Repo2docker: Add support for R 4.0

[nuest]

Proposed change

Add support for R 4.0 to the R version map at https://github.com/jupyter/repo2docker/blob/master/repo2docker/buildpacks/r.py#L71

Alternatives:

• Installation via conda, but there the latest version is also 3.6.
• Use holepunch and the rocker/binder image, though no 4.0.0 tag there either yet.

Alternative options

IMO: None.

Who would use this feature?

People demonstrating latest R features, or relying on latest packages requiring R 4.0, or trying to match their own environment closely.

How much effort will adding it take?

Not much, _if/when the version is available via the default Ubuntu sources.

Who can do this work?

Anyone with experience in running repo2docker tests and some R experience for testing.

[betatim]

We should remove/cut down most of the version support in https://github.com/jupyter/repo2docker/blob/master/repo2docker/buildpacks/r.py#L71 as it is broken. We built that on top of functionality which the creators of those R APT packages never intended :-/

My vote is to suggest using https://github.com/binder-examples/r-conda as the way to get different R versions installed. To get R 4.0 support someone needs to update the conda-forge package (which means repo2docker gets it "for free").

[scottyhq]

@betatim and @nuest - just noting that in order to install R4.0 from conda-forge with repo2docker I had to use the following environment.yml:

channels:
  - conda-forge
dependencies:
  - r-base>=4
  - r-irkernel>=1.1

Otherwise it appears repo2docker has an incompatible r-irkernel version pinned, as seen in the output from this build step:
Step 42/62 : RUN conda install -p ${NB_PYTHON_PREFIX} r-base=4 r-irkernel=1.0.2 r-devtools && conda clean --all -f -y && conda list -p ${NB_PYTHON_PREFIX}

[betatim]

Interesting. https://github.com/jupyter/repo2docker/pull/892 bumped the version of irkernel and I had thought that would fix it. Which version of repo2docker did this happen with?

[scottyhq]

Ah, thanks @betatim - i missed those recent changes since I was using the latest release locally (jupyter-repo2docker==0.11).

[betatim]

Following up my comment from May 2rd: https://stat.ethz.ch/pipermail/r-sig-debian/2019-November/003121.html is the message that made me think that we got lucky with using the different R versions from that PPA ever worked.