README.md

Developing Jade

We use docker to setup the development environment for Jade. We need to build the docker image once; after that, we can iterate:

1. Develop Jade feature.
2. Build a python wheel of Jade.
3. Install the newly built wheel in a python envrionment (conda, virtualenv, e.t.c.).
4. Test the new feature.

You need to install docker (we assume that you are using Ubuntu). Follow the instruction. After the installation, you should be able to run sudo docker run hello-world.

Then you need to do the docker post installation config so that you can invoke docker command without elevating to root. The key steps are:

sudo groupadd docker
sudo usermod -aG docker $USER

To be sure, reboot after the post-installation config; after rebooting, you should be able to run docker run hello-world without sudo.

You need to ensure your glibc>=2.28. To check, run

ldd --version

Build the docker image

You only need to execute this once. In this directory, run

python bdocker_image.py

Alternatively, get a pre-built image

docker pull linfenglee/adamas-env:latest
docker tag linfenglee/adamas-env:latest adamas-env:latest

Build the python wheel

In this directory, run

python bwheel.py

The output wheel should be located at <repo root>/wheelhouse/nimblephysics-0.7.7-cp38-cp38-manylinux_2_28_x86_64.whl

Reuse CMake build tree

The bwheel.py script has one opton --cache. When supplied, it will not clean the previous CMake cache. This speeds up the build.

python bwheel.py --cache

Install the python wheel

In the this directory, with your own python environment activated, run

pip uninstall -y nimblephysics && pip install ../wheelhouse/nimblephysics-0.7.7-cp38-cp38-manylinux_2_28_x86_64.whl

The flag -I removes existing installed nimblephysics.

You can install other dependencies (e.g., torch) as usual.