Hi @QuLogic, I think I might have found an issue with this commit. I'm not sure if this is the right forum (or if I should create an issue?). I suspect this commit is causing a memory usage growth, even after running the garbage collector. I've only tested it on Windows 10 with python3.11. Can you reproduce this behavior? Details below.

memory_usage_test.py

import gc

from matplotlib.figure import Figure

from numpy.random import rand

import psutil



def plot(num_points):

    fig = Figure()

    ax = fig.add_subplot()

    ax.plot(rand(num_points))

    fig.clear()



def delta_mb_after_plot_and_gc(num_points):

    proc = psutil.Process()

    before_mb = proc.memory_info().rss / 1e6

    plot(num_points)

    gc.collect()

    after_mb = proc.memory_info().rss / 1e6

    delta_mb = after_mb - before_mb

    print(f"plotting {num_points} points leaks {delta_mb:.3f} MB")



def main():

    gc.disable()

    delta_mb_after_plot_and_gc(100_000)

    delta_mb_after_plot_and_gc(1_000_000)

    delta_mb_after_plot_and_gc(10_000_000)



if __name__ == "__main__":

    main()

Running it against this commit and the previous commit (I found this commit via git bisect)

(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> git checkout 597554db667344c4c4ad026ec1c6dd5f4c688d8f

HEAD is now at 597554db66 Add a pybind11 type caster for GCAgg and its requirements



(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> python ..\memory_usage_test.py

plotting 100000 points leaks 1.978 MB

plotting 1000000 points leaks 16.077 MB

plotting 10000000 points leaks 159.719 MB



(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> python ..\memory_usage_test.py

plotting 100000 points leaks 1.970 MB

plotting 1000000 points leaks 16.081 MB

plotting 10000000 points leaks 159.625 MB



(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> git checkout a47e26bd8583f0fcacc1dd83c3f2ac39b1f7a091

Previous HEAD position was 597554db66 Add a pybind11 type caster for GCAgg and its requirements

HEAD is now at a47e26bd85 Add a pybind11 type caster for agg::rgba



(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> python ..\memory_usage_test.py

plotting 100000 points leaks 0.303 MB

plotting 1000000 points leaks 0.041 MB

plotting 10000000 points leaks -0.279 MB



(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> python ..\memory_usage_test.py

plotting 100000 points leaks 0.197 MB

plotting 1000000 points leaks 0.041 MB

plotting 10000000 points leaks -0.311 MB

Environment Details

(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> python --version

Python 3.11.9



(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> [System.Environment]::OSVersion.Version

Major  Minor  Build  Revision

-----  -----  -----  --------

10     0      19045  0



(mpl_venv) PS C:\Users\justi\projects\matplotlib_ref_cycle\matplotlib> pip freeze | findstr matplotlib

-e git+https://github.com/justinjhendrick/matplotlib.git@a47e26bd8583f0fcacc1dd83c3f2ac39b1f7a091#egg=matplotlib

matplotlib-inline==0.1.7

Originally posted by @justinjhendrick in 597554d