PR Summary

As noted in #6664, Axes.__init__ is fairly slow due to the eventual call to Axis.reset_ticks. @efiring suggested lazy instantiation and @Kojoley was implementing, but I'm unsure of the status. In the meantime though, we still speed things up by avoiding calling cla too many times.

Using the example from #6664 with v2.0.x:

$ python3 -m timeit \
  -s "from matplotlib.figure import Figure; from matplotlib.axes import Axes; fig = Figure()" \
  "Axes(fig, (0, 0, 1 ,1))"
10 loops, best of 3: 29.7 msec per loop

and master is already a bit faster:

$ python -m timeit \
  -s "from matplotlib.figure import Figure; from matplotlib.axes import Axes; fig = Figure()" \
  "Axes(fig, (0, 0, 1 ,1))"
100 loops, best of 3: 17.7 msec per loop

The trouble with Axes.__init__ is that it does a bit too much clearing; Axes.__init__ calls:

• Axes._init_axis
  • Creates self.xaxis and self.yaxis
    • Axis.__init__ -> Axis.cla (2 calls per Axes)
  • Registers each Axis with two spines via Spine.register_axis
    • Axis.cla is called on the registered Axis (2 Spine * 2 Axis = 4 calls per Axes)
• Axes.cla - Necessary to finish initializing the Axes in a clean state, though it's generic for any caller.
  • self.xaxis.cla() / self.yaxis.cla() (2 calls per Axes)
  • self.spines[:].cla() (2 spines per Axis = 4 calls per Axes)

That makes 12 calls to Axis.cla for only 2 Axis objects. Adding some hidden argument to skip these calls as in this PR would reduce those calls by 6, though the speedup is only approximately 4 times:

$ python -m timeit \
  -s "from matplotlib.figure import Figure; from matplotlib.axes import Axes; fig = Figure()" \
  "Axes(fig, (0, 0, 1 ,1))"
100 loops, best of 3: 4.45 msec per loop

PR Checklist

• [N/A] Has Pytest style unit tests
• Code is PEP 8 compliant
• [N/A] New features are documented, with examples if plot related
• [N/A] Documentation is sphinx and numpydoc compliant
• [N/A] Added an entry to doc/users/whats_new.rst if major new feature
• [N/A] Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way