PR Summary

This is a WIP, but for discussion as to whether this is worth the extra code.

It is easy to make negative datetime64 objects, or large datetime64 objects that are greater than 10000 years, whereas right now we restrict to between year 1 and 9999 because we usedatetime.datetime.

The obvious solution is to usedatetime64, but that falls down because a) we want to be able to deal with timezones, anddatetime64 doesn't by default. b) we want to usedateutil.rrule for things like "put a tick every other Wednesday".

So, the solution proposed here is to derive a subclass ofdatetime that keeps track of more years than 1-9999 (I'm not actually clear whydatetime restricts to these values in the first place). I've attempted (probably poorly) to make the subclass as minimal as possible, and to call the super as often as possible.

import matplotlib.pyplot as pltimport numpy as npfig, ax = plt.subplots()dates = np.arange('-20000-01-01', '-20000-02-01', 10, dtype='datetime64[h]')y = np.arange(len(dates))ax.plot(dates, y, 'd')plt.show()

Before

ValueError: view limit minimum -7305216.541666667 is less than 1 and is an invalid Matplotlib date value. This often happens if you pass a non-datetime value to an axis that has datetime units

This PR:

• Make tests with this feature.
• write up docs....

PR Checklist

• Has Pytest style unit tests
• Code isFlake 8 compliant
• New features are documented, with examples if plot related
• Documentation is sphinx and numpydoc compliant
• Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
• Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way