fix: strided interpretation for data with extra offsets

[ioanaif]

No description provided.

[ioanaif]

pre-commit.ci autofix

[ioanaif]

I compared the last issue of the non-primitive dtypes being passed on to awkward in the test added by this PR with tests/test_0034-generic-objects-in-ttrees.py::test_jagged_strided_awkward

I tracked where the byte order changes. This happens in interpretation.jagged.final_array():



print("content dtype ", cnt['fP/fX'].dtype) => content dtype  >f8
content = numpy.empty((before,), self.content.to_dtype)
before = 0
for cnt in contents:
    content[before : before + len(cnt)] = cnt
    before += len(cnt)

print(“content dtype after “, content['fP/fX'].dtype) => content dtype after  float64

For the test I added, in interpretation.numerical.final_array() this transformation doesn’t take place so the byteorder is preserved and thus awkward complains about the fact that it gets a non-native dtype. This is solved by adding output = output.newbyteorder('=') before self.hook_before_library_finalize.

[jpivarski]

Only change (which we discussed on Zoom): hide the new (None, None) elements of members by determining _all_headers_prepended once for each AsStridedObjects in its __init__, and then remove the (None, None) from the AsStridedObjects._members that it presents to the world (through the members property).

[jpivarski]

Yes, this is much better. Previously, this logic was happening in AsDtype, but now it's in AsStridedObjects, and that's good because it can only affect this one subclass.

Everything looks ready to merge. Go ahead and do it when you're done with the git branch!

[jpivarski]

Right, because the file already exists. This will allow the test to be tested on more systems. Thanks for catching this!