Problem

We broke this in 3.5.0, but in 3.4.3

import matplotlib.pyplot as plt
import numpy as np
import matplotlib as mpl

fig, ax = plt.subplots()
pc = ax.imshow(np.random.randn(10, 10), cmap='RdBu_r')
cb = fig.colorbar(pc, ax=ax, boundaries=np.linspace(-4, 4, 5))
plt.show()

would do:

I'm not a fan of us enabling digitizing the colorbar differently than the data. If someone wants boundaries, they should be doing them with a BoundaryNorm:

fig, ax = plt.subplots()
norm = mpl.colors.BoundaryNorm(np.linspace(-4, 4, 5), 256)
pc = ax.imshow(X, cmap='RdBu_r', norm=norm)
cb = fig.colorbar(pc, ax=ax)

Proposed solution

deprecate the boundaries kwarg....