axvspan is useful with plot_date.
Two thoughts addressing only convenience not functionality.
1. Might it make sense for the first argument to be a list?
This would be interpreted as
allowing multiple shadings in one go. If breaking the
interface is too costly, perhaps allowing matched lists
for the xmin and xmax arguments would be an alternative.
2. If I wish to shade date ranges, am I right that I now
should produce date objects and then use date2num to get
xmin and xmax? Would it be useful for axvspan to
directly accept the tuples that would be fed to date (or
datetime) for xmin and xmax?