Bug summary

When a Path contains one or more NaN vertices along with valid vertices, Path.get_extents() returns an all-NaN Bbox. As a result, a single NaN vertex causes the entire bounding box to become invalid, even though the remaining vertices are valid.

I expected the bounding box to be computed from the valid vertices instead.

Code for reproduction

import numpy as np
from matplotlib.path import Path

verts = np.array([
    [1, 1],
    [np.nan, np.nan],
    [2, 2],
])

path = Path(verts)
print(path.get_extents())

Actual outcome

Bbox(x0=nan, y0=nan, x1=nan, y1=nan)

Expected outcome

I expected the bounding box to be computed from the valid vertices, resulting in a bounding box covering (1, 1) to (2, 2).

i.e Bbox(x0=1.0, y0=1.0, x1=2.0, y1=2.0)

Additional information

Path.iter_segments() defaults to remove_nans=True, treating NaN vertices as missing points during iteration. Since Path.get_extents() computes the bounding box for the same path, I expected it to handle NaN vertices consistently.

Operating system

Ubuntu 24.04

Matplotlib Version

3.12.0.dev229+g1e6bcafd1

Matplotlib Backend

tkagg

Python version

3.12.3

Jupyter version

No response

Installation

git checkout