I'm reporting a digression from previous behaviour of the nbagg backend in Jupyter notebooks. Until matplotlib 3.2.2, when I refreshed an interactive figure in a notebook cell by re-executing the cell, the existing figure would be closed and a new one created using the old figure number (when the cell also created the original figure). In matplotlib 3.3.1 the old figure is replaced by the new figure (with a new figure number) in the notebook output, but the old figure still exists in the kernel.

I guess this is related to "NbAgg and WebAgg no longer use jQuery & jQuery UI" from the release notes of matplotlib 3.3.0. Also, ipympl has the same issue: matplotlib/ipympl#4.

This can be tested in a notebook with:

%matplotlib notebook
import matplotlib.pyplot as plt

fig = plt.figure()
plt.get_fignums()

I saw this behaviour in recent versions of Firefox and Google Chrome.

I have manually applied 2c9a85b in case this would reintroduce the old behaviour (which I prefer), but it didn't.