Repository navigation
TST: devdeps segfault #19621
Description
Started noticing this morning. Was fine last night.
Notable details:
...
Package versions:
Numpy: 2.5.0.dev0+git20260428.6ec400a
Scipy: 1.18.0.dev0+git20260425.f1f7a63
Matplotlib: 3.11.0rc1
h5py: not available
Pandas: not available
PyERFA: 2.0.1.7.dev138+g46feb489b
Cython: not available
Scikit-image: not available
pyarrow: not available
asdf-astropy: not available
Using Astropy options: remote_data: any.
hypothesis profile 'ci' -> database=None, deadline=None, print_blob=True, derandomize=True, suppress_health_check=(HealthCheck.differing_executors,)
CI: true
ARCH_ON_CI: normal
IS_CRON: false
...
plugins: astropy-0.11.0, cov-7.1.0, doctestplus-1.7.1, astropy-header-0.2.2, filter-subpackage-0.2.0, xdist-3.8.0, mock-3.15.1, remotedata-0.4.1, arraydiff-0.6.1, hypothesis-6.152.4
...
astropy/uncertainty/tests/test_distribution.py::TestStructuredDistribution::test_setitem_field[b] FAILED
...
corrupted size vs. prev_size
Fatal Python error: Aborted
Thread 0x00007f6d93bff6c0 (most recent call first):
File "/home/runner/work/astropy/astropy/.tox/py313-test-devdeps/lib/python3.13/site-packages/astropy/samp/client.py", line 200 in _serve_forever
File "/opt/hostedtoolcache/Python/3.13.13/x64/lib/python3.13/threading.py", line 995 in run
File "/opt/hostedtoolcache/Python/3.13.13/x64/lib/python3.13/threading.py", line 1044 in _bootstrap_inner
File "/opt/hostedtoolcache/Python/3.13.13/x64/lib/python3.13/threading.py", line 1015 in _bootstrap
...
Activity
neutrinoceros commentedon Apr 30, 2026
I know I said I wouldn't touch this before FF but I couldn't resist. I haven't managed to reproduce this on MacOS yet, gonna try on a Linux machine.
neutrinoceros commentedon Apr 30, 2026
scipy is off the hook, I hit the segfault (well, I hope it's the same one) without it on the stack.
neutrinoceros commentedon Apr 30, 2026
same goes for matplotlib
neutrinoceros commentedon Apr 30, 2026
It's almost certainly numpy. I'll bisect for it
neutrinoceros commentedon Apr 30, 2026
bisection took a while because my linux machine is much less capable, but I got a result out: numpy/numpy#29270
neutrinoceros commentedon Apr 30, 2026
another important result, that I wish I found earlier, is that the segfault is easily triggered by running just
astropy/uncertaintytests.neutrinoceros commentedon Apr 30, 2026
discussion moving upstream to numpy/numpy#31363
pllim commentedon Apr 30, 2026
Thank you! I was getting worried because ideally we want a green CI (even for devdeps) in RC period. Looks like progress is being made in upstream discussions. 🤞
xref #15304
neutrinoceros commentedon May 2, 2026
This is fixed on numpy main. Devdeps is expected to turn green again with the next nightly batch.
pllim commentedon May 4, 2026
devdeps no longer segfault, closing. Thanks, all!