uv lock should invalidate lockfile if registries change

[charliermarsh]

If a lockfile lists a package from a registry that isn't part of the index URLs passed to uv lock, we should probably not reuse it.

[charliermarsh]

I'm actually unsure about this. Like if you do:

uv add idna --index-url https://test.pypi.org/simple

And then uv lock later (no arguments)... should we "allow" the lockfile to use the existing entries, which reference https://test.pypi.org/simple?

[zanieb]

Definitely! Seems weird to force everything to use another index URL after an add operation.

[charliermarsh]

The only quirk is that if you don’t specify any index URL, we treat it as PyPI, so the lock would be invalidated if you “stop” passing it on the command-line for any command.

[charliermarsh]

Actually, I think the type is rich enough that we can avoid enforcing this if no index URLs are provided (i.e., assume the lockfile is fine). It's a little strange because if we fail, we'll then resolve against PyPI. But probably fine in practice.