[Python] Support `mask` in `FixedSizeListArray.from_arrays`

[kylebarron]

Describe the enhancement requested

In ListArray.from_arrays it's possible to pass in a mask array defining which list elements are null. FixedSizeListArray.from_arrays does not currently have a mask parameter, so it appears impossible to create a fixed size list array where some elements are null?

pa.array([1, 2, 3, 4], mask=[False, True, False, False])
# <pyarrow.lib.Int64Array object at 0x126eb29e0>
# [
#   1,
#   null,
#   3,
#   4
# ]
pa.FixedSizeListArray.from_arrays([1, 2, 3, 4], 2)
# <pyarrow.lib.FixedSizeListArray object at 0x127255900>
# [
#   [
#     1,
#     2
#   ],
#   [
#     3,
#     4
#   ]
# ]
pa.FixedSizeListArray.from_arrays([1, 2, 3, 4], 2, mask=[False, True])
# TypeError                                 Traceback (most recent call last)
# /Users/kyle/fused/application/job/job/raster/naip/async_zonal_stats.py in line 1
# ----> 449 pa.FixedSizeListArray.from_arrays([1, 2, 3, 4], 2, mask=[False, True])
#
# File ~/fused/application/job/.venv/lib/python3.11/site-packages/pyarrow/array.pxi:2141, in pyarrow.lib.FixedSizeListArray.from_arrays()
#
# TypeError: from_arrays() got an unexpected keyword argument 'mask'

Component(s)

Python

[rok]

This works:

import pyarrow as pa
array = pa.array([1, 2, 3, 4, 5],
                 mask=pa.array([True, False, True, False, True]))
pa.FixedSizeListArray.from_arrays(array, 1)

I assume you're looking to store multidimensional arrays? These discussions might be relevant: #33925 #8510

[kylebarron]

That doesn't appear to work:

In [8]: primitive_array = pa.array([1, 2, 3, 4, 5], mask=pa.array([True, False, True, False, True]))

In [9]: list_arr = pa.FixedSizeListArray.from_arrays(primitive_array, 1)

In [10]: list_arr.is_valid()
Out[10]:
<pyarrow.lib.BooleanArray object at 0x1028a4dc0>
[
  true,
  true,
  true,
  true,
  true
]

In your example the validity is assigned to the underlying primitive array, not the fixed size list array itself. According to https://arrow.apache.org/docs/format/Columnar.html#buffer-listing-for-each-layout the FixedSizeListArray should store its own validity array. I want to ensure FFI compatibility between Arrow implementations, which means it's important to be able to set the fixed size list array's mask correctly.

I'm looking to store geoarrow points, not multidimensional arrays, so those discussions don't appear to be relevant to me

[rok]

Indeed from_arrays doesn't pass the mask to FixedSizedListArray. Did you try using from_buffers and passing the validity bitmap? Some examples here.

[jorisvandenbossche]

@kylebarron thanks for raising the issue! Similarly as we added this keyword to ListArray.from_arrays recently(#13894), we can also add this to FixedSizeListArray.from_arrays

Short term, as @rok notes, from_buffers might be a reasonable alternative (and actually also ensures the validity buffer doesn't need to be inverted and can be used zero-copy):

>>> arr = pa.array([1, 2, 3, 4])
>>> pa.Array.from_buffers(pa.list_(pa.int64(), 2), 2, [None], children=[arr])
<pyarrow.lib.FixedSizeListArray object at 0x7f6ac7eb7ac0>
[
  [
    1,
    2
  ],
  [
    3,
    4
  ]
]

>>> validity = pa.array([True, False])
>>> pa.Array.from_buffers(pa.list_(pa.int64(), 2), 2, [validity.buffers()[1]], children=[arr])
<pyarrow.lib.FixedSizeListArray object at 0x7f6ac408e380>
[
  [
    1,
    2
  ],
  null
]

[LucasG0]

HI, I am working on this issue.