Repository navigation
Python 3.15.0-beta.2-linux 22.04 x64 seems to have issues. #1321
Description
Activity
One more note for context in case this is upstream: I'll change the use in NumPy to use
PyMem_Reallocwhich is the correct API here and makes more sense, it may be that the realloc on null behavior is funny for actual python objects.
This is also not much of a problem in practice since 24.04 works fine. But it seems to me there is still something weird here (unless Python forgot to update their docs and changed the guarantees).OK, I was pointed to python/cpython#151297 so this is fixed in Python and all that is needed is a rebuild.
Hello @seberg,
Thank you for reporting this issue. We will investigate it and get back to you as soon as we have some feedback.Hi @seberg,
Thank you for reporting this issue. We've have investigated on the same and wanted to let you know that, The
3.15.0-beta.2artifact (3.15.0-beta.2-26891732290) was built on June 3, 2026, prior to the upstream fix (python/cpython#151358) landing on June 12, 2026, which is why you had come across with the error.However, we would like to inform you that,
3.15.0-beta.3has now been released and the updated artifacts are available in actions/python-versions (via PR #401). We have verified that the fix is included on the free threaded build for Ubuntu 22.04 and no longer segfaults and completes successfully. Users can now usepython-version: '3.15.0-beta.3'withfreethreaded: truewithout encountering this issue. Closing this issue as resolved. Please feel free to reopen if you run into any further issues.
Description:
It seems that 3.15.0-beta.2-26891732290 and specifically
python-3.15.0-beta.2-linux-22.04-x64-freethreaded.tar.gzhas some build issue.I.e. this artifact: https://github.com/actions/python-versions/releases/download/3.15.0-beta.2-26891732290/python-3.15.0-beta.2-linux-22.04-x64-freethreaded.tar.gz
(I can locally verify that 24.04 works just fine with the reproducer below.)
This causes segfaults (one example
from numpy import random), but it can also be locally reproduced nicely with gdb:Inside gdb:
rthen Ctrl+c and then, let's do something that is definitely valid code:which then fails with a Segfault in
Action version:
See above for precise artifact.
Platform:
Runner type:
Tools version:
Repro steps:
Sorry, I don't have exact reproducer steps at hand right now since I debugged it locally. But I expect that is easier anyway.
As mentioned, installing NumPy and runningpython -c 'from numpy import random'will reproduce it, though.EDIT: Ah silly me to repro should just be something like:
(that also simplifies the gdb call)
(Sorry if this is the wrong place, I guess there is a chance it is yet again an upstream issue...)