore the reading of directories that end with the globstar. I file PR #49 for this case.

[mrmlnc]

Hello, @pvdlg,

Thanks for good catch. Yeap, I know about this problem. The current behavior is described in «How to exclude directory from reading?» section.

Unfortunately, I don't remember why I decided to only describe the current behavior. It seems that were borderline cases when the directory still need to read the case even if you specify a pattern that excludes this directory. I will try to find artifacts notes of these conversations.

[mrmlnc]

Landed by [email protected].