Renv in `quarto_publish.yaml`

[amyheather]

In commit 8420b0c, I finally got a GitHub action thay successfully builds the quarto book running Python and R.

I had tried directly installing the environment in an action then rendering, but it failed to use both environments.

I switched to a Dockerfile, and used that container when rendering. However, I found that it did not find the R environment, and I had to re-run renv::restore() in the GitHub action for it to finally run successfully.

It would be good to understand why this is the case, and whether there is any changes I can make to the Dockerfile or quarto_publish.yaml which would avoid us needing to re-run the restore, as that is an intensive and long thing to run, and I would ideally like to make the GitHub action quicker. It also seems a little redundant, having already run renv::restore() in the Dockerfile.

[amyheather]

Could try adding this to .Rprofile?

Sys.setenv(RENV_PATHS_LIBRARY = "/workspace/renv/library")

[amyheather]

Created a repository where I am experimenting with options here (wanted a minimal example to test, as this is too big - long time setting up environment etc. - to be playing around with).

https://github.com/amyheather/quarto_githubactions_python_and_r

[amyheather]

🎉 Successfully avoided needing to re-run renv::restore() in https://github.com/amyheather/quarto_githubactions_python_and_r.

How does it differ from des_rap_book?

Difference 1

DES RAP Book has these cells at start of quarto files:

#| include: false
library(reticulate)
use_condaenv("des-rap-book", required = TRUE)

Whilst the other repo takes a similar approach to the vidigi documentation, where it just has this in the title yaml:

engine: knitr

Difference 2

DES RAP Book installs chrome and chrome system dependencies as required by plotly.write_image().

This is not included in the other repo.

Difference 3

Before installing renv in the Dockerfile, the other repo adds this line:

# Set path to renv
ENV RENV_PATHS_LIBRARY=/workspace/renv/library

Comments

In other repo, did have issue that conda's matplotlib was not compatible with the base rocker image. The matplotlib in conda was built against a newer libstdc++ ABI than rocker/r-ver:4.4.1 so when reticulate imports it we get a CXXABI_1.3.15 mismatch. Attempting to resolve by forcing Python to use conda's libstdc++ at runtime.

I tried setting it to use conda's libstdc++ in the Dockerfile but that didn't work - in the end, I resolved by downgrading matplotlib to a version that doesn't require CXXABI_1.3.15.

TODO: Try again, setting up on a different base image.