[C++] "iso_calendar" kernel returns incorrect results for array length > 32

[mroeschke]

Describe the bug, including details regarding any error messages, version, and platform.

In [1]: import pyarrow as pa

In [2]: pa.__version__
Out[2]: '14.0.0'

In [3]: import pyarrow.compute as pc

In [4]: from datetime import  datetime

In [5]: arr = pa.array([datetime(2019, 1, 3, 5, 11)]*50)

In [6]: pc.iso_calendar(arr)
Out[6]: 
<pyarrow.lib.StructArray object at 0x1500ff640>
-- is_valid: all not null
-- child 0 type: int64
  [
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    ...
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0
  ]
-- child 1 type: int64
  [
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    2019,
    ...
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0
  ]
-- child 2 type: int64
  [
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    1,
    ...
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0,
    0
  ]

In [7]: pc.iso_calendar(arr)
zsh: segmentation fault  ipython

Component(s)

Python

[jorisvandenbossche]

Thanks! I can't reproduce the segfault, but it's of course obviously wrong anyway.

It seems the repeated values don't matter, I can also reproduce this with a date range, and it starts to happens once above 32 values:

arr = pa.array(pd.date_range("2019-01-01", periods=33, freq="3H"))
pc.iso_calendar(arr)

cc @rok

[rok]

Oh, that is interesting. I'll take a look this week.