Find redundant GHA and remove them

[MichaelChirico]

Lately I've found there's a lot of lethargy in our review process because of queueing in the GHA for the repo.

My suspicion is that many of the actions we run are totally redundant -- they provide signal that is completely identical to at least one other action. That is, they always share failure status with some other action. Such actions can be removed.

I will be extending this script to analyze this numerically:

https://gist.github.com/MichaelChirico/d5a1a0d864864e0d30f24e3eb87afecf

[IndrajeetPatil]

In my view, we can remove:

• R-CMD-check / ubuntu-latest (devel) zh_CN (this workflow is actually quite flaky and often fails unpredictably)
• test-package (redundant if we run R-CMD-check with options(warn = 2))
• R-CMD-check / windows-latest (4.1) (if this is only for checking with a newer Rtools version, isn't that also done via R-release?)

[MichaelChirico]

Hmm GHA API is doing something weird:

runs[, sum(conclusion == "failure"), by = name]
#                          name  V1
# 1:             check-link-rot   0
# 2:                    pkgdown   0
# 3:                       lint   0
# 4:                R-CMD-check 346
# 5:         check-all-examples   0
# 6:               test-package 173
# 7:           R-CMD-check-hard   0
# 8:              test-coverage   0
# 9: pages build and deployment   0

I'm fairly confident some of those other actions have failed! Especially we saw check-lint-rot fail when CRAN was down this week.

[AshesITR]

The zh_CN run was added to test a non-UTF 8 locale because of some encoding related bug I think.

[MichaelChirico]

Yea, I wouldn't remove zh_CN -- it has a clear purpose marking it distinct from the other runs. Flakiness can be addressed separately.

[IndrajeetPatil]

Gabor had checked this with R-devel after discussion in r-lib/actions#628, and they had also mentioned not to use the non-UTF8 locale.

If we choose to retain this, we shouldn't be running this workflow on R-devel, but maybe on R-oldrel. Otherwise, it's just going to fail sporadically.

[AshesITR]

Maybe switch to r-release?
Dropping non UTF-8 testing seems like a bad trade. We are also working with user files that might have any encoding, so we should test that it's handled correctly.

[MichaelChirico]

I haven't noticed this as much lately, or maybe I'm just used to it. I don't think there's anything to do here. The slowest GHA is usually the Windows one, which is kind of expected IME.