Indeed! If anything, the bottom subplot should expand. Definitely not the top one.

The problem seems to be that

So self.mappable.axes is not set correctly(?).

It is set correctly, it is "just" that the logic doesn't consider this case.

The code assumes that the mappable provided (the scatter plot) is also the object that got the colorbar attached to it and therefore the part that should be updated when the colorbar is removed. Probably one should use ax instead here. (Althiguh that may cause other issues...)

Thanks @oscargus that does indeed seem to be the problem. I'm skeptical we should resize anything, but if that is the existing behaviour we should be storing the result of ax=ax and operating on that (if it is a single axes).

The problem is that the colorbar axes is created outside of colorbar, so the colorbar method never knows if it stole from space anyone or not. I think its overreach for colorbar to then try to give space back...

Just to clarify here:

import matplotlib.pyplot as plt

fig, ax = plt.subplots(2, layout='constrained')
s = ax[0].scatter([0, 1, 2], [0, 1, 2], c=[0, 1, 2])
plt.colorbar(s, ax=ax[0])

plt.show()

Does what I think the OP wants.

If they really don't want to use constrained layout, they can also do various other hacks (inset_axes is the best way), or make a mock scalarmappable

fig, ax = plt.subplots(2)
s = ax[0].scatter([0, 1, 2], [0, 1, 2], c=[0, 1, 2])
plt.colorbar(s, ax=ax[0])
cbar = plt.colorbar(mpl.cm.ScalarMappable(norm=plt.Normalize(), cmap=plt.get_cmap('RdBu_r')),
ax=ax[1])
cbar.remove()