Fortran kernel for Jupyter with Coarray support

---

Quick start

• Try Coarray Fortran in the cloud with GFortran and OpenCoarrays using Binder: http://bit.ly/TryCoarrays
  This will open the index.ipynb file in an interactive environment using Binder.
• Launch a full JupyterHub Binder Fortran environment in the cloud with OpenCoarrays: http://bit.ly/CAF-Binder

Run locally

docker pull sourceryinstitute/jupyter-caf-kernel:latest
docker run -i -t -p 8888:8888 sourceryinstitute/jupyter-caf-kernel:latest

You should see output similar to the following which will instruct you to point your web browser to http://localhost:8888/?token=...

[I 19:29:08.089 NotebookApp] Writing notebook server cookie secret to /home/jovyan/.local/share/jupyter/runtime/notebook_cookie_secret
[W 19:29:08.188 NotebookApp] WARNING: The notebook server is listening on all IP addresses and not using encryption. This is not recommended.
[I 19:29:08.221 NotebookApp] JupyterLab alpha preview extension loaded from /opt/conda/lib/python3.5/site-packages/jupyterlab
[I 19:29:08.233 NotebookApp] Serving notebooks from local directory: /home/jovyan
[I 19:29:08.233 NotebookApp] 0 active kernels
[I 19:29:08.233 NotebookApp] The Jupyter Notebook is running at: http://[all ip addresses on your system]:8888/?token=67e0bf26a40f863c8af93c030de57b4802cd7b3d9f10ae17
[I 19:29:08.233 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[C 19:29:08.234 NotebookApp]

    Copy/paste this URL into your browser when you connect for the first time,
    to login with a token:
        http://localhost:8888/?token=67e0bf26a40f863c8af93c030de57b4802cd7b3d9f10ae17

When running locally (this way) you won't have access to the examples folder or index.ipynb.

About

This Coarray Fortran kernel is still a bit rough around the edges. It is a shameless fork of the jupyter-c-kernel, quickly thrown together. This is licensed under the MIT license, as is jupyter-c-kernel. I also drew inspiration from the jupyter-fortran-kernel.

Whats up with the prebuild directory?

The prebuild directory is there so that the docker image used to launch the Coarray Fortran kernel on Binder can be built automatically on Docker Hub. The Dockerfile at the project root then wraps the one located in prebuild. This is because we must compile GCC from source to get a new enough version and this takes a very long time. We don't want users to have to wait forever to spawn a new notebook on Binder, so we ensure that the base image is built ahead of time, locally, or through the Docker Hub automated build process.

Manual local installation

The recommended usage of this kernel is through Binder in the cloud, or via Docker, as described above. If you decide to install this kernel locally on your system you will need the following software packages installed:

• GCC and GFortran >= 8.2
• MPICH >= 3.2
• OpenCoarrays
• Jupyter
• Python 3
• pip

Step-by-step

1. git clone https://github.com/sourceryinstitute/jupyter-CAF-kernel
2. cd jupyter-CAF-kernel
3. pip install -e prebuild/jupyter-caf-kernel
4. jupyter-kernelspec install prebuild/jupyter-caf-kernel/Coarray-Fortran/
5. jupyter-notebook .

Contributing

Contributions are most welcome!

ToDo

• Use proper Python OOP class extension & better understand base class
• Improve I/O
• Improve exception handling and error reporting
• Improve magics
• Use MetaKernel to build the kernel and magics
• Allow use of all aplicable MetaKernel magics
• Don't attempt to compile code when only magics are present
• Different flavors of Docker images (so that less space is used). See jupyter/docker-stacks

Version control

1. Fork
2. Create branches named issue-X where X is the number of the issue.
3. Commit changes
4. Pull Request

License

MIT