Using globs for `directory` is slow

[freshollie]

The recommended config for multiple directories is very slow (./packages/**/tsconfig.json)

This is probably due to the library walking the node_modules directories? Would it be possible to ignore the node_modules folders?

For now, I recommend explicit levels in the directories glob, which is much faster.

"settings": {
    "import/resolver": {
    "ts": {
      "directory": "./*/*/tsconfig.json" // Any package, but don't go deeper than the package root
    }
  }

[JounQin]

@freshollie Sorry I didn't notice this issue previous, I'll take a look ASAP.

[JounQin]

@teoxoy Would you like to help with this?

[teoxoy]

./packages/**/tsconfig.json shouldn't look in the node_modules folder.

Having said that, the example glob can be optimized like this: ./packages/*/tsconfig.json - this way the star (*) will only match one layer deep instead of the globstar (**) matching n layers

@freshollie can you test the proposed glob and see if it's faster?

If it is, we should update the readme.

[JounQin]

@teoxoy Should directory matched in node_modules be ignored always? What means we can use glob.sync(pattern, { ignore: 'node_modules' }).

[teoxoy]

ignore should only be used when the pattern would include a path that then ignore will remove.

Using it in our case, will make things slower and with no real benefits since ./packages/*/tsconfig.json will not match anything inside node_modules.

As I said above, the issue was that I used a globstar instead of a star for the example.

[JounQin]

@teoxoy Thanks for clarifying. Maybe another stupid question, should we replace glob with tiny-glob according to its benchmark?

[JounQin]

@freshollie v0.4.0 has been released.

[freshollie]

Hey @JounQin, thanks for this, updating now.

@teoxoy, yes, using the single star glob was always faster. I had already discovered this before posting the issue, so I was more trying to inform that using ** would make eslint very slow and that having it in the readme was a bad idea.

[JounQin]

@freshollie Yep, we've changed ** to * in README, thanks for reporting any way.