Im thinking to further avoid this issue, we could also type this correctly. Maybe we can just represent the root subdirectory as "" too? Or if we do want to makenthe distinction between not specified and empty we could add a new type?

There are a few (serialization) failures now because e.g. subdir: . serializes to None now. I implemented normalization for the path (./foobar == foobar == foobar/ == foobar/././// etc.).

@baszalmstra do you know the impact of this? I don't think there will be a lot of subdir: . in the wild, so hopefully not a crazy amount of lockfile churn. Alternatively, we can do normalization only when comparing the subdirs and keep faithful serialization.

@wolfv If you can get it to work with a subdir, I think that that would be great, but I must agree, as a person with a git sourced PyPI package in my pixi project's pyproject.toml, I do not have a subdir. If you can maybe merge the change with churn on subdir and address that issue in another PR later on, so that those like me who are facing the lock file issues now can get this fix in the next pixi version and those with the subdir may get a fix later on, I think that this would be the better approach 🙏 (Just my 2 cents as a user of pixi)