We wrap the git executable in a shell script that touches one source
file after 'git pull'. This makes the file newer than the index even
though it has not actually changed. If CTest does not refresh the index
properly then the test will fail with a bogus modified file.