Repository navigation
[Bug]: Incorrect offsets and scaling of EllipseCollection #30866
Description
Bug summary
This is a regression bug. The position and scaling of EllipseCollection do not follow the specification of offsets.
Code for reproduction
import matplotlib.pyplot as plt
import numpy as np
from matplotlib.collections import EllipseCollection
x = np.arange(2)
y = np.arange(2)
X, Y = np.meshgrid(x, y)
XY = np.column_stack((np.zeros_like(X.ravel()), Y.ravel()))
fig, ax = plt.subplots()
ec = EllipseCollection(X, Y, X, units='x', offsets=XY, offset_transform=ax.transData)
ax.add_collection(ec)
plt.show()Actual outcome
Expected outcome
Additional information
This works in v3.10.8.
Operating system
No response
Matplotlib Version
0.2.0.dev53474+unknown.g9e210e864
Matplotlib Backend
No response
Python version
No response
Jupyter version
No response
Installation
None
Activity
rcomer commentedon Dec 16, 2025
The change comes from #29958, but I would not say it is a regression as such. Autoscaling is now automatic when you add a collection, as implied by the default value of the autolim parameter. To preserve the old behaviour, you can do
If you call
ax.autoscale_view()after adding the collection then you get the same behaviour in v3.10.8 and inmain.The fact that the automatically determined limits seem to be wrong in this case probably should be investigated.
timhoffm commentedon Dec 16, 2025
Indeed automatic scaling is an intended change.
In fact, they are correct:
Bbox(x0=0.0, y0=0.0, x1=0.0, y1=1.0). Note that data limits are calculated based on the offsets only.It's just that the data is chosen maximally confusing: There are 4 ellipses, two at (0, 0), two at (0, 1), which explains the data limits. Widths and heights are permutated between 0 and 1, so that only the 4th ellipse has non-zero width and height and is thus visible. 🤪 @williamlus please take a little more care when setting up an example that you don't use data leading to such degenerate cases.
williamlus commentedon Dec 17, 2025
@rcomer Yes, setting
autolim=Falsecan preserve the old behaviour. However, I found a new issue with autoscaling. After callingax.autoscale_view(), the scaling is incorrect. Is this behavior intended or not? @timhoffmActual: With
ax.autoscale_view()Expected: Without
ax.autoscale_view()timhoffm commentedon Dec 17, 2025
This is correct.
autolim=Falsemeans the artist is not included in the limits calculation. Since no limit-setting data is contained in the Axesautoscale_view()falls back to a small margin around (0, 0).