Release GIL inside optimize?

[safijari]

Wondering if there any reason to retain the GIL in this function?

g2o/python/core/py_sparse_optimizer.cpp

Line 47 in e8d3ccf

.def("optimize", &CLS::optimize, "iterations"_a,

I'm running into a situation where some times an optimization can take a very long time (in the order of minutes) and because this call does not release the GIL it starves out some background threads meant for telemetry.

[RainerKuemmerle]

I have to admit that I did so far not think about this that much. Mostly used the Python binding so far to quickly prototype and visualize some data.

Reading the documentation (https://pybind11.readthedocs.io/en/stable/advanced/misc.html#global-interpreter-lock-gil) I believe we can release the GIL.

Would be great if you can test the branch and provide feedback here. Thanks.

[safijari]

Are wheels saved anywhere that I could grab and install?

[RainerKuemmerle]

Not for updated g2o. You'd have to build yourself. Probably best done in a python venv.

git clone --recurse-submodules https://github.com/miquelmassot/g2o-python.git
cd g2o-python/g2o
git remote set-url origin https://github.com/RainerKuemmerle/g2o.git
git fetch -tp
git co pymem-gil-optimize
cd ..
pip install -v .