Dask awkward segfault (for https://github.com/dask-contrib/dask-awkward/issues/140)

dak_test.py

# coding: utf-8


import dask_awkward as dak
import awkward as ak


# located at https://cernbox.cern.ch/remote.php/dav/public-files/rvipIIozKQAypwu/calib.parquet
input_path = "calib.parquet"

# same file, saved with extensionarray=False
# located at https://cernbox.cern.ch/remote.php/dav/public-files/xQTpSqrvZQSSTZp/calib_no_extensionarray.parquet
# input_path = "calib_no_extensionarray.parquet"


def load(**kwargs):
    dak_array = dak.from_parquet(input_path, **kwargs)
    print(f"npartitions: {dak_array.npartitions}")
    print(f"divisions  : {dak_array.divisions}")

    # compute
    ak_array = dak_array.compute()
    print(f"ak array   : {ak_array}")

    return ak_array


print(f"dask_awkward {dak.__version__}")
print(f"awkward {ak.__version__}")

print(80 * "-")

# load all columns
load()  # works nicely

print(80 * "-")

# load some 1st-level columns
load(columns=["mc_weight", "deterministic_seed"])  # works nicely

print(80 * "-")

# load all (!) 2nd-level columns under "Jet"
load(columns=["Jet.*"])  # works nicely

print(80 * "-")

# load a subset of 2nd-level columns under "Jet"
load(columns=["Jet.pt"])  # segfaults at compute()