fix: array.dtype.type will never be in ak.types.numpytype._dtype_to_primitive_dict

[lgray]

Fixes #1840

Original input to square brackets would never work?

[codecov]

Codecov Report

Merging #1841 (678aed9) into main (c033697) will increase coverage by 0.02%.
The diff coverage is 81.25%.

Additional details and impacted files

Impacted Files	Coverage Δ
src/awkward/operations/ak_type.py	73.91% <81.25%> (+18.64%)	⬆️
src/awkward/contents/indexedarray.py	76.77% <0.00%> (-0.19%)	⬇️
src/awkward/contents/recordarray.py	83.94% <0.00%> (-0.19%)	⬇️
src/awkward/_util.py	82.15% <0.00%> (-0.18%)	⬇️
src/awkward/contents/indexedoptionarray.py	88.96% <0.00%> (-0.16%)	⬇️
src/awkward/_connect/avro.py	87.17% <0.00%> (-0.14%)	⬇️
src/awkward/types/_awkward_datashape_parser.py	47.72% <0.00%> (-0.01%)	⬇️
src/awkward/contents/content.py	77.31% <0.00%> (+0.10%)	⬆️

[agoose77]

Would you be able to add a test_1840-type-numpy-array.py to tests/ containing the following:

# BSD 3-Clause License; see https://github.com/scikit-hep/awkward-1.0/blob/main/LICENSE

import numpy as np  # noqa: F401
import pytest  # noqa: F401

import awkward as ak  # noqa: F401


def test():
    array = np.random.random(size=512).astype(dtype=np.float64)
    assert ak.type(array) == ak.types.ArrayType(
        ak.types.NumpyType("float64"), 512
    )

[lgray]

Sure! Just a bit.

[jpivarski]

I'm working on fixes to this, but I'm in a meeting now, so that's paused.

[jpivarski]

(Do not merge!)

[agoose77]

Weirdly enough, I was actually looking at this code for another issue when I noticed that we have a dtype_to_primitive function that we can use here.

[jpivarski]

I got the 10 minutes to do this, and now realize that I had misunderstood the original issue #1840. I thought it was about making ak.type(np.float64) work, but it had been more broken than that: ak.type(np.array([1, 2, 3])) didn't work.

Well, all of these cases work now. Since it's doubly approved, I'll reenable auto-merge.

[agoose77]

I've held this from merging, as we probably want to use numpytype.dtype_to_primitive - but we might need to do more with respect to the loop that checks for membership in the type dictionary. We could use a try/except, but perhaps there's a more elegant way of writing this code.

[jpivarski]

The dtype_to_primitive function is much better. I think it includes some logic for getting the time units of datetime/timedelta types right. Even if it doesn't, it's the right function to use here.

Just one problem below. I'll self-accept my suggestion and enable auto-merge.

[jpivarski]

I think this is done now; I'm reenabling auto-merge.