Trouble skipping R installation in setup-r GitHub Action v1

[lianos]

My package development related GitHub actions that use r-lib/actions/setup-r started breaking a few weeks ago.

After a bit of sleuthing, I've been able to identify that this introduction of the install-r variable to the action is the reason why.

I think if I can successfully set install-r to false, things would work again for me, but I'm having trouble getting that to stick.

The relevant portion of a GitHub action definition where I try to set install-r to false, but it's not sticking looks like this:

      - uses: r-lib/actions/setup-r@master
        with:
          install-r: false

For now I've pinned my dependency on the setup-r action to the commit right before this change, as shown below:

      - name: Setup R
        uses: r-lib/actions/setup-r@2fac31217e602dd30286b4c984a0ccf1afb18cbd

This works, but I'd like to get it back to using @v1 or @master of the setup-r action, if possible.

Am I not setting install-r to false correctly in the yaml file, or is there something else going on?

Some other bits of things I can think of that might be relevant:

1. These actions are set to run inside a Docker container which has all of the system dependencies, R, and R package versions already setup so I can avoid long GitHub action runtimes.

2. The links in this post point to do different, but largely similar, actions on the same GitHub repository, just to show that one of them works when pinned to setup-r@2fac3 and the other doesn't.

Thanks for any help!
-steve

[jimhester]

I am not sure why this would be happening. It would be useful to see a build log of what the error message you are seeing is.

[lianos]

Hi Jim,

Thanks for the response. Is the build log something other than the log for the action itself?

If that's what you're after, here are some links that I think you're after:

1. The top-level result for the failed action itself. It's running out of time in the action since it's trying to build all the underlying R mojo, as opposed to skipping it and just using what is setup in the container that it's running in.
2. This is the log page itself.
3. This is the workflow file

Is that what you're after?

In the meantime, I've been meaning to setup a test repository for a minimal R package that then has R-CMD-check actions that use a more standard image, like rocker/shiny-verse and see if the install-r: false sticks there ...

[jimhester]

I was able to reproduce the issue, it should be fixed by https://github.com/r-lib/actions/commit/08590c77b4f4c84ecaf8637a1c92cba2d3b3b146.

Sorry for the trouble, thank you for posting!

[lianos]

Awesome, thanks for smoking that out so quickly.

I pinned my the setup-r action to the commit you referenced here, and can confirm that the fix now works as expected.

Thanks for all of your work on developing these actions, they are a huge help.