GLCI fixes

[jangorecki]

towards #5727

• removed 3.4.4, 3.5.0 test jobs
• updated urls to windows R binaries
• cleanup old comments
• remove docker builds

[jangorecki]

if we merge it to master before hotfix release, we can revert #5719 because #5718 will be solved at the root. Important to note that PR does not touch any files that are being part of the package so is safe to merge.

[tdhock]

hi Jan it looks ok to me, but I am not really familiar with this, so not sure if I would be able to catch any issues.

[MichaelChirico]

I would start (e.g.) .ci/scripts/ and add .ci/scripts/test-install-deps.R with:

source(".ci/ci.R")
install.packages(
  dcf.dependencies("DESCRIPTION", which="all"),
  repos=file.path("file:", normalizePath("bus/mirror-packages/cran", mustWork=FALSE)),
  quiet=TRUE
)

This will be more readable and avoid duplication (and hence double maintenance/risk of getting out of sync) across test-intall-deps and test-install-deps-win jobs.

[MichaelChirico]

Or, we can find a way to put RUN_SCRIPT=if(win) Rscript.exe else Rscript into env and unify the jobs.

(just noticing the other small diff on Windows getwd(), so it's probably best to write this into the suggested script)

[jangorecki]

Quoting arguments from shell also differs

[jangorecki]

I would start (e.g.) .ci/scripts/ and add .ci/scripts/test-install-deps.R with:

source(".ci/ci.R")
install.packages(
  dcf.dependencies("DESCRIPTION", which="all"),
  repos=file.path("file:", normalizePath("bus/mirror-packages/cran", mustWork=FALSE)),
  quiet=TRUE
)

This will be more readable and avoid duplication (and hence double maintenance/risk of getting out of sync) across test-intall-deps and test-install-deps-win jobs.

Unless we will need distinction for some reason, then we ended up having two scripts :)

[MichaelChirico]

Quoting arguments from shell also differs

Can't we unify if there's no env inside "? Anyway, if the shell command is just Rscript -e "source(...)" it will still be easier to glance & see it looks ~OK.

Unless we will need distinction for some reason, then we ended up having two scripts :)

Looks like not the case yet, we can do some .Platform/Sys.info() switching to handle the small differences. Still almost all of the logic between the two jobs is the same, I find it more confusing needing to read long one-line scripts without syntax highlighting.

[MichaelChirico]

lin is linux here? I would just write out linux I don't see much benefit to saving 2 characters

[jangorecki]

Yes, it is 3 letters so it aligns better when printing, to win and mac. Maybe lnx then?

[MichaelChirico]

That would be less ambiguous, yes

[MichaelChirico]

hi Jan it looks ok to me, but I am not really familiar with this, so not sure if I would be able to catch any issues.

Ditto, basically LGTM but I don't follow 100%. It's just CI though, issues should be easy to detect/fix later, effects on package code are second-order only (e.g. if a mistake in CI eventually lets a mistake pass through PR review unnoticed)

[jangorecki]

These CI runs on master so after PR being merged already. Design was made that ghci gives fast feedback and let us do merge, while glci tests thoroughly and we don't have to wait for it.