len(np.lib.__all__) drops from 87 to 71 with this PR. Everything removed is still available in the top-level namespace.

It would probably be worthwhile to add errors redirecting people to the main namespace or whichever other migration path is appropriate for that submodule if they try to access any np.lib submodules we're removing.

This could be a followup though, no need to hold up merging this, which looks good otherwise except for the one issue I pointed out in the docs.

It would probably be worthwhile to add errors redirecting people to the main namespace or whichever other migration path is appropriate for that submodule if they try to access any np.lib submodules we're removing.

This could be a followup though, no need to hold up merging this, which looks good otherwise except for the one issue I pointed out in the docs.

@ngoldbaum I think that's a good point - I added it here to def __getattr__(): in numpy/lib/__init__.py. If a user wants to access a function from a lib module like:

np.lib.arraysetops.unique(src_ids) 

The message says:

AttributeError: `np.lib.arraysetops` is now private. Function that you're looking for should be available from the main namespace `np.*`

Let's merge this one as well so the other np.lib PRs can get updated to add to the __getattr__ error. Thanks for dealing with all the merge conflicts and making the PR diffs a lot smaller!