The height of axes of colorbars and parent plots do not exactly match in the inline backend

[javadnoorb]

When I run the following in Jupyter notebook:

%matplotlib inline
import numpy as np
import matplotlib.pyplot as pl

random_array = np.random.random((100, 100))
fig = pl.pcolormesh(random_array);
cbar = pl.colorbar();
points = fig.axes.get_position().get_points()
cbar_points = cbar.ax.get_position().get_points()
print(points[0,1], cbar_points[0,1])

I end up with:

0.125 0.1250000000000001

and the following figure:

Since the above two numbers are not equal, the height of the bottom of the figure doesn't exactly match the bottom of the colorbar.

This apparently is an IPython inline backend issue. Because when I try the above code but comment out the magic function I get equal values:

0.10999999999999999 0.10999999999999999

This is not a very obvious behavior and leads to undesirable outcome in Pandas plots. Pandas determines existence of the components of a figure (such as subplots and colorbars) by comparing their heights. So if colorbar and parent figure do not have the exact same vertical coordinates it will imply that a super tiny extra subfigure is present right underneath the colorbar. This in turn leads to erroneous behavior of sharex argument, leading to missing axis labels as noted here and here. I have proposed a PR on pandas page to deal with this. But since this is stemming from how IPython interacts with matplotlib, I was wondering if there is any interest to dig into its roots in IPython?