[ci] [R-package] add CI jobs covering more CRAN "additional checks", fix R_NO_REMAP warnings (fixes #6369)

[jameslamb]

Contributes to #6522
Fixes #6116
Fixes #6369
Replaces #6381 and #6276

This attempts to cover more of the CRAN "Additional Checks" (link) with the new rhub container images.

I originally started this just to try to get a CI job that could have caught #6522. But then realized that this could also be extended, with minimal extra complexity, to significantly improve the coverage of the CRAN checks.

So this PR now does the following:

• adds a build compiling with icc 🎉
• adds a build compiling with gcc 14 (this would have caught [R-package] CRAN warnings on gcc-14 #6522)
• replaces the custom clang-* builds added in [R-package] [ci] remove unnecessary include in linear_tree_learner (fixes #6264) #6265 with the container images provided by https://r-hub.github.io/containers/ (covering clang 16, 17, 18, and 19)
• guards defines of R_NO_REMAP, to avoid these compilation warnings:

./include/LightGBM/utils/log.h:19:9: warning: 'R_NO_REMAP' macro redefined [-Wmacro-redefined]
   19 | #define R_NO_REMAP
      |         ^
<command line>:8:9: note: previous definition is here
    8 | #define R_NO_REMAP 1
      |         ^
1 warning generated.

(ref: dmlc/xgboost#10461)

Notes for Reviewers

Happy to say that this would have caught #6522!

I saw exactly that WARNING in the build logs: https://github.com/microsoft/LightGBM/actions/runs/9803295968/job/27069256021?pr=6523

And that then went away once I merged in #6514 : https://github.com/microsoft/LightGBM/actions/runs/9803499820/job/27069753658?pr=6523

[StrikerRUS]

Could you please elaborate on this particular change?

[jameslamb]

R CMD check redirects the output of compiling the library + installing it to a separate file. To be able to always see those logs in CI, we cat out that file:

LightGBM/.github/workflows/r_package.yml

Lines 299 to 307 in 6662d55

	echo ""
	echo "install logs:"
	echo ""
	cat lightgbm.Rcheck/00install.out
	echo ""
	if grep -q -E "NOTE|WARNING|ERROR" lightgbm.Rcheck/00check.log; then
	echo "NOTEs, WARNINGs, or ERRORs have been found by R CMD check"
	exit 1
	fi

This || exit 1 prevents those lines from ever being reached.

Setting it to || true will allow them to run, and then we'd be relying on the check on NOTEs / WARNINGs / ERRORs to appropriately raise a non-0 exit code and fail the build.

[StrikerRUS]

Thanks, got it! I guess that NOTEs also cause exit code to be non-zero, that's why we cannot rely on it anymore like here

LightGBM/.ci/test_r_package.sh

Lines 200 to 214 in 6662d55

	# fails tests if either ERRORs or WARNINGs are thrown by
	# R CMD CHECK
	check_succeeded="yes"
	R CMD check ${PKG_TARBALL} \
	--as-cran \
	--run-donttest \
	|| check_succeeded="no"
	echo "R CMD check build logs:"
	BUILD_LOG_FILE=lightgbm.Rcheck/00install.out
	cat ${BUILD_LOG_FILE}
	if [[ $check_succeeded == "no" ]]; then
	exit 1
	fi

[jameslamb]

I'm really glad you asked about this, because I realized some of the assumptions being made in this existing code are not correct.

I tested today with R 4.3.3 and found that NOTEs and WARNINGs do not lead to a non-0 exit code for R CMD check. ERRORs do.

So I think we do still want something like this ||, to be sure that you can see all the build logs (content of 00install.out) in CI logs for a build where R CMD check found ERRORs.

To avoid duplicating that in multiple places, I just pushed commits moving that into a new script, .ci/run-r-cmd-check.sh. In that script, I made "allowed number of NOTEs" parameterizable, so we can continue not allowing any NOTEs in most CI jobs, but selectively allow 1 in these new jobs.

[jameslamb]

This was a fix found in #6501. That PR was abandoned accidentally and I'd still like to give @Kunal-Singh-Dadhwal an opportunity to submit that work again.

But I want this particular fix, to get more confidence that the R CI jobs are working as expected.

context: #6501 (comment)

[Kunal-Singh-Dadhwal]

@jameslamb Thanks for the opportunity will make a commit really soon

[StrikerRUS]

Thanks for rethinking this PR!

LGTM except two minor comments below.

[StrikerRUS]

🚀