[Python] Negative out of range slices yield invalid arrays

[asfimport]

Tested on pyarrow 2.0 and pyarrow 4.0 wheels.  The errors are slightly different between the 2.0.  Below is a script from 4.0

 

This is taken from the result of test_slice_array

 
>>> import pyarrow as pa
>>> pa.array(range(0,10))
<pyarrow.lib.Int64Array object at 0x7f59b8bdab20>
[
0,
1,
2,
3,
4,
5,
6,
7,
8,
9
]
>>> a=pa.array(range(0,10))
>>> a[-9:-20]
<pyarrow.lib.Int64Array object at 0x7f59b8bdaa00>
[]
>>> len(a[-9:-20])
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
SystemError: <built-in function len> returned NULL without setting an error

Reporter: Micah Kornfield / @emkornfield
Assignee: Joris Van den Bossche / @jorisvandenbossche

Related issues:

• [Python] Slicing array can lead to negative length (is cloned by)

PRs and other links:

• GitHub Pull Request #10341

_{Note: This issue was originally created as ARROW-12769. Please see the migration documentation for further details.}

[asfimport]

Eduardo Ponce / @edponce:
This error occurs in the following combinations of the range indices:

1. (Reverse range) First index > 0 and first index > len(arr), and second index < 0

2. (Reverse range) First index < 0, and 0 < second index < len(arr) + first index

3. (Reverse range) First index < 0, and second index < first index

   All of these case do result in an empty IntXArray but the len() operation triggers the error.

[asfimport]

Joris Van den Bossche / @jorisvandenbossche:
It also seems to happen simply when start > stop with positive indices (eg arr[5:3])

[asfimport]

David Li / @lidavidm:
Issue resolved by pull request 10341
#10341