Description
It seems that a generic "pip install gtsam" does not install compatible requirements. In particular, "numpy" gets installed and by default is a new version (e.g., 2.0.0) that seems to not be compatible with the C++ binary that the python bindings are linking to. Downgrading "numpy" (e.g., to 1.26.4) seems to fix the issue, but this is not obvious and should either be documents or specified in "requirements.txt" so a compatible version gets automatically installed.
Steps to reproduce
- pip install gtsam (in a python 3.9.6 venv - this automatically installed numpy 2.0.0 for me)
- run any example - fails in a segmentation fault
- downgrade numpy to 1.26.4 and the examples work as expected
Expected behavior
Examples should run properly without having to figure out that the wrong numpy gets installed.
Environment
python 3.9.6 in venv
Additional information