ak.type does not understand numpy.<type> style dtypes

[lgray]

Version of Awkward Array

2.0.0rc1

Description and code to reproduce

numpy = 1.23.4

>>> import awkward as ak
>>> import numpy as np
>>> x = np.random.normal(size=100)
>>> ak.type(x)

results in:

Traceback (most recent call last):
  File "/Users/lgray/miniforge3/envs/coffea-dev/lib/python3.8/site-packages/awkward/operations/ak_type.py", line 99, in _impl
    out = ak.types.numpytype._dtype_to_primitive_dict[array.dtype.type]
KeyError: <class 'numpy.float64'>

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/lgray/miniforge3/envs/coffea-dev/lib/python3.8/site-packages/awkward/operations/ak_type.py", line 60, in type
    return _impl(array)
  File "/Users/lgray/miniforge3/envs/coffea-dev/lib/python3.8/site-packages/awkward/operations/ak_type.py", line 101, in _impl
    raise ak._errors.wrap_error(
TypeError: while calling

    ak.type(
        array = numpy.ndarray([ 0.27824033 -1.483569   -0.61108357 ...
    )

Error details: numpy array type is unrecognized by awkward: <class 'numpy.float64'>

np.float64 (or np.<type> in general) is a fairly common way for folks to denote typing, we should probably support it.

[jpivarski]

Here's where it can go: in awkward/operations/ak_type.py, there's a case-catcher for

awkward/src/awkward/operations/ak_type.py

Lines 67 to 73 in c033697

	elif isinstance(
	array,
	tuple(x.type for x in ak.types.numpytype._dtype_to_primitive_dict),
	):
	return ak.types.NumpyType(
	ak.types.numpytype._dtype_to_primitive_dict[array.dtype]
	)

The keys of that dict are actual np.dtype objects, but you want the type instances, which are obtained via some_dtype_object.type.

They can be identified as

isinstance(array, type) and issubclass(array, np.generic)

[lgray]

Wouldn't this be fixed instead by asking for array.dtype instead of array.dtype.type (which gives the class as opposed to the dtype). Here: https://github.com/scikit-hep/awkward/blob/main/src/awkward/operations/ak_type.py#L99

?

[agoose77]

Thanks for this @lgray. We should be testing the comprehensive dtype here rather than the underlying type.