Currently blocked by multiple regular but not deterministic segmentation faults, that seem to be most commonly from fffpy_multi_iterator_delete or elsewhere in the fffpy_multi_iterator stuff. Test files currently causing segmentation faults are:

• nipy/labs/bindings/tests/test_numpy.py
• nipy/labs/glm/tests/test_glm.py
• nipy/labs/utils/tests/test_repro.py
• nipy/labs/group/tests/test_onesample_stat.py
• nipy/labs/spatial_models/tests/test_parcel_io.py
• nipy/labs/group/tests/test_permutation_test.py

No such segmentation faults on Numpy 1.26.4, which tests clean.

@alexis-roche - do you have any time to have a look at what's going on here with fff and Numpy 2.0 crashes?

Procedure is:

virtualenv nipy-env
. ./nipy-env/bin/activate
git clone https://github.com/matthew-brett/nipy
cd nipy
git checkout numpy-20-fixes
pip install meson-python ninja cython numpy pytest spin
spin install
pytest nipy/labs/bindings/tests/test_numpy.py

If it doesn't segfault first time, try running the pytest line a few times.

In case it's useful - Numpy 2.0 changes : https://numpy.org/doc/stable//numpy_2_0_migration_guide.html

I'm not getting segfaults, but I am getting:

corrupted size vs. prev_size
[1]    769785 IOT instruction (core dumped)  pytest -sv nipy/labs/bindings/tests/test_numpy.py

double free or corruption (out)
[1]    769607 IOT instruction (core dumped)  pytest -sv nipy/labs/bindings/tests/test_numpy.py

I don't have any insights into what's causing this.

Codecov Report

Attention: Patch coverage is 98.68421% with 1 line in your changes missing coverage. Please review.

Project coverage is 84.61%. Comparing base (3e8409c) to head (d8989ab).
Report is 40 commits behind head on main.

@@            Coverage Diff             @@
##             main     #565      +/-   ##
==========================================
- Coverage   84.70%   84.61%   -0.09%     
==========================================
  Files         297      293       -4     
  Lines       27590    27479     -111     
  Branches     3361     3362       +1     
==========================================
- Hits        23370    23252     -118     
- Misses       3266     3273       +7     
  Partials      954      954              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Given Py3.8 is end-of-life next week, might make more sense to just remove support and end testing on it.

Good call - dropped. Tests passing now.

Ah - that turns out to be because Python 3.8 pulls in Numpy < 2, which is the actual problem.

OK - added back 3.8 to trigger a Numpy < 2 build - and added some Numpy 1.2x fixes to the original Numpy 2.0 Nipy port. I think this is ready now. And it will fix a few of the outstanding issues.

Numpy issue with description, discussion, resolution of segfault: numpy/numpy#26765