fix: canonicalize the path of TemporaryFs

[siketyan]

Summary

Tests added in #5200 are failing on macOS. I found that the directory path of the TemporaryFs used in tests is not canonicalized, which leads path inconsitency. On macOS, temp_dir() returns a path under /var by default, which is a symlink to /private/var. oxc_resolver follows symlinks to resolve a module, so the resolved path will be under /private/var.

This pull request adds canonicalization to TemporaryFs::new to avoid this inconsitency.

Test Plan

Unfortunately, this PR will NOT make the failing tests green. At least the current failing position will be passed, but there is another problem failing the cases. 😢

[siketyan]

I wanted to do std::fs::canonicalize for all platforms here, but it will change the path to \\?\C:\... on Windows, which is not properly handled in this project. 😢

[arendjr]

I wonder if we should make the watcher smart enough that any sources outside a project will be automatically watched in destinations pointed to by symlinks. It might be a more cross-platform approach that could help users too.

Not sure yet how complex that would get…

[siketyan]

@arendjr The problem here is that the workspace is opened in a symlinked path. The diagnostics should appear before the watcher detects a change, but it's not appeared now as oxc_resolver resolves the module to an actual path (symlink target). Workspace watcher should also follow symlinks, but it's not yet a problem in this case.

[arendjr]

Workspace watcher should also follow symlinks, but it's not yet a problem in this case.

I think this requires some delicate interplay between the scanner and its watcher. When the watcher receives events, it would only be for the destination paths, so it cannot traverse backwards to see which symlinks pointed there. But the scanner should be able to catch when it discovers symlinks and tell the watcher to watch those destinations if they are outside the project folder. At least, that's how I think it should work.

For now I'll merge this, because it's a step towards getting the macOS tests green. I'll try to get the remainder of the tests green as well. I've created #5271 for improving the symlink handling.

[arendjr]

Huh, I just pulled after merging, and it did make the tests green locally. Watching the CI now 👀

[arendjr]

Oh, I see. They're sometimes green, but indeed unreliable now.