Allow watching of arbitrary files for nix flake updates

[polygon]

For more complex flake projects that branch out into other nix-files, it is currently not possible to trigger a rebuild of the environment when one of these files changes. Even if they are added to the direnv watches using watch_file, use_flake will check only some pre-determined files and otherwise will re-use the cached environment resulting in an out-of-sync environment.

This is a first suggestion on how an approach to watching multiple files might look and will likely require some re-work. Essentially, instead of using watch_file <file>, you now write nix_watches+=(<file1> <file2>). The functionality was changed such that all the files in that array are now checked and whenever one of them changed, it will correctly trigger rebuilding the environment.

This would also properly fix #51

[Mic92]

cc @bbenne10

[bbenne10]

@Mic92: I suppose the question is whether we want a difference between invalidation of nix cache independent of the direnv cache. This would work fine for my use case and is a lot simpler than the PR I've opened since my approach uses functions rather than a variable. Simplicity, clarity, and aligning the nix cache and the direnv cache are all good things and so I am all for this approach over the one taken in #145.

The only trouble that I see with this is that it doesn't seem to touch use_nix, which (imo) it should. That should be trivial.

(As an aside: What do we think of maybe overriding watch_file? Doing so would allow us to unify the nix-direnv "watches" with the direnv watches (by simply dispatching to the original function) and would provide new users a discoverable and expected entry point that's already documented and utilized by the upstream. I don't want to cause more churn, but that seems like a pretty good solution to me?)

[polygon]

@bbenne10: Thanks for the feedback. Sorry, I guess we solved a similar problem at the same time. I haven't seen your PR before opening this one. In any case, If needed, I think the function approach would still be compatible with this. We could call this custom function if it exists, and then set need_update if it returns 1. However, I don't have a use-case myself beyond watching files.

Indeed, I was going for a minimum viable version to show at first. I've added support for use_nix now. I also like the idea of overriding watch_file and implemented this. I've had to add a check if the function was already renamed because for some reason, this file is sourced twice for me.

[bbenne10]

No need to be sorry for solving the same problem but better. I believe that your use-case is the same as mine overall and that overriding watch_file is a clever solution. If use_flake or use_nixis called, we unify the concepts of the two unique caches. If not, nothing unexpected happens (presuming no one else reads nix_watches). Sounds like a win to me :)

I'm hoping to get feedback from @dvzubarev on one or both of these PRs soon, but I understand that life happens and we may need to move forward without that.

I may suggest also updating the README to describe the functionality

[bbenne10]

Thinking on it - if we're overriding watch_file we may also need to override watch_dir to keep everything consistent, no?

[polygon]

Function is called nix_direnv_watch_file now to not override any existing functionality. Open for other name suggestions. If everyone is fine with this name, it should be ready now.

[bbenne10]

I think that some documentation is in order for this, but otherwise I'm happy to sign off on it.

[polygon]

I've replaced the Manually re-triggering evaluation chapter in the readme. Please have a look.

[bbenne10]

Lgtm

[Mic92]

Thanks!