My employer produces a Windows desktop app and I maintain the GitLab CI and develop product installers using InstallAnywhere, among other things. We have a new product so I developed a functional installer for it. Our other products rely on a DLL library of ~6.5 GB called spaceLib which sits in its own repo, and the new software relies a subset of this library (reduced to 200 MB), which is where my problem stems.

Our other software uses everything in spaceLib, so my predecessor configured the other installer project repos with spaceLib as a submodule. Due to redundant maintenance of spaceLib, file size, possible errors in maintenance (I might miss a file update to spaceLib in my installer project repo), and the fact that I only need a fraction of the files, I first decided I’d perform a shallow clone of the spaceLib subset into my new installer project when GitLab builds my installer but this results in empty files (1kB).

I wrote a batch script and use a file list to handle this:

@echo off
setlocal

echo CD is %CD%

:: Set vars
echo Setting environment variables...
set SCRIPTS_DIR=%~dp0
set BRANCH=dev
set SPACELIB_REPO=my.gitlab/installers/dependencies/spaceLib.git
set SPACELIB_DIR=windows64spaceLib
pushd source

:: Clone empty spaceLib repo
echo Cloning empty spaceLib repository...
git clone --branch %BRANCH% --no-checkout http://%SPACELIB_REPO% %SPACELIB_DIR% || goto GIT_ERROR
pushd %SPACELIB_DIR%
echo CD is %CD%
:: Restore necessary spaceLib files
echo Restoring necessary spaceLib files...
for /f %%F in (%SCRIPTS_DIR%spaceLib_components.txt) do (
    echo Restoring file %%F...
    git restore --staged %%F || goto GIT_ERROR
    git restore %%F || goto GIT_ERROR
)
popd

popd
echo Cloning partial repositories successfull...
goto END

:GIT_ERROR
echo Failed one of the Git operations...
exit /b 1

:END
endlocal

When the script runs, spaceLib is cloned into my installer project repo in sourcewindows64spaceLib but all my files are empty (1kB). I started testing on my end and found the --no-checkout option causes this. Why would the --no-checkout option cause git restore to produce empty files?

To my understanding, git clone will fetch everything from the remote > download the git objects > rebuild the files in the index using the downloaded git objects > rebuild the files in the working area using what is in the index. Adding the --depth=1 option will only fetch the latest commit. Adding the --no-checkout option will fetch everything from the remote > download the git objects and leave the index and working area empty. And running git restore on each file in spaceLib_components.txt would first restore my index files to match my remote, and then restore my working area files to match my index. I originally thought git was unable to reconstruct my files because of the --depth=1 restriction, but even after increasing the number of commits my files are still appearing as empty. If I run the command git clone --branch %BRANCH% http://%SPACELIB_REPO% %SPACELIB_DIR%, I get my files as desired but end up with all the spaceLib files. I’m not sure if I’m misunderstanding how --no-checkout works or how git restore works.

I’ve also read other posts on SO and see different suggestions on other possible solutions, I might use the last option for my case but don’t like that it can introduce more errors:

• git clone --depth=1 <URL> <DIR> to create a shallow clone (pulls all files)
• git clone --depth=1 --no-checkout <URL> <DIR> to create a shallow clone without any files, then git checkout pathtofile to restore my file (results in empty files when I try this)
• Create an empty git repo locally, set the remote origin, run git config core.sparseCheckout true, add the file subset I want to clone to .git/info/sparse-checkout, and run git pull