Adds a patchfile for fsevents

[arcanis]

What's the problem this PR addresses?

FsEvents isn't aware of the virtual filesystem ($$virtual, more details at the end of this post), and that breaks Webpack when requiring files from a workspace that lists peer dependencies.

How did you fix it?

As a short-term fix I've implemented the fix in fsevents via our new patch: protocol. I wonder if we could convince the fsevents maintainers to merge the change upstream? (cc @paulmillr / @pipobscure)

What is this $$virtual thing?

First read that. Now, as the article says:

Virtual packages [...] are now implemented through a virtual filesystem layer. This circumvents the need to create hundreds of confusing symlinks, improving compatibility with Windows and preventing issues that would arise with third-party tools calling realpath

The way we do this by injecting a fs layer that applies the following algorithm:

• Find occurances of /$$virtual/<whatever>/<number>/
• Replace them by '../'.repeat(number)

So for example, /foo/$$virtual/abcdef/0/hello becomes /foo/hello, and /foo/$$virtual/abcdef/1/hello becomes /hello. This technic allows us to generate an infinity of paths for a given package.

[paulmillr]

Could you get the PR to our fsevents repo? We will discuss it there!

[pipobscure]

Though I have to say that fsevents seems like the worst possible layer to do this.

[arcanis]

Damn, for some reasons fsevents just doesn't seem to work on our GitHub Actions, even though you actually use it in your own tests 🙁

[arcanis]

Oh interesting - FSEvents 2.x works on GitHub Actions, but not 1.x 🤔

[arcanis]

Errata: I had a typo in the variable name that was causing undefined to be sent to FsEvents, triggering a segfault. For some reason it still worked on my computer (the segfault's ways are mysterious indeed), but that's why it failed once on CI. Now everything's passing!

[arcanis]

Merging for now, will open a PR against FSEvents later.