Check if tests are coupled by randomizing order in which tests are run

[IndrajeetPatil]

closes #1937

[MichaelChirico]

it looks like shuffle=TRUE is designed to make this trivial to test:

https://tidyverse.org/blog/2025/11/testthat-3-3-0/

[IndrajeetPatil]

Yes :)

r-lib/testthat#1942

[IndrajeetPatil]

The failing tests with testthat 3.3.0 are expected to fail (cf. #2937). The incompatibility needs to be resolved in a separate PR.

[IndrajeetPatil]

@MichaelChirico Took only 3 years but the builds are now passing 😆

[MichaelChirico]

could you run this locally, say, 10 times to guard against flakiness before we do the initial merge?

I'm also wondering if this should be run on every commit or just during release wdyt?

[MichaelChirico]

we don't technically need this since it's GHA-only requirement right?

[IndrajeetPatil]

We don't, but it's a good way to enforce the dev team to upgrade to use the latest version, which will become especially relevant if we run into a flaky test.

I know it's not ideal because they are forced to make this update globally, since we aren't using project dependency manager (like {renv}).

[IndrajeetPatil]

I'm also wondering if this should be run on every commit or just during release wdyt?

SGTM. Do we have a standardized release process? E.g. a branch name that always starts with rc-*?