It's unclear to me why we wouldn't want to fail with datetime types. After all you can't add datetimes so using them as x/yerr should fail, afaict (you can add a datetime to a timedelta, but timedeltascan be meaningfully compared to zero).

Thanks@anntzer! Yeah, I added this line to not fail ondatetime.timedelta

Ah, I was thinking about np.timedelta, not datetime.timedelta, sorry for the careless reading (np.timedelta(...) > 0 works). Well, I guess my point remains: it would be nice to have the check also for datetime.timedelta inputs, but I don't know how easy that is.

@anntzer I've added a check for timedelta types. However, I'm a bit uncomfortable with checking onlytimedelta type.datetime.datetime handles only positive values -> should not be an issue. What aboutdatetime.date?

you can add a datetime to a timedelta

Do you mean, that there is a way to convert anything totimedelta and check it? If yes, could you elaborate a bit more?

I agree the special-casing is a bit annoying, I don't have any good solution to offer right now though.