tests: make test-convert-git.t reproducible
For reasons I can't explain, Git's copy detection code was identifying
different source files on OS X and (presumably) Solaris versus Linux
(which the test was originally authored against). This was causing
unstable test output.
Changing the test to use a non-ambiguous source file appears to make
the test stable.
The test was introduced recently in
ea3540e66fd8.
from __future__ import absolute_import
import pkgutil
__path__ = pkgutil.extend_path(__path__, __name__)