Segmentation fault (core dumped) when running example simulation

[ncr292]

Hello,
I am trying this awesome package but I run into some small problems which I wanted to report.
Emerge 1.0.3 crashes with a segmentation fault when running the example simulation, in particular when generating the mesh.

Steps to reproduce:
conda create -n emerge
conda activate emerge
conda install -c conda-forge pip
pip install emerge
Then copy-paste the first example in the user manual -- where btw there is a typo in line 23, missing capital "w". Eventually
python example.py
gives
[INFO] Setting frequency as [...] [INFO] Adding PEC boundary condition with tags []. Segmentation fault (core dumped)

Environment:

Emerge 1.0.3
Python 3.13.7
NumPy 2.2.6
Numba 0.61.2

OS: Linux (Ubuntu 24.04.3 LTS x86_64)

Notes
The crash happens on model.generate_mesh(), and it is almost instantaneous, no system resource is being consumed. Both libopenblas-dev and libsuitesparse-dev are installed system wide.

Temporary solution
Downgrading to Python 3.11 with compatible NumPy/Numba resolves the issue.

[FennisRobert]

Hey! Thanks for sharing this information. If possible (the environment still exists), can you set the option write_log=True in your Simulation object function and share the generated log file by some means that is placed in a newly created subdirectory. Maybe I can figure out at what point GMSH is crashing.

[FennisRobert]

I am suspecting GMSH to be the culprit. Can you also share the GMSH version? I might have to tighten the dependency requirements a bit.

[ncr292]

Of course and you might be right. So, GMSH version is 4.13.1 and you can find the log here.

[FennisRobert]

Hmm complicated. I don't have access to a linux machine myself so its hard to reproduce. You can try forcing an older version of GMSH although the allowed version range currently isn't that big. It may be something with the Linux build. If you find out more about it, please let me know

[damor-rbz]

Same issue on my side, I checked with python 3.11 and had to downgrade numba to version 0.60.0 to make it work.

Also checked with python 3.12 and same, just downgrading numba to version 0.60.0 makes it work.

[FennisRobert]

Ok interesting, also on Linux? Where along the process did you get segmentation faults? During meshing or somewhere else?

[damor-rbz]

Yes, under ubuntu linux 24.04, and also during the meshing phase

[FennisRobert]

Yes, under ubuntu linux 24.04, and also during the meshing phase

Ok and if you lower the Numba version, GMSH no longer crashes? That is actually fascinating. I'll do my best to figure out what is going on. I'm still learning more and more about dependency management and mutual compatibility. Any suggestions for resolutions are welcome!

Appreciate reaching out and raising this issue :)

[damor-rbz]

Yes, I see no problem once I lower numba, I'll try to make some tests on different versions

[FennisRobert]

Can you try running the following two fixes by adding either of the sections before importing EMerge?

Option 1:

import os
os.environ.setdefault("NUMBA_THREADING_LAYER", "workqueue")

import emerge as em

Option 2:

import gmsh

import emerge as em

The hypothesis is that there is some issue with the binaries that run parallel OpenMP runtimes which cause this issue. Either or both options should fix it. If so then I can add these as default in EMerge :)

[damor-rbz]

Tried both and none worked.

[FennisRobert]

That is unfortionate. I hope running it on 3.11 is ok for now! I'll get back to it at some point if I can test it out on Linux.

[damor-rbz]

sure, no problem, will keep testing in future versions.