BUG: np.bool_ alias not exported in __init__.pyi with 2.0.0rc1

[jorenham]

Describe the issue:

Type checkers (tested with pyright==1.1.356 in strict mode) don't recognize np.bool_ as alias for np.bool on numpy==2.0.0rc1.

Unless I'm missing something, this is because bool_ is missing in __init__.pyi.

Reproduce the code example:

import numpy as np

type NpBool = np.bool_  # results in pyright errors

Error message:

error: Type of "bool_" is unknown (reportUnknownMemberType)
error: "bool_" is not a known member of module "numpy" (reportAttributeAccessIssue)

Python and NumPy Versions:

2.0.0rc1
3.12.2 (main, Feb 25 2024, 16:35:05) [GCC 11.4.0]

Runtime Environment:

No response

Context for the issue:

This makes it impossible to add support for both numpy 1 and 2 in typed python packages, e.g. in https://github.com/jorenham/Lmo/blob/master/lmo/typing.py

[rgommers]

Unless I'm missing something, this is because bool_ is missing in __init__.pyi.

This was done in gh-25080. There's some discussion in that PR; if needed we can revisit the decision, but in principle non-recommended aliases (as well as deprecated items) are not exposed in .pyi files.

In principle I don't see an issue with adding back np.bool_ as an alias of np.bool for a while (eventually we should remove it though I think). I think it wasn't discussed carefully enough in gh-25080. The hard work done there was to update all the .pyi files to make np.bool canonical and used everywhere in the 2.0 type stubs. At the end np.bool_ could have been added back though, since package authors cannot unconditionally use np.bool yet as long as they support numpy 1.x.

[jorenham]

The np.bool_ being an alias is mentioned in
https://github.com/numpy/numpy/blob/v2.0.0rc1/doc/source/release/2.0.0-notes.rst#changes.

But surprisingly, I can't seem to find anything about this (breaking) np.bool change in the migration guide. I'll open a new issue for this

[seberg]

Since np.bool_ is the only way to write things compatible and it is not yet deprecated, I think it should be re-added.