`uproot.dask` raises error on empty branches

[masonproffitt]

See example below. The exact file used is https://github.com/iris-hep/func_adl_uproot/blob/61b35593b4512dfaf3a75054aa13ffe2fbc602b6/tests/empty_branches_tree_file.root. It's a perfectly valid tree; there are just no entries, so I expect to get empty arrays rather than an error.

>>> import uproot
>>> uproot.__version__
'5.0.0rc2'
>>> uproot.dask('tests/empty_branches_tree_file.root')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/user/miniconda3/envs/func_adl_uproot_rc/lib/python3.10/site-packages/uproot/_dask.py", line 143, in dask
    return _get_dak_array(
  File "/home/user/miniconda3/envs/func_adl_uproot_rc/lib/python3.10/site-packages/uproot/_dask.py", line 483, in _get_dak_array
    ].basket_entry_start_stop(0)
  File "/home/user/miniconda3/envs/func_adl_uproot_rc/lib/python3.10/site-packages/uproot/behaviors/TBranch.py", line 2112, in basket_entry_start_stop
    raise IndexError(
IndexError: branch 'int_branch' has 0 baskets; cannot get starting entry for basket 0
in file tests/empty_branches_tree_file.root

[jpivarski]

That's right, and also this is handled correctly by concatenate, arrays and array:

>>> uproot.concatenate('empty_branches_tree_file.root')
<Array [] type='0 * {int_branch: int32, float_branch: float32}'>
>>> uproot.open('empty_branches_tree_file.root:tree').arrays()
<Array [] type='0 * {int_branch: int32, float_branch: float32}'>
>>> uproot.open('empty_branches_tree_file.root:tree/int_branch').array()
<Array [] type='0 * int32'>
>>> next(uproot.open('empty_branches_tree_file.root:tree').iterate())
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
StopIteration

So it's something in the uproot.dask logic, and @kkothari2001 should take a look. I'll put this file in scikit-hep-testdata so that we can add it to the tests.

Thanks!

[jpivarski]

It's ready: scikit-hep-testdata 0.4.20 has uproot-issue-697.root, which should be used in a test to ensure that uproot.dask can make an empty Dask array from it. (Maybe that array will need to have one partition, but that partition can be an empty array.)

[kkothari2001]

Hey @masonproffitt , thanks for finding the bug! I'm solving it in #700.

@jpivarski please confirm the behaviour:

• When there are one or many files, each of which is empty, a single dask chunk is returned which will return an empty numpy/awkward array.
• When there are multiple files with some (but not all) empty branches, we just skip over this file , no empty-array dask chunk is created here.

(PS: Sorry, I didn't mean to be nitpicky with the PR title, I accidentally thought this issue was the PR I created because of the similar titles and intended to correct my own mistake).

[jpivarski]

What you describe sounds like a good behavior to adopt. For any valid ROOT files, the user will get an array out, though it might be empty.

No problem changing the issue title name. I will sometimes change an issue title name as I find out more about it (if, for instance, the original title described a symptom and we later discover the cause). The issue titles are our list of to-do items, and we should feel free to make them more understandable to ourselves so that it's easier to find actionable tasks.