[Python] Calling pyarrow.dataset.ParquetFileFormat.make_write_options as a class method results in a segfault

[nonatomiclabs]

Description

I had issues finding good examples of how to use make_write_options and ended up calling it without having instantiated the ParquetFileFormat, which resulted in a segmentation fault:

>>> import pyarrow.dataset
>>> pyarrow.dataset.ParquetFileFormat.make_write_options(42)
[1]    64569 segmentation fault  python

Expected behavior

Even though the way I called the method was nonsense, I'm guessing there could be better error handling than a segfault 🙂

Additional information

• The issue happens regardless of the parameters passed (as long as there is at least one)
• Tested on Python 3.11.6 and 3.8.14
• PyArrow version 14.0.2
• macOS 14.2 ARM

Component(s)

Python

[raulcd]

This probably should be a blocker for the release, right?

[jorisvandenbossche]

Nice to have fixed given that it is a segfault, but it's not new, so I wouldn't say it's a blocker.

We have this issue in general, when calling cython methods as class instead of instance methods, many of them will segfault. For example also:

In [9]: pa.Array.get_total_buffer_size(1)
Segmentation fault (core dumped)

Because we nowhere do any type checking on self and just assume it is the correct type.

Now, in this case of ParquetFileFormat.make_write_options, it's more important I think, because it makes more sense to indeed try to call this thinking it is a static class method.

[AlenkaF]

Issue resolved by pull request 40976
#40976