Recommend .Rbuildignore the spellcheck test?

[MichaelChirico]

I think better than using NOT_CRAN to disable the {spelling} tests on CRAN would be to avoid shipping these tests to CRAN in the first place.

Maybe spell_check_setup() should prefer setting up a GHA workflow to do the spell-checking instead? Then backing up to tests/ if no appropriate CI tool is detected. That could also help trim {spelling} as a somewhat spurious dependency.

[jeroen]

I'm not sure I like the idea of limiting this to GHA. Many people don't use GitHub or CI, and still like to run this as part of their testing set.

Of course you are free to create a GHA workflow that performs a spell check independent of the package tests, or not remove these things in the shipped R package, if you want to avoid declaring a the dependency on test tooling, but I think this could hold for anything you want to test.

[MichaelChirico]

Right, that's why I suggested to look for .github/workflows first. And indeed I think maintainers should consider what tsts they actually want to ship to CRAN more carefully!

I took a random sample of 10 {spelling} downstreams,

• https://github.com/Bisaloo/authoritative
• https://github.com/epiverse-trace/finalsize
• https://github.com/vpparkkinen/frscore
• {pcds}: none, but on GitHub: https://github.com/elvanceyhan/pcds
• https://github.com/quanteda/quanteda
• {rzentra}: none
• https://github.com/matthiasgomolka/sdcLog
• https://github.com/MSKCC-Epi-Bio/tidycmprsk
• https://github.com/Trackage/trip
• https://github.com/matrix-profile-foundation/tsmp

80% are using GHA. I think recommending this path first is a good default.