cc @lorenzwalthert -- any ideas on the R failures here?

yes, a new version of the package {rlang} got released on CRAN. Since binaries take sometimes a few days to arrive, this CI run tried a source installation, and failed because of some system requirements for compiling code. The problem should go away in a few days. Alternative is to re-write the test to not involve a package like {rlang} that has system requirements as build time dependencies.

Maybe a more general solution that would also serve the user (instead of just hacking a test) would be to always prioritise binary packages over source package when initializing an environment, even if the source package offer a newer version, e.g. with options(install.packages.compile.from.source = "never")

yes, a new version of the package {rlang} got released on CRAN. Since binaries take sometimes a few days to arrive, this CI run tried a source installation, and failed because of some system requirements for compiling code. The problem should go away in a few days. Alternative is to re-write the test to not involve a package like {rlang} that has system requirements as build time dependencies.

Maybe a more general solution that would also serve the user (instead of just hacking a test) would be to always prioritise binary packages over source package when initializing an environment, even if the source package offer a newer version, e.g. with options(install.packages.compile.from.source = "never")

since you know way more about this -- could you help with a PR to add that?