from datetime import timedelta from boltons.timeutils import total_seconds def test_float_total_seconds(): """Check for floating point precision loss per http://bugs.python.org/issue8644 and tests in the corresponding diff, spurred by https://github.com/mahmoud/boltons/pull/13#issuecomment-93835612 . The funny thing is that floating point precision loss is definitely happening. With or without true division, in Python 2.7's native timedelta.total_seconds() as well as dateutils.total_seconds. The constants in the tests below are from manual tests on Python 2.7.6 final. 2.6 does vary slightly, but it might just be a repr change. With these tests in mind, I'm not sure why the Python issue got created in the first place. """ assert total_seconds(timedelta(microseconds=1)) == 1e-06 assert total_seconds(timedelta(microseconds=-1)) == -1e-06 assert total_seconds(timedelta(microseconds=-2)) == -2e-06 assert total_seconds(timedelta(days=2 ** 29, microseconds=1)) == 46385646796800.0 assert total_seconds(timedelta(seconds=123456.789012)) == 123456.789012 assert total_seconds(timedelta(seconds=-123456.789012)) == -123456.789012