Should we do the same for np.object_ vs np.object?

+1 from me, np.object is perfectly clear. Less important than bool, but has the same rationale.

In general, will bool still be an alias/converted for the underlying numpy boolean type? Is using bool as the dtype the most compatible way across <1.20.0, >=1.20.0, <1.25, and going forward?

Hi, if this issue is untaken, I would like to have this as my first issue. Can I please get some pointers on where to start?

Hi, I come from #23494.

For now, to avoid confussion, perhaps the text of the FutureWarning should be changed?
From the curent text to something along the lines of "Starting in v1.24, 'np.bool' was deprecated in favour of 'np.bool_', but will be aliased to 'np.bool_' in the future. For now, please use 'np.bool_'"?

Provides more information about what is going on, and seems less confusing/conflicting when paired with the AttributeError raised later.

IMO, the confussion comes mainly from the current wording. The warning states that np.bool will be deprecated "in the future", but then an attritube exception is raised. Changing the text to reflect that np.bool has already been deprecated, then raising the exception prevents confussion about why the exception has been raised.

np.bool is back alive and the preferred name for numpy 2.0 (np.bool_ is an alias to np.bool now):

>>> np.bool
<class 'numpy.bool'>
>>> np.bool_
<class 'numpy.bool'>

Should we do the same for np.object_ vs np.object?

This wasn't done. Can still be done at any time but isn't a blocker for 2.0, so I'll open a new issue for it now to not forget about it. Closing this one to mark that the issue with bool has been addressed.

What is the purpose of np.bool ? Should it be preferred to bool when instantiating numpy arrays? Is it faster or more memory efficient?

It's a different scalar that behaves slightly differently (there are two reasons for this, historically because numpy scalars pretend to be arrays in part, and in part because Python bools behave more like ints sometimes.)

EDIT: and no, as dtype= it doesn't matter, Python bool is just converted to NumPy bool. There is no dtype that behaves exactly like the Python bool.