The major formatter is definitely not getting propagated through the set_position call:

In [15]: ax2.yaxis.get_major_formatter()
Out[15]: <matplotlib.ticker.ScalarFormatter at 0x55d0410>

In [16]: ymajorFormatter
Out[16]: <matplotlib.ticker.FormatStrFormatter at 0x55bce90>

However I am not sure what the 'correct' behavior is.

My understanding was basically that unless two calls were mutually exclusive (e.g., the same parameter getting set twice) all commands prior to plt.show() or equivalent should be order agnostic.

If this is the design goal, it seems this should propagate through changing the position of the spines. Users would expect spine position and formatting of the axis label to be independent.

@JDWarner duh, yes, you are right. Looking at this again I was confused about what was going on (thought that the non-commuting operation was setting the ticks to be on the right axis, not the left, which is not what is happening).

Re-milestoned and labeled accordingly.

Digging into this, it seems the offending line (https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/spines.py#L382) clear the axis (which resets everything to the defaults) after adjusting the spine position.

This looks like it was a feature of the original introduction of this code in 2009 (43ca13b) !

I suspect the fix is to just remove the self.axis.cla() block. @astraw, you have a comment?

I don't think this is a bug per se, just a wart.

I don't remember exactly why the call to self.axis.cla() was necessary, but I remember that it was. You're welcome to play with removing it. Check especially the spine_placement_demo.py example and make sure that zooming and panning continues to work. And spine placement code should also work with the basemap projections.

By design, the matplotlib API is not order-independent. For example, this code shows that most basic plotting commands get drawn in the order they were called:

import matplotlib.pyplot as plt

fig=plt.figure()
ax1=fig.add_subplot(211)
ax1.plot([1,2,3],[4,5,4],'b-')
ax1.plot([1.5],[4.5],'ro',ms=20.)

ax2=fig.add_subplot(212)
ax2.plot([1.5],[4.5],'ro',ms=20.)
ax2.plot([1,2,3],[4,5,4],'b-')

plt.show()

@astraw Granted, as I noted before when one or more command either references the same space or is mutually exclusive the order matters. Ergo the zorder= kwarg.

The behavior in this issue, however, is not mutually exclusive nor is it intuitive to the user. When setting independent (or seemingly independent) features about a figure/axis, it's not expected that the order would matter.