Fix restoring 'auto' aspect in 3D axes after switching from 'equal' by jayaprajapatii · Pull Request #31304 · matplotlib/matplotlib

jayaprajapatii · 2026-03-14T09:38:16Z

Switching the aspect of a 3D axis from "auto" to "equal" and then back to "auto" did not restore the original plot correctly. The logic for handling "auto" continued executing the equal-aspect adjustments, which modified the axis limits and box aspect.

This change ensures that when aspect == "auto", the box aspect is reset and the function exits early, preventing further equal-aspect logic from running.

The baseline image for test_aspects has been updated to reflect the corrected rendering.

rcomer · 2026-03-14T12:00:22Z

I'm not too familiar with 3D plotting, but the test explicitly set the box aspect. If a user did similarly, shouldn't we respect it?

matplotlib/lib/mpl_toolkits/mplot3d/tests/test_axes3d.py

Lines 88 to 89 in 32805b1

    
           ax.set_box_aspect((3, 4, 5)) 
        
           ax.set_aspect(aspects[i], adjustable='datalim')

jayaprajapatii · 2026-03-14T12:52:44Z

Thanks @rcomer
My intention with the early return was to avoid the equal-aspect adjustment logic from running when aspect="auto", since it was continuing to modify the axis limits and box aspect.
You're right that the test explicitly sets a box aspect. In that case we should respect a user-defined set_box_aspect.

Would it make more sense to only reset the box aspect when it was implicitly modified by the equal-aspect logic, rather than when it was explicitly set by the user?

scottshambaugh · 2026-03-19T03:57:36Z

I think that overriding a previously set aspect with "auto" is the correct behavior. But we should have different behavior with setting "auto" on the "datalim" - the box shouldn't be modified in that case.

scottshambaugh

The test image was changed in separate commits, so that needs to be git squashed or rebased to fix the "PR Cleanliness" failure.

Also, this new behavior needs to be tested.

scottshambaugh

Ok, I think this is looking good for the adjustable='box' case. What's the current behavior for 'datalim'? That should be fixed at the same time as well.

jayaprajapatii · 2026-03-24T21:15:07Z

Hello @scottshambaugh

Updated set_aspect('auto') to respect adjustable. For adjustable='box', the box aspect is reset as expected.
For adjustable='datalim', the existing behavior is preserved.

I verified that in the datalim case, switching to 'equal' does not lead to a meaningful change in limits, so I did not add an assertion for limit differences.
Added a test to cover the adjustable='datalim' behavior.

scottshambaugh · 2026-03-24T21:23:30Z

What is the current behavior for 'datalim'? Please share screenshots, and try to set up a case where this is exercised. I don't think the test added for the datalim case is meaningfully doing anything right now.

jayaprajapatii · 2026-03-24T22:12:07Z

Hello,
I rechecked the datalim case with a minimal example.

Observed behavior:
Initial limits, after setting 'equal', and after switching back to 'auto' all remain the same. So in the 'datalim' case, aspect changes do not seem to affect limits.

Given this behavior, I’m trying to determine what the intended assertion should be for the datalim case. Could you clarify what aspect of the behavior we should validate here?

scottshambaugh · 2026-03-25T17:39:03Z

The auto with datalim should give the same limits in both cases here (which should both match the default autoscaling):

import matplotlib.pyplot as plt
import numpy as np

fig, ax = plt.subplots(1, 1, subplot_kw={'projection': '3d'})

ax.plot([0, 1], [0, 1], [0, 1])
print('default:', np.array(ax.get_w_lims()))
ax.set_aspect('auto', 'datalim')
print('auto:', np.array(ax.get_w_lims()))

ax.set_aspect('equal', 'datalim')
print('equal:', np.array(ax.get_w_lims()))
ax.set_aspect('auto', 'datalim')
print('auto:', np.array(ax.get_w_lims()))

default: [-0.07291667  1.07291667 -0.07291667  1.07291667 -0.02083333  1.02083333]
auto: [-0.07291667  1.07291667 -0.07291667  1.07291667 -0.02083333  1.02083333]
equal: [-0.19444444  1.19444444 -0.19444444  1.19444444 -0.02083333  1.02083333]
auto: [-0.19444444  1.19444444 -0.19444444  1.19444444 -0.02083333  1.02083333]

scottshambaugh · 2026-04-07T18:48:37Z

Hi @jayaprajapatii are you still interested in working on this? I think you may need to rebase to main to get the tests working again

jayaprajapatii · 2026-04-08T16:00:45Z

Thankyou @scottshambaugh
Yes, I’m still working on this. I’ll rebase onto main and update the PR shortly.

scottshambaugh · 2026-05-22T16:59:35Z

Hi @jayaprajapatii are you still interested in working on this? The baseline image test failure is real.

scottshambaugh · 2026-05-24T01:25:08Z

Ok almost there - you'll need to squash /rebase your commits so the aspects.png baseline image isn't modified in multiple of them. We do that to help limit the size of the git history.

scottshambaugh · 2026-05-24T18:46:54Z

I don't see any visual changes in the aspects.png baseline image now, and it passes locally with the new code. I don't think it needs to be updated unless CI complains - could you please remove that image change?

scottshambaugh

This looks good for me, thanks for putting it together!

QuLogic · 2026-05-26T03:07:52Z

+            self._aspect == "auto"
+            and aspect in ('equal', 'equalxy', 'equalxz', 'equalyz')
+        ):
+            self._auto_aspect_restore_limits = self.get_w_lims()


Why is this an attribute? It only appears to be used in this method.

Hii @QuLogic ,
I think the limits need to persist across separate set_aspect calls when switching back from "equal" to "auto", which is why I stored them on the instance.

I understood that part between switching modes. What I don't understand is how it manages to work if you change w_lims (i.e., xlim, ylim, or zlim) between switching.

~That is, this does work now:~

ax2.set_aspect('auto', adjustable='datalim') ax2.set_xlim([-50, 50]) fig2.savefig('3d-plot1.png') # Correct ax2.set_aspect('equal', adjustable='datalim') ax2.set_xlim([-50, 50]) ax2.set_aspect('auto', adjustable='datalim') fig2.savefig('3d-plot2.png') # Previously wrong (inconsistent with 3d-plot1.png) but fixed in this PR.

~~but I'm missing where the persisted limits get refreshed.~~

No wait, I'm dumb; my example doesn't work. It's just that I set the xlim on the Axes before the first save, so of course it carried through to the second one. When doing it individually, this PR is still incorrect:

fig2 = plt.figure() ax2 = fig2.add_subplot(projection='3d') X, Y, Z = axes3d.get_test_data() ax2.contourf(X, Y, Z, cmap="coolwarm") ax2.set_aspect('auto', adjustable='datalim') ax2.set_xlim([-50, 50]) fig2.savefig('3d-plot1.png') # Correct fig3 = plt.figure() ax3 = fig3.add_subplot(projection='3d') X, Y, Z = axes3d.get_test_data() ax3.contourf(X, Y, Z, cmap="coolwarm") ax3.set_aspect('equal', adjustable='datalim') ax3.set_xlim([-50, 50]) ax3.set_aspect('auto', adjustable='datalim') fig3.savefig('3d-plot2.png') # Wrong (inconsistent with 3d-plot1.png)

jayaprajapatii · 2026-06-16T14:25:20Z

Hello @QuLogic

Just a gentle ping, please have a look at this PR.

…d auto

github-actions Bot added the topic: mplot3d label Mar 14, 2026

QuLogic requested a review from scottshambaugh March 17, 2026 21:33

scottshambaugh reviewed Mar 21, 2026

View reviewed changes

Comment thread lib/mpl_toolkits/mplot3d/axes3d.py Outdated

Comment thread lib/mpl_toolkits/mplot3d/axes3d.py Outdated

Comment thread lib/mpl_toolkits/mplot3d/axes3d.py Outdated

jayaprajapatii force-pushed the fix-3dplot-aspect-back-to-auto branch from e8db81d to a4d63fa Compare March 22, 2026 17:10

scottshambaugh reviewed Mar 23, 2026

View reviewed changes

Comment thread lib/mpl_toolkits/mplot3d/tests/test_axes3d.py

github-actions Bot added the status: needs rebase label Apr 10, 2026

jayaprajapatii force-pushed the fix-3dplot-aspect-back-to-auto branch from 3d6cd1f to 2c505d5 Compare April 11, 2026 10:26

github-actions Bot removed the status: needs rebase label Apr 11, 2026

jayaprajapatii force-pushed the fix-3dplot-aspect-back-to-auto branch from 1b03c4c to cc7babe Compare May 24, 2026 06:50

jayaprajapatii force-pushed the fix-3dplot-aspect-back-to-auto branch from cc7babe to 672356e Compare May 25, 2026 06:11

scottshambaugh approved these changes May 25, 2026

View reviewed changes

QuLogic reviewed May 26, 2026

View reviewed changes

scottshambaugh added this to the v3.11.1 milestone Jun 30, 2026

github-actions Bot added the status: needs rebase label Jul 9, 2026

jayaprajapatii force-pushed the fix-3dplot-aspect-back-to-auto branch 2 times, most recently from d889829 to 1d08ff6 Compare July 10, 2026 18:06

github-actions Bot removed the status: needs rebase label Jul 10, 2026

Fix stale limit restoration when explicit limits set between equal an…

7e7d7f8

…d auto

jayaprajapatii force-pushed the fix-3dplot-aspect-back-to-auto branch from 1d08ff6 to 7e7d7f8 Compare July 10, 2026 18:17

Uh oh!

Uh oh!

Conversation

jayaprajapatii commented Mar 14, 2026

Uh oh!

rcomer commented Mar 14, 2026

Uh oh!

jayaprajapatii commented Mar 14, 2026

Uh oh!

scottshambaugh commented Mar 19, 2026

Uh oh!

scottshambaugh left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

scottshambaugh left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jayaprajapatii commented Mar 24, 2026

Uh oh!

scottshambaugh commented Mar 24, 2026

Uh oh!

jayaprajapatii commented Mar 24, 2026

Uh oh!

scottshambaugh commented Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

scottshambaugh commented Apr 7, 2026

Uh oh!

jayaprajapatii commented Apr 8, 2026

Uh oh!

scottshambaugh commented May 22, 2026

Uh oh!

scottshambaugh commented May 24, 2026

Uh oh!

scottshambaugh commented May 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

scottshambaugh left a comment

Choose a reason for hiding this comment

Uh oh!

QuLogic May 26, 2026

Choose a reason for hiding this comment

Uh oh!

jayaprajapatii May 26, 2026

Choose a reason for hiding this comment

Uh oh!

QuLogic Jul 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

QuLogic Jul 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jayaprajapatii commented Jun 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

scottshambaugh commented Mar 25, 2026 •

edited

Loading

scottshambaugh commented May 24, 2026 •

edited

Loading

QuLogic Jul 9, 2026 •

edited

Loading

QuLogic Jul 9, 2026 •

edited

Loading