[Bug]: savefig(bbox_inches='tight') clips axis labels on 3D axes

[buddy0452004]

Bug summary

savefig(bbox_inches='tight') clips the zlabel and xlabel on 3D axes. This happens because Axes3D.get_tightbbox() does not include axis labels in its bounding box calculation. On 2D axes this works correctly.

Code for reproduction

import matplotlib.pyplot as plt

fig = plt.figure(figsize=(3, 3))
ax = fig.add_subplot(1, 1, 1, projection='3d')
ax.plot([0, 1], [0, 1], [0, 1])
ax.set_zlabel("Z axis label")
ax.set_xlabel("X axis label")

fig.savefig("3d_tight.png", bbox_inches='tight')

Actual outcome

zlabel and xlabel are clipped at the edges of the saved image.

Confirmed numerically:

tightbbox:  x1=296.0
zlabel x1=321.2  →  25px outside tightbbox on the right
xlabel y0=1.0   →  12px outside tightbbox at the bottom

Expected outcome

Expected outcome
Labels should be fully visible, as they are when saving without`

`.

On 2D axes, get_tightbbox() always includes xlabel tightbbox.y0 == xlabel.y0. On 3D it does not.

Additional information

Related to the broader known issue of 3D axes not fully contributing to bounding box calculations, noted by @QuLogic in #31277. This specific symptom axis labels excluded from get_tightbbox() produces a concrete user-visible artifact in saved figures.

Operating system

No response

Matplotlib Version

3.10.8

Matplotlib Backend

Agg

Python version

No response

Jupyter version

No response

Installation

pip

[rcomer]

Possible duplicate of #28117.

[buddy0452004]

Thanks @rcomer. I looked at #28117 related but I think not a duplicate. That issue is specific to %matplotlib inline in Jupyter. This one reproduces on Agg with a plain savefig() call, no Jupyter involved. The root cause identified by @QuLogic in #28117 (zlabel skipped when for_layout_only=True in axis3d.py) likely applies here too, but the trigger is different. Happy to work on a fix if this is confirmed as separate.

[rcomer]

@buddy0452004 the inline backend automatically uses bbox_inches='tight'.

[buddy0452004]

Thanks @rcomer, understood. Since #28117 has been open for over a year with no fix merged, I'd like to work on it. Will open a PR there.