Consider something like a tidal height time series, simulated here:

Now zoom in to the first 3 days, with v2.x defaults, and notice that the x-ticks in the plot below are nearly useless:

The following is the result of this PR:

Here is the example code, so you can try zooming in to smaller and smaller intervals, to see how bad the present defaults are.

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates

x = mdates.datestr2num('2017-01-01') + np.linspace(0, 100, 5000)
y = np.sin(2 * np.pi * x * 24/12.42) + 0.5 * np.sin(2 * np.pi * x * 24/12)
fig, ax = plt.subplots()
ax.xaxis_date()
fig.autofmt_xdate()
ax.plot(x, y)
plt.show()

I have also noted this, but never mustered the energy to do anything about it 👍

Thanks @efiring