Bug summary

The current implementation seems to assume that the lower left ll, upper left ul, and lower right lr corners are preserved under transformation. This need not be the case. For example, an affine transformation that simply switches coordinates produces a wrong result.

Code for reproduction

import numpy as np
from matplotlib.transforms import Affine2D, Bbox

# Switch 'x' and 'y' coordinates
T   = Affine2D( 
        np.array([
          [0,1,0],
          [1,0,0],
          [0,0,1]
        ])
      )

# Unit vectors
pts = np.array([
  [1, 0],
  [0, 1]
])

# Produce a Bbox from the unit vectors
bb = Bbox.null()
bb.update_from_data_xy(pts)

# Result
bb.transformed( T )

Actual outcome

In [84]: bb.transformed( T )
Out[84]: Bbox([[0.0, 0.0], [0.0, 0.0]])

Expected outcome

bb.transformed( T ).height == 1.0
bb.transformed( T ).width == 1.0

Additional information

The current implementation (v3.8.0) seems to assume that the lower left ll, upper left ul, and lower right lr corners are preserved under transformation. A more general solution should probably transform each corner and fully recompute the bounding box.

Operating system

linux

Matplotlib Version

3.8.0

Matplotlib Backend

TkAgg

Python version

3.11.5

Jupyter version

No response

Installation

pip