Issue with current documentation:

After reading Data type promotion in NumPy, NEP50, and Changes to NumPy data type promotion, the following type promotion scenario with Python's builtin scalar types (i.e., not scalar values) is still surprising to me:

>>> np.result_type(7, np.array([1], np.float32))
dtype('float32')  # Ok, the scalar value '7' does not impact type promotion per NEP50

>>> np.result_type(type(7), np.array([1], np.float32))
dtype('float64')  # Surprisingly, the *type* of the scalar value '7' does impact promotion

# Similar situations happen with Python's float and complex types.

What I assume happens is that Python's int type is converted internally to an np.int64 (this is on a 64-bit system):

>>> np.result_type(int)
dtype('int64')

, and np.int64 is then used for the type promotion.

However, I could not find this scenario explained in the docs.

cc @inducer

Idea or request for content:

I'd suggest adding an explanation to Data type promotion in NumPy, perhaps in the "Exceptions to the general promotion rules" section. I'd be happy to open a PR if that is helpful.