[ARROW-10194] [Python] Array.to_numpy() with type fixed_size_list(int64(), 1) doesn't roundtrip for large integer values

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Python
• Labels: python-conversion
• External issue URL: https://github.com/apache/arrow/issues/26199
• Assignee: Unassigned
• Reporter: Krisztian Szucs
• Created: 06/Oct/20 13:55
• Updated: 11/Jan/23 08:11

Description

Reproducer:

data = [None, [9007199254740993]]
arr = pa.array(data, type=pa.list_(pa.uint64(), 1))
ndarray = arr.to_numpy(zero_copy_only=False)
restored = pa.array(ndarray, type=arr.type)
assert restored.equals(arr)

Error:

E       assert False
E        +  where False = <built-in method equals of pyarrow.lib.FixedSizeListArray object at 0x7fbdb0239e20>(<pyarrow.lib.FixedSizeListArray object at 0x7fbdb0262220>\n[\n  null,\n  [\n    9007199254740993\n  ]\n])
E        +    where <built-in method equals of pyarrow.lib.FixedSizeListArray object at 0x7fbdb0239e20> = <pyarrow.lib.FixedSizeListArray object at 0x7fbdb0239e20>\n[\n  null,\n  [\n    9007199254740992\n  ]\n].equals

The inner numpy array (ndarray[1]) has float64 dtype where the integer gets truncated because of the precision.