Testing GitHub actions

[jameslamb]

No description provided.

[jameslamb]

looks promising: https://github.community/t5/GitHub-Actions/Powershell-error-output-not-shown/m-p/57949

[jameslamb]

copied from microsoft#3119 (comment) for documentation purposes

---

Ok I have some evidence of what I suspected to be true:

the way GitHub Actions powershell handles treatment of stderr is nondeterministic

I just pushed a commit that redirects the output of Rscript -e "install.packages(..." to $null. This is an extreme example of suppressing stderr.

Then I pushed 10 empty commits.

commit	build	MINGW build	MSVC build
link	link	✅	✅
link	link	✅	❌ (ERR1)
link	link	❌ (ERR2)	✅
link	link	❌ (ERR1)	❌ (ERR2)
link	link	❌ (ERR1)	✅
link	link	❌ (ERR1)	✅
link	link	✅	✅
link	link	✅	❌ (ERR1)
link	link	✅	❌ (ERR1)
link	link	❌ (ERR1)	✅

I really really think these "errors" are just "some code writing benign messages to stderr", and not actual failures of the calls. The two types of errors I encountered are shown below. The most common is what I've called ERR1, which comes from the use of message() inside install.packages(). I have not had success suppressing this output by using install.packages(..., quiet = TRUE, verbose = FALSE).

ERR1

ERR2

Given this finding, I'm sad to say that at this point, we should not move any of the Windows CI builds too GitHub Actions. It is too unreliable.

I'm planning to freeze this PR and use it in a post on the Github Actions Community Forum. If that post returns any useful information, or GitHub Support ever responds to the ticket I submitted, maybe we'll be able to add GitHub Actions jobs for Windows in the future.

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.