Strange and inconsistent ordering behavior with ak.sum

[masonproffitt]

Version of Awkward Array

1.8.0

Description and code to reproduce

I basically expect ak.sum with axis=0 to add all rows together. So behavior like this makes sense:

>>> ak.sum([[[], [1]],
...         [[], [3]]], axis=0)
<Array [[], [4]] type='2 * var * int64'>

Just appending one element to each of the non-empty lists in the above example somehow changes ordering in the output:

>>> ak.sum([[[], [1, 2]],
...         [[], [3, 4]]], axis=0)
<Array [[4, 6], []] type='2 * var * int64'>

I of course expected to get [[], [4, 6]], so I'm not sure what's going on here.

[agoose77]

I can confirm this in both v2 and v1, and I can't find a particular commit that introduces it within 1.5 → HEAD