Any way we can document this new behavior with a test?

I took some influence from the diff::submodules test and the rest of the stash tests, and kinda smashed them together. Let me know if that works. 🔍

I like it, thanks! 🎇 Anybody else have an opinion?

The janky build on Windows reported a failure in the tests. Has it been checked out?

stash::submodules::does_not_stash_modified_submodules [c:\program files (x86)\jenkins\jobs\libgit2-win32-msvc-f5ec724b4d32\workspace\tests-clar\clar/fs.h:76]
Expression is not true: DeleteFileW(buffer)

Looks like a file access issue in Windows. What's the janky test exactly? We should re-run it.

It's this one https://janky.rs.github.com/634985/output which I think is the only one that's been run. No idea if it's just the CI being dirty.

Yeah, I've no idea what that failure means. If it's a problem in the PR, I'm happy to fix it, but I'd need some pointers. :\

👆 Proposed fix to the win32 errors. Basically, the cleanup method being injected by using setup_fixture_submodules was running before test_stash_submodules__cleanup could close down the submodule repo it had opened.

This patch makes the test code a bit noisier, but ensures that those resources are freed before the files are deleted. @jspahrsummers, what do you think?

@ben Thanks!

👍 from me.

Looks perfect. Thanks for looking into this @ben!