Memory leak

Describe the issue

Plotting a 30M row data frame causes memory only to increase. Also memory isn't collected by gc.collect.
As discussed with matplotlib guys in https://github.com/matplotlib/matplotlib/issues/21259 (and Jupyter guys: https://github.com/jupyterlab/jupyterlab/issues/11182) - we tested and it works as intended. I tested further and think I found the issue - %matplotlib widget:

%matplotlib widget  # if this is commented out, gc.collect releases the memory

import gc

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd


zeros = pd.Series(np.zeros(30_000_000))
df = pd.concat([zeros, zeros, zeros, zeros], axis=1)


FROM = 0
TO = 10_000_000  # df.shape[0]
color_spread = "grey"

fig, axs = plt.subplots(1, 1, sharex=True)

ax2 = axs.twinx()
ax2.plot(df[df.columns[0]][FROM:TO], color="blue")
ax2.plot(df[df.columns[1]][FROM:TO], color="red")
ax2.tick_params(axis="y", labelcolor="red")

ax1 = axs.twinx()
ax1.plot(df[df.columns[2]][FROM:TO], color="orange")
ax1.plot(df[df.columns[3]][FROM:TO], color="yellow")

fig.tight_layout()
plt.show()
gc.collect()

Versions

3.7.10 | packaged by conda-forge | (default, Feb 19 2021, 16:07:37) 
[GCC 9.3.0]
ipympl version: 0.7.0
jupyter core     : 4.7.1
jupyter-notebook : 6.4.3
qtconsole        : not installed
ipython          : 7.27.0
ipykernel        : 6.4.1
jupyter client   : 7.0.2
jupyter lab      : 3.1.11
nbconvert        : 6.1.0
ipywidgets       : 7.6.5
nbformat         : 5.1.3
traitlets        : 5.1.0
Config option `kernel_spec_manager_class` not recognized by `ListNBExtensionsApp`.
Known nbextensions:
  config dir: /home/test/PROGS/miniconda3/envs/puma-lab/etc/jupyter/nbconfig
    notebook section
      jupyter-matplotlib/extension  enabled 
      - Validating: OK
      nbdime/index  enabled 
      - Validating: OK
      jupyter-js-widgets/extension  enabled 
      - Validating: OK
      nbextensions_configurator/config_menu/main  enabled 
      - Validating: problems found:
        - require?  X nbextensions_configurator/config_menu/main
      contrib_nbextensions_help_item/main  enabled 
      - Validating: OK
    tree section
      nbextensions_configurator/tree_tab/main  enabled 
      - Validating: problems found:
        - require?  X nbextensions_configurator/tree_tab/main
Config option `kernel_spec_manager_class` not recognized by `ListLabExtensionsApp`.
[W 2021-10-07 12:20:00.836 LabApp] Config option `kernel_spec_manager_class` not recognized by `LabApp`.
JupyterLab v3.1.11
/home/test/PROGS/miniconda3/envs/puma-lab/share/jupyter/labextensions
        jupyter-matplotlib v0.9.0 enabled OK
        nbdime-jupyterlab v2.1.0 enabled OK
        @jupyterlab/git v0.32.2 enabled OK (python, jupyterlab-git)
        @pyviz/jupyterlab_pyviz v2.1.0 enabled OK (python, pyviz_comms)
        @jupyter-widgets/jupyterlab-manager v3.0.1 enabled OK (python, jupyterlab_widgets)

Other labextensions (built into JupyterLab)
   app dir: /home/test/PROGS/miniconda3/envs/puma-lab/share/jupyter/lab

Beginne mit dem bereitgestellten minimalen Reproduzierer in einem Jupyter-Notebook und vergleiche das Speicherverhalten mit aktiviertem und auskommentiertem %matplotlib widget. Untersuche die ipympl-Widget-Integration und überprüfe, dass nach dem Plotten und dem Aufruf von gc.collect() der Speicher des DataFrames und des Plots wie erwartet freigegeben wird.