#345 concerned an issue where git wasn't producing output quite as we expected. There are several tests in test_build.py and test_sdist.py which mock out the git command so they run independently of the system state. But it would be good to have an integration test which sets up a real git repository in a temporary directory, which we could add cases like #345 to.

This should cleanly skip itself if the git command is not available.