Equivalent of Make.user for git-ignored local dependencies

[MilesCranmer]

It seems there is no way to add a development dependency to a virtual environment without modifying pyproject.toml. The issue with this approach is that pyproject.toml is often tracked by git, so if I want to add personal development dependencies to a python project, I either need to:

1. Work with a dirty git state, or
2. Stash and unstash the changes repeatedly, or
3. Use git update-index to ignore the pyproject.

While it is useful to have the [tool.rye] for global settings across multiple users, I think it would also be good to have a separate per-project file so I can set local settings. .rye.toml might make the most sense – this would basically allow a user to set all of the settings in [tool.rye], but locally, without changing the tracked pyproject settings.

[lorenzogatti]

You'd need to merge conflicting directives from two files, consistently for each command, and pyproject.toml would become unreliable.
What "personal development dependencies" are you unable to configure and share normally? Should they be managed by Rye in the first place?

[MilesCranmer]

.rye.toml would override pyproject.toml in the case of conflicts. You can think of it like a Make.user file for a Makefile based build, if that’s familiar at all.

Personal development dependencies are things like IPython. I want them in the virtualenv because a global IPython (as its warning message states) can fail to load the correct versions of libraries from a virtual environment. So I always want to put it in the local environment too.

But if I’m using rye, this means I perpetually have a pyproject.toml that is showing up in the unstaged changes for my git repo. If I had a separate .rye.toml or whatever, I could put it in the gitignore and not have this issue. It’s similar to why Make.user exists and why you don’t just have people editing a Makefile directly - because then git would always be asking if you want to commit those changes. A separate file = I can put it in the gitignore.

[MilesCranmer]

@charliermarsh what do you think?