watchman integration for faster incremental runs

[9999years]

Is your feature request related to a problem? Please describe.

Formatting is slow in large repos, even with caching enabled (4 seconds the first time, 0.5s on subsequent runs with no changes). It would be vastly faster if treefmt could narrow down which files had changed.

Describe the solution you'd like

I already have a watchman server running for tools like jj and buck2. treefmt could simply connect to that server and query it for a list of files which changed since the last time treefmt ran. Then, it could look at only those files to vastly cut down on the required work.

Additional context

Unlike #509, I am not proposing a new mode of operation in which treefmt continuously watches files and re-runs its formatters. Instead, I would like it to only query the watchman server once in order to ask it which files have changed since the last run.

[brianmcgee]

I like it.

[jfly]

As with #509, I'm opposed to this. IMO, it's something your text editor/IDE should be doing instead.

[9999years]

@jfly I'm not proposing that we add any sort of file watching to treefmt. I'm simply proposing that we (optionally) use watchman to determine which files have changed since the previous run. This is an enhancement to the caching mechanism that reduces duplicate work for users who are already running watchman to track file changes for tools like jj or buck2.

Does that make sense?

[jfly]

Thanks for the explanation! Please disregard my comment, I incorrectly assumed this was about automatically invoking treefmt when files change.

It looks like @brianmcgee is supportive, so I'll wait to see the PR =)