This should probably be dealt with while fixing #2294.

I'm not seeing exactly what you are talking about with this. Has it already been fixed?

L350 of matplotlib.dates.py is an indirect call to L217, which I think just deals with ordinal floats and uses datetime pretty extensively.

strftime in the current version is L434-L466, but I don't see anything obviously wrong with it.

I no longer remember what prompted me to create this issue.

I think it boils down to the fact that we have all manner of logic for converting dates <-> scalars that should be handled by an external library.

Closing this as the fix is probably a complete overhaul of dates which is sort of already on the radar.

Well with regards to the question of strftime, I think the problem is that as far as I can tell there's nothing in the standard library that will work for dates < 1900 AD. A fix has apparently been implemented in 3.x, but is still limited to dates > 1000 AD. In my opinion, it's probably better to leave that code as is if it's working rather than add any additional dependencies.

The main other place that I see a problem is in the functions which call _from_ordinalf and _to_ordinalf, which gets the Gregorian days from datetime, then has its own method for converting fractional days into hours, minutes, seconds, etc. Again it seems like there's no library function that converts a fractional day into a timedelta, and the way it's done in dates.py is standard. See Issue 12772. Actually, I think I found a way to do this. See PR #3947.