post-render R script can't find renv installed R library when used in a Github action

[yulric]

I have:

• searched the issue tracker for similar issues
• installed the latest version of Quarto CLI
• formatted my issue following the Bug Reports guide

Bug description

I have an R project with a Quarto project within it. The R project uses renv to manage its dependencies. The Quarto project uses the post-render option to run an R script that adds a file to the output directory. The Quarto project has no issues rendering locally. However, when running in a Github action it gives an error in the post render script saying it can't find the R libraries it uses.

Steps to reproduce

This repo has the code to reproduce the issue https://github.com/yulric/quarto-issue.

Actual behavior

The Github action fails at the last step that runs the quarto render command.

Expected behavior

The Github action should succeed.

Your environment

It looks like the action is running Ubuntu version 24.04.2

Quarto check output

The failing build is using quarto v1.7.32

[cscheid]

I'm fairly sure this isn't a quarto bug; Quarto itself has no say about where your R libraries are installed.

I'm going to convert this to a discussion.

[mcanouil]

@yulric you should add some checks to verify where your command is running and if it does indeed use your renv project for example by adding a non R-base package to your renv.lock and use that in the script. "work locally" could be only because the R packages are available globally.

Note that Quarto's design is to have everything inside the project but you have chosen to put things outside of it which can lead to unexpected issues.

[mcanouil]

As a "workaround", in your project script @yulric you can add:

setwd("../")
source(".Rprofile")

[cderv]

The question now is: should pre/post-render be executed from where the quarto render command is called?

I don't think this should be changed by default. This is the current design, and this is documented
https://quarto.org/docs/projects/scripts.html#pre-and-post-render

Pre and post-render scripts are run with the main project directory.

Changing it would be a breaking change and would also make the problem harder the other way around (calling in sub dir but main renv is at project root).

As @mcanouil said

Note that Quarto's design is to have everything inside the project

and pre/post render script are project features, so having the Quarto project as the root seems to be the correct logic for pre/post render so that this is standard for everyone.

However, maybe adding a config to change the working directory for Project script can be useful 🤔
Not changing the default but allowing to pass a working dir.

project:
  type: website
  pre-render: 
  	script: make prepare
	working-directory: ../

or

project:
  type: website
  pre-render: make prepare
  pre-render-wd: ../

Just idea to offer configuration.

It can easily be tweaked from script using custom command like your shared @mcanouil.
Regarding renv usage, a note its functions supports a project = argument, so there may be ways to work with configuration

[cscheid]

The question now is: should pre/post-render be executed from where the quarto render command is called?

I don't think this should be changed by default. This is the current design, and this is documented
https://quarto.org/docs/projects/scripts.html#pre-and-post-render

I agree with Christophe here.

[yulric]

@mcanouil Thanks for your help with this, your fix worked!

Regarding the other comments made, from a user/black-box perspective its odd to me when R code within a Quarto document works but an R script within post-render fails due to a dependency issue. My assumption would have been that it all uses the same infrastructure to run R code so that either both should fail or succeed re dependency issues.

[cderv]

Regarding the other comments made, from a user/black-box perspective its odd to me when R code within a Quarto document works but an R script within post-render fails due to a dependency issue. My assumption would have been that it all uses the same infrastructure to run R code so that either both should fail or succeed re dependency issues.

This is the problem of being too clever: When calling quarto render quarto in your project root, Quarto will see the .Rprofile and the renv activation in it. So it will use this root directory to start R, and then in R itself, use setwd() to be in the right place.

In pre/post render, it is harder to be that clever, and it is not the user running pre/post, but the project process. So, a clearer situation would be to have both fail. Only having renv activation at the root of the Quarto Project could work. But if we change that, we break some usage. Hard choice...

Sometimes trying to be too helpful have side effect of being less obvious for more advanced usage. 🤷‍♂️