[Python] segfault on s390x with the latest version (6.33.0)

[frenzymadness]

What version of protobuf and what language are you using?
Version: 6.33.0
Language: Python

What operating system (Linux, Windows, ...) and version?
Fedora Linux 41 in a container on s390x (emulated via qemu)

What runtime / compiler are you using (e.g., python version or gcc version)
Python 3.13.7 installed from Fedora 41 repositories

What did you do?
Steps to reproduce the behavior:

1. Clone https://github.com/Netflix/nflxprofile on s390x machine
2. Go to python directory
3. Run tests via tox, or
4. install the project pip install -e . and run python3 -m unittest

What did you expect to see

Tests passing successfully like

# python -m unittest

…

----------------------------------------------------------------------
Ran 4 tests in 0.069s

OK

What did you see instead?

# python -m unittest
qemu: uncaught target signal 11 (Segmentation fault) - core dumped
Segmentation fault (core dumped)

Anything else we should know about your project / environment

We are using the project mentioned above in our CI to test fedora-python-tox container image, and the CI started to fail when protobuf 6.33.0 went out. I don't know the project I'm using here, and there are no commits in it for the last 3 years, so the codebase might be outdated.

I've tried to investigate what is wrong but my knowledge is not deep enough.

What I know:

• It works fine with the previous version (6.32.1)
• Installing the latest version from sources (instead of the pre-built wheel for s390x) doesn't help.

When I tried to switch protobuf to use the pure Python implementation, it failed as well, but surprisingly only after the tests had run:

# PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python python -m unittest
…
----------------------------------------------------------------------
Ran 4 tests in 0.058s

OK
qemu: uncaught target signal 11 (Segmentation fault) - core dumped
Segmentation fault (core dumped)

[mkruskal-google]

While we release s390x binaries, it's community-supported and rely on PRs to keep it working because we don't currently have any way to catch or debug issues like this. If someone can build python-upb on s390x in ASAN and give a stack trace that might help narrow down this problem more.

Also note: 33.0 was the first release where we released s390x upb binaries to PyPi. So in 32.x you would have been using pure python.

[frenzymadness]

Thank you for the quick reply. Let's wait and see whether others hit the same problem or not. I have a simple workaround that is fine for my purpose.

[barracuda156]

If this happens to be endianness-caused, then might be related: #24158 (this issue is on powerpc).