feat: replace sane with chokidar

[SimenB]

Summary

sane pulls in old version of micromatch. We're also already vendoring their watchman watcher (#5387) and the fsevents watcher they removed (#8258). Replacing with chokidar allows us to drop that last vendoring as chokidar abstracts over it.

We might need to create a chokidar abstraction here, I just did the minimal to make it work.

This is very untested

Test plan

Dunno. It works for the simple cases I played with locally, but it needs real usage.

[cpojer]

Surprising but .. they expose almost the same API? :O

[SimenB]

I dunno if the ignore pattern works, should probably verify

[SimenB]

this is the part that makes me think creating some abstraction makes sense. There might be other things we want to tweak as well

[cpojer]

Goodbye sane, and goodbye @cnakazawa/watch. Nice work, I'm excited about this change as it took years :D

[SimenB]

Hmm, #5387 imports some stuff from sane

We should vendor it, I guess

[arhelmus]

Hey @SimenB, thanks for your amazing work, its actually should also improve DX for Windows users of Metro. What is currently blocking you from merging this PR?

[SimenB]

@arhelmus no real blocker, just time and energy to work on this. It'll require extensive testing, so I haven't really found the time to do it.

This is the part of the code base I've touched the least (along with watch mode, which also mostly is haste-map), so I'm a bit hesitant to merge it. I can rebase it to resolve conflicts, tho

[arhelmus]

@SimenB I would love to help you to test it on Mac and Windows, but I missing lots of context. Can you provide some test plan which I can follow. The best I see now is: create new file/rename it/upper lower case it/delete and see how watchers will react.

[SimenB]

Yeah, that's essentially it - make sure you're disabling watchman (by either not having it installed or by passing --no-watchman).

Note that the linting errors now (#10048 (comment)) need to be solved first. We don't really have many tests for watch mode on CI which is why it's passing even though those imports does not resolve presumably).

[mikemoate]

Given amasad/sane#159 (upstream dependency execa has critical vuln) and that sane is basically unmaintained, this PR would be really helpful!

[SimenB]

Feel free to open PRs against my branch (or opening a new PR based on this branch) fixing the type error. This will need to be tested thoroughly, but once we have a green PR I can make alpha releases

[SimenB]

Testing this, it's horribly slow during first run - about 50x regression over what's currently in master. I think this is due to it reading (or at least stating) all files it finds during initial crawl instead of just registering a watcher.

In addition, on change it emits change events for node_modules - I was unable to figure out how or where. In monorepos any local files are usually symlinked in node_modules, meaning they all triggered a change event (even worse, it happened staggered. So changing one file which ended up regenerating a bunch of .d.ts changes made the tests re-run 10-12 times on my machine).

---

So instead of spending time trying to figure out how to make chokidar do what I want, I think it's better to just vendor the last watcher from sane (the node watcher - we already vendor both the watchman watcher and the fsevents one).

#10919

[github-actions]

This pull request has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.
Please note this issue tracker is not a help forum. We recommend using StackOverflow or our discord channel for questions.