WIP: [ci] fix R 3.6 Windows build (fixes #5036)

[jameslamb]

Attempting to investigate #5036.

First, I'm pushing a commit just re-running all the R CI to see if the same error shows up.

After that, I'll start investigating to try to find the root cause of new failures, like making logs more verbose.
This might be a slow process involving a lot of just printing this to logs, since https://github.com/nelsonjchen/reverse-rdp-windows-github-actions is no longer available.

[jameslamb]

I see the same error on https://github.com/microsoft/LightGBM/runs/5353387904?check_suite_focus=true as reported in #5036.

Pushed e93cb85, adding more logs to try to understand what happened.

[jameslamb]

I think that we have tar issues again with Windows builds. See the third item in #3946 (review) for context.

In the logs for the most recent build (link), I see the following

removing object files created by vignettes
untarring lightgbm_3.3.2.99.tar.gz
done untarring lightgbm_3.3.2.99.tar.gz
re-tarring lightgbm_3.3.2.99.tar.gz
Running R CMD check
* using log directory 'C:/tmp-r-cmd-check/lightgbm.Rcheck'

Note that the line "Done creating lightgbm_3.3.2.99.tar.gz" from that section of build-cran-package.sh is not shown.

LightGBM/build-cran-package.sh

Lines 207 to 221 in 01568cf

	echo "re-tarring ${TARBALL_NAME}"
	tar \
	-czv \
	-C ./_tmp \
	--exclude=*.a \
	--exclude=*.dll \
	--exclude=*.o \
	--exclude=*.so \
	--exclude=*.tar.gz \
	--exclude=**/conftest.c \
	--exclude=**/conftest.exe \
	-f "${TARBALL_NAME}" \
	lightgbm \
	> /dev/null 2>&1
	echo "Done creating ${TARBALL_NAME}"

I just pushed 2c0e362 to try printing the locations of tar, gzip, and gunzip.

That commit also reverts e93cb85... I realize now that e93cb85 wouldn't have helped anyway, since it only affects CMake-based builds and the problem is related to CRAN builds.

[jameslamb]

try printing the locations of tar, gzip, and gunzip.

For the failing R 3.6 CRAN build (link), I see

--- location of tar ---
CommandType     Name                                               Version    Source
-----------     ----                                               -------    ------
Application     tar.exe                                            0.0.0.0    C:\Rtools\bin\tar.exe
--- location of gzip ---
Application     gzip.exe                                           0.0.0.0    C:\msys64\usr\bin\gzip.exe
--- location of gunzip ---
Application     gunzip                                             0.0.0.0    C:\msys64\usr\bin\gunzip

For the succeeding R 4.x CRAN build (link), I see that all three utilities are coming from Rtools

CommandType     Name                                               Version    Source
-----------     ----                                               -------    ------
Application     tar.exe                                            0.0.0.0    C:\rtools40\usr\bin\tar.exe
--- location of gzip ---
Application     gzip.exe                                           0.0.0.0    C:\rtools40\usr\bin\gzip.exe
--- location of gunzip ---
Application     gunzip                                             0.0.0.0    C:\rtools40\usr\bin\gunzip

I wonder if the root issue is some combination of:

• mixing different builds of those tools can cause problems
• some recent change to the windows-latest environment in GitHub actions remove C:\msys64\usr\bin\tar.exe

I found that some other utilities provided by the MSYS2 distribution used in these images are available at C:\msys64\mingw64\bin.

https://github.com/actions/virtual-environments/blob/3b5c4ebd39ce3d5812e130938bb066f67d90b54e/images/win/scripts/Installers/Install-Msys2.ps1#L75-L76

I just pushed 3e1d5f8, to see if adding that to PATH helps.

[jameslamb]

I tried just using the tools bundled with Rtools35 (by ensuring that Rtools paths are at the beginning of PATH) and unfortunately there is not a gunzip there.

(failing R 3.6 CRAN build link)

CommandType     Name                                               Version    Source
-----------     ----                                               -------    ------
Application     tar.exe                                            0.0.0.0    C:\Rtools\bin\tar.exe
--- location of gzip ---
Application     gzip.exe                                           0.0.0.0    C:\Rtools\bin\gzip.exe
--- location of gunzip ---
Get-Command: D:\a\LightGBM\LightGBM\.ci\test_r_package_windows.ps1:180
Line |
 180 |      Get-Command gunzip
     |      ~~~~~~~~~~~~~~~~~~
     | The term 'gunzip' is not recognized as a name of a cmdlet, function, script file, or executable
     | program. Check the spelling of the name, or if a path was included, verify that the path is correct
     | and try again.

There is one in newer versions of RTools.

(successful R4.X CRAN build link)

CommandType     Name                                               Version    Source
-----------     ----                                               -------    ------
Application     tar.exe                                            0.0.0.0    C:\rtools40\usr\bin\tar.exe
--- location of gzip ---
Application     gzip.exe                                           0.0.0.0    C:\rtools40\usr\bin\gzip.exe
--- location of gunzip ---
Application     gunzip                                             0.0.0.0    C:\rtools40\usr\bin\gunzip

[jameslamb]

Ok I was able to get a few more logs from the failing line. (build link)

re-tarring lightgbm_3.3.2.99.tar.gz
lightgbm/
tar (child)lightgbm/build/
lightgbm/build/vignette.rds
lightgbm/cleanup
lightgbm/configure
: gzip: Cannot exec: No such file or directory
tar (child): Error is not recoverable: exiting now
tar: lightgbm_3.3.2.99.tar.gz: Cannot write: Broken pipe
tar: Child returned status 2
tar: Error is not recoverable: exiting now
Running R CMD check

That error message is what I'd expect to see if gzip wasn't installed at all, but I can tell from the logs of Get-Command that it is!

I just pushed ab59cfb adding -z to the original tar command (based on this very old Stack Overflow answer, https://stackoverflow.com/questions/9749466/cant-untar-a-complete-directory-using-tar-cvpzf#comment12404275_9749491), but that did not work.

I'm traveling right now and will be away for the next 8 hours or so, sorry I'm not able to devote more time to this right now.

[guolinke]

Hi, @jameslamb is the CI job normal now?

[jameslamb]

is the CI job normal now?

@guolinke I will come back to this draft PR once LightGBM's CI is fixed (#5362 (comment)). I had been prioritizing other things, like getting R 4.2 support added (#5274).

[jameslamb]

to make things easier for reviewers, I'm closing this PR with lots of debugging commits and comments. It is replaced by #5479.

[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.