Bug summary

In Triangulation, when calling get_trifinder, the trifinder is stored is self._trifinder, however TriFinder itself also stores a reference to the triangulation in self._triangulation. As a result, neither object is ever destroyed.

Code for reproduction

import matplotlib.tri as tri
import numpy as np
import psutil

n_points = 500000
x = np.random.rand(n_points) * 100
y = np.random.rand(n_points) * 100

while True:
    triangulation = tri.Triangulation(x, y)
    trifinder = triangulation.get_trifinder()

    mem = psutil.virtual_memory().available * 100 / psutil.virtual_memory().total
    print(f"Available memory: {mem:.2f}%")

Actual outcome

Available memory: 75.65%
Available memory: 69.80%
Available memory: 63.91%
Available memory: 58.04%
Available memory: 52.15%
Available memory: 46.27%
Available memory: 40.40%
Available memory: 34.53%
Available memory: 28.65%
Available memory: 22.87%
Available memory: 17.03%
Available memory: 11.12%
Available memory: 6.25%
Available memory: 0.95%
Available memory: 0.62%
Available memory: 0.73%
Available memory: 0.95%
Available memory: 0.90%
Killed

Expected outcome

The memory should remain the same after each iteration.

Additional information

The _triangulation attribute in TriFinder is never actually used and can be safely removed: this fixes the issue.

Operating system

Linux (Debian 12), Windows

Matplotlib Version

3.9.0

Matplotlib Backend

qtagg

Python version

3.11.2

Jupyter version

7.1.2

Installation

pip