C++ refactoring: ak.type and ak.values_astype

[ioanaif]

No description provided.

[codecov]

Codecov Report

Merging #1234 (5816e05) into main (0a0e9be) will decrease coverage by 0.49%.
The diff coverage is 78.51%.

Impacted Files	Coverage Δ
src/awkward/_v2/contents/bitmaskedarray.py	64.45% <ø> (-0.79%)	⬇️
src/awkward/_v2/contents/bytemaskedarray.py	82.81% <ø> (+3.66%)	⬆️
src/awkward/_v2/contents/emptyarray.py	69.54% <ø> (ø)
src/awkward/_v2/contents/indexedarray.py	60.74% <ø> (+0.98%)	⬆️
src/awkward/_v2/contents/content.py	81.57% <28.57%> (-1.62%)	⬇️
src/awkward/_v2/_typetracer.py	68.94% <33.33%> (-0.25%)	⬇️
src/awkward/_v2/operations/describe/ak_type.py	45.71% <38.70%> (-34.29%)	⬇️
src/awkward/_v2/highlevel.py	65.59% <60.00%> (+0.60%)	⬆️
...c/awkward/_v2/operations/reducers/ak_linear_fit.py	85.71% <66.66%> (-1.52%)	⬇️
src/awkward/_v2/operations/reducers/ak_all.py	91.66% <75.00%> (-8.34%)	⬇️
... and 50 more

[jpivarski]

On the whole, it's looking good, but I have a few requests. Most of them are about using centralized utilities, reducing the proliferation of type-lists that were in v1.

The type name for leaves of the type tree changed from PrimitiveType to NumpyType. As far as I can see, there is no ak._v2.types.PrimitiveType, so how are these tests working? Virtually every attempt to build a type would have to terminate on these leaf-nodes, so you ought to be getting

AttributeError: module 'awkward._v2.types' has no attribute 'PrimitiveType'

(I don't see a PrimitiveType being defined in this diff.)

Anyway, in an attempt to use terminology more consistently, the NumpyArray has NumpyForm and NumpyType. We still use the word "primitive," but it refers to the string naming the type in Forms. v1 and v2 have to be able to accept each others' Forms so that pickled data from each version can be used in the other (that was a goal, so that users could build work-arounds if they need to), and that means the JSON field name "primitive" in the Form can't be changed. (Otherwise, a better name might be "dtype", though that would also be misleading because dtypes are Python objects, and this is a string. So it's just going to be "primitive".)