Segfault on Python versions >=3.9

[elkalt]

Describe the bug
CollisionMesh constructor segfaults on Python versions >=3.9.

To Reproduce
I used pyenv to get python versions 3.8.19 and 3.9.19 and ran the same test file (below) with both. Python 3.8.19 raised no errors while 3.9.19 gave a segfault.

from numpy import array
import ipctk

points = array([[-0.5, 0.5], [0.5, 0.5], [0.5, -0.5]])
edges = array([[0, 1], [1, 2], [2, 0]])
faces = array([[0, 1, 2]])

mesh = ipctk.CollisionMesh(points, edges, faces)

Environment:
Tested on both Debian Sid and macOS Monterey 12.2.1 with both raising the same issue

Additional context
I haven't tested any other functions, but I assume that there might be other issues as well.

[zfergus]

Thank you for opening this issue. I tried it on macOS with Python 3.11.7 and it works. I will install Python 3.9 and test it.

Can you provide me with some further information in the meantime?

• Which version of the toolkit are you using?
• How did you install the toolkit (e.g., pip install ipctk)?

[zfergus]

I installed Python3.9 and got the same segmentation fault 😞. I will try to determine the issue, but I recommend using a different Python version if possible, for now.

[elkalt]

I installed the toolkit using pip, and I'm using the latest version there (1.2 I believe). I was originally trying to use it with Python 3.12 on Debian and had the segfault as well (which I forgot to mention in the original post, sorry). I just retested with both Python 3.11.7 and 3.11.9 and got hit a segfault with both, so it might be somewhat platform dependent.

[zfergus]

Hmm, how strange. I will try to reproduce this on my end.

[zfergus]

I may have figured it out. It looks like the version of Pybind11 I am currently using (2.11.1) is incompatible with Numpy 2.0. I updated my Numpy to 2.0 and got a segfault with Python 3.11.

I plan to update Pybind11 in #100 which should fix the problem in main. I'll think about how to best address this in a way that fixes the Pypi installation, but for now, ipctk v1.2 is only compatible with numpy<2

[zfergus]

You can now install v1.2.1 with pip which contains a patch fix to use the latest pybind11. This should fix your issue without requiring you to update to our latest API.