[Python][C++] Segfault when casting a sliced ListArray of int64 in v4.0.0

[asfimport]

Hi !

I just updated to pyarrow 4.0.0 and I noticed that a segfault happens when casting a sliced ListArray of int64. It crashes on both linux and windows. Here is a code to reproduce:

import pyarrow as pa

arr = pa.array([[i * 10] for i in range(4)])
print(type(arr))  # <class 'pyarrow.lib.ListArray'>
print(arr.type)  # list<item: int64>
arr.cast(pa.list_(pa.int32()))  # works
print(">>> Casting without slicing works")

arr = arr.slice(1)
print(type(arr))  # <class 'pyarrow.lib.ListArray'>
print(arr.type)  # list<item: int64>
arr.cast(pa.list_(pa.int32()))  # crashes !!
print(">>> This message isn't printed because casting with slicing causes a segfault")

On pyarrow 3.0.0 I used to get

ArrowNotImplementedError("Casting sliced lists (non-zero offset) not yet implemented")

instead of the segfault.

Thanks in advance for your help !

Reporter: quentin lhoest / @lhoestq

PRs and other links:

• GitHub Pull Request #10184

_{Note: This issue was originally created as ARROW-12568. Please see the migration documentation for further details.}

[asfimport]

yibocai#1:
From gdb, the error is at

arrow/cpp/src/arrow/compute/kernels/scalar_cast_nested.cc

Line 69 in 3a6f605

CopyBitmap(ctx->memory_pool(), in_array.buffers[0]->data(),

Second augment "in_array.buffers[0]->data()" deferences nullptr as "in_array.buffers[0]" can be null.

Looks a critical bug.

[asfimport]

Antoine Pitrou / @pitrou:
Issue resolved by pull request 10184
#10184

[asfimport]

quentin lhoest / @lhoestq:
Thanks !