choice of GitHub Actions runners

[wlandau]

@shikokuchuo, I remember you expressed a preference for Mac OS runners for GitHub Actions early on. Do you remember why? For a scenario this lightweight, I think ubuntu-latest may be more minimal.

[shikokuchuo]

I don't actually remember. The only reason I can think of is to take advantage of MacOS binaries rather than building packages from source. But personally I don't use them - on one project I use Windows binaries on a cron job. So happy to go with Linux.

[szhorvat]

adds igraph as a dependency, which has a long and brittle compilation

As for long compilation, ccache can help tremendously. Parallel build timings on my Apple M2 machine:

Executed in   20.41 secs    fish           external
   usr time   48.34 secs    0.08 millis   48.34 secs
   sys time    7.85 secs    1.23 millis    7.85 secs

As for brittle compilation, things got a bit more difficult in 2.0.0, but all issues should have been resolved by 2.0.3. If you see a reproducible problem, please report it. In my experience, the majority of issues people see come from having a conda environment active at the time of compilation, which can cause all sorts of problems, and not just for igraph. That shouldn't be a problem for R-multiverse.

[shikokuchuo]

Thanks for the note. Personally I've not come across any issues, but I'm operating in a pretty standard Linux environment. igraph is doing wonders for our R-Multiverse infrastructure though from the work @wlandau has done so far.

[wlandau]

Yes, R-multiverse needs to quickly traverse large dependency graphs, and igraph is excellent. It is also an invaluable piece of targets.

We will keep ccache in mind and report reproducible compilation issues if we encounter them going forward.

[wlandau]

I noticed all our Mac runners began failing due to r-lib/actions#948 / r-lib/actions#950, and I am converting them to ubuntu-latest to put out this fire. As it turns out, it is super easy to use p3m for Ubuntu binaries. All we need is:

      - name: Install R.
        uses: r-lib/actions/setup-r@v2
        with:
          extra-repositories: https://community.r-multiverse.org
          use-public-rspm: true

and then

      - name: Install helper R package.
        run: install.packages("multiverse.internals")
        shell: Rscript {0}

and the action installs binaries and runs super fast. I'm recommending we stick with this approach long-term.

[shikokuchuo]

Sounds good.