I think it's probably worth writing a new function called something like _dtime64_to_ordinalf(dts) with the new code to handle np.datetime64s. (for clarity)

ooops, I guess I missed a rebase here. Will fix after

It runs for me without converting a single item into array and back. This is needed for some older numpy (mine is 1.13.3)?

Hmmm, I didn't even try! You're right..

t0 = np.datetime64('0001-01-01T00:00:00', 's')

any reason this isn't done with base/time/timenp being pytest fixtures and the test parameterized or split into two?

Because I can understand this, and I have no idea what a pytest fixture is. Is there an advantage to doing something more obscure than a simple assert?

Only cause you're repeating yourself a lot so it's hard to tell what the specific tests are. Factoring out common features into a fixture means you can make two tests that are uniquely named for the things they're testing.

I'll look into it. OTOH, I find some of the tests aren't very easy to follow. Sometime verbosity is a good thing!

OK, happy to take suggestions, but I find the uses of fixtures in the existing tests makes for pretty obfuscated code.

shrugs can agree on that somewhat, either way it's not something I'd hold the PR up on.

Well, I didn't mean to say I wasn't willing to learn, just that I don't get it... More than happy if you wanted to push a change or give me some pseudo code to get it to work... Thanks!

Sorry for missing your reply (the embedded ones seem to not pop up as notifications), and will totally try to remember your suggestion for the future. I should have been way clearer on what I wanted and given you an example. :/

same here, could this be parameterized with ids?

I sympathise with @jklymak re. pytest, (I still have no idea what a fixture is) but find paramaterising fairly easy to understand (https://docs.pytest.org/en/latest/parametrize.html)

Something like this should work:

@pytest.mark.parametrize('t0', [datetime.datetime(2017, 1, 1, 0, 1, 1),

                                [datetime.datetime(2017, 1, 1, 0, 1, 1),
                                 datetime.datetime(2017, 1, 1, 1, 1, 1)],

                                [[datetime.datetime(2017, 1, 1, 0, 1, 1),
                                  datetime.datetime(2017, 1, 1, 1, 1, 1)],
                                 [datetime.datetime(2017, 1, 1, 2, 1, 1),
                                  datetime.datetime(2017, 1, 1, 3, 1, 1)]]])
@pytest.mark.parameterize('dtype', ['datetime64[s]',
                                    'datetime64[us]',
                                    'datetime64[ms]'])
def test_date_numpyx(t0, dytpe):
    time = mdates.date2num(t0)
    tnp = np.array(t0, dtype=dtype)
    nptime = mdates.date2num(tnp)
    assert time == nptime

This will run through every possible combination of t0 and dtype, and avoids copy/pasting the actual testing code.

OK, I agree that parametrize works well here (despite pytests odd way of spelling it) Thanks for the example!