Under Windows, when running test-cases creating different temporary repositories consecutively, I get sporadic failures when executing the git process:

 File "D:\Apps\WinPython-64bit-3.5.2.1\python-3.5.2.amd64\lib\site-packages\git\repo\base.py", line 872, in init
    git.init(**kwargs)
  File "D:\Apps\WinPython-64bit-3.5.2.1\python-3.5.2.amd64\lib\site-packages\git\cmd.py", line 466, in <lambda>
    return lambda *args, **kwargs: self._call_process(name, *args, **kwargs)
  File "D:\Apps\WinPython-64bit-3.5.2.1\python-3.5.2.amd64\lib\site-packages\git\cmd.py", line 910, in _call_process
    return self.execute(make_call(), **_kwargs)
  File "D:\Apps\WinPython-64bit-3.5.2.1\python-3.5.2.amd64\lib\site-packages\git\cmd.py", line 630, in execute
    raise GitCommandNotFound(str(err))
git.exc.GitCommandNotFound: [WinError 6] The handle is invalid

• Digging a bit further, I noticed that before executing the failing Popen(), the subprocess._active list contains 2 leftovers processes, while this list is always empty when not failing.
• I tried to add some delays with time.sleep() but it did not help.
• It did help when I simplified my test-cases, i.e. dropping some commit actions.
• Tested under:
  • Windows 10,
  • WinPython 3.5.2:
  • gitpython: both latest in PyPi v2.0.8 and latest master from github(df65f51) (2.0.9-dev0).
  • smmap: 0.9.0
  • gitdb: 0.6.4

An minor note: the try-catch Popen() code assumes that any exception caught is because git was not found, while above it obviously not the case.