Add datascience-python-r-notebook

[massongit]

fix #1551

I add datascience-python-r-notebook, which is a multi-architecture stack based on jupyter/datascience-notebook without Julia.
If it add, users who do not use Julia will be able to use jupyter/datascience-notebook on M1 Mac and latest Docker Desktop for Mac.

[mathbunnyru]

Also https://github.com/jupyter/docker-stacks/blob/master/docs/using/selecting.md#image-relationships has to be upgraded.

To be honest, I don't think adding new image is what we should do.
I do understand, that you want the image to work under arm, but the better approach would be to fix our build system to use native arm runners.
It's difficult, but it's worth it.

[mathbunnyru]

Taggers and manifests are inherited, so you only need to add julia here.

[mathbunnyru]

Links seem to be wrong.

[mathbunnyru]

Let's keep it sorted.

[consideRatio]

I'm strongly opposed to adding another image to this repo. If we should do that, I think we should first have clear guidelines on when and why we do decide to add another images. My opposition stems from a care of this projects long term maintainability, not that the addition has no value for an end user.

I've seen several open source projects run into trouble for accepting more and more variations of something when there are infinite amounts of variations possible. I think they all need to say stop after a while as it becomes unsustainable to keep adding new parts.

@romainx @mathbunnyru what do you think?

[romainx]

@consideRatio and @mathbunnyru agree for the past years the answer for new images--even if they make sense and are interesting--was

We've been requesting that all new image definitions take the community images path, primarily because this repository is unwieldy and we're already finding it difficult to maintain the images that already exist here. It sounds like you already have a nice project setup for building the images. Would you be open to submitting a pull request linking to it from the documentation instead of this PR?
-- #1023

So I think we are in this case here and moreover this image is more a workaround to fix issues with ARM and Julia than a new features based image.

[massongit]

Thank you for your reviews!
I agree that this approach is not good for maintainability.
I would like to cooperate as much as possible to find to install Julia under ARM.

[benz0li]

@massongit You may be interested in https://gitlab.b-data.ch/jupyterlab/r/docker-stack, which combines https://github.com/rocker-org/rocker-versioned with https://github.com/jupyter/docker-stacks (± minimal-notebook) including code-server and further customisations.

See also https://demo.jupyter.b-data.ch for a minimal deployment (CPU: 2 cores, RAM: 4 GB) based on https://gitlab.com/b-data/docker/deployments/jupyter serving either registry.gitlab.b-data.ch/jupyterlab/r/verse:latest-root or registry.gitlab.b-data.ch/jupyterlab/julia/ver:latest-root.
ℹ️ Both JupyterHub and JupyterLab images are multi-arch (linux/amd64, linux/arm64/v8), the latter Debian-based.

Both JupyterLab images have stock Python (currently v3.9.2) installed and focus on Julia and R, respectively. Thus, apart from jupyter-server-proxy jupyterhub jupyterlab jupyterlab-git notebook nbconvert [radian] and its dependencies, no further Python packages are installed.