Bug report

Bug summary

When adding a ConnectionPatch that spans multiple axes with constrained_layout=True, the layout becomes broken, with the axes getting shrunk considerably. This can be resolved by calling patch.set_in_layout(False) before adding the artist to the axes. It may be useful to include a note about this to the documentation for ConnectionPatch.

Thanks to ImportanceOfBeingErnest for helping me figure this out on stackoverflow: https://stackoverflow.com/questions/57237711/connectionpatch-breaks-constrained-layout-when-crossing-axes/57241978#57241978

Code for reproduction

I followed the example provided here, except that I changed line 4 to use a managed layout.

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(6, 3), constrained_layout=True)

Actual outcome

Running the code with constrained_layout produced the following figure:

However, if I add con.set_in_layout(False) before the ax2.add_artist(con) on line 26, it produces a nicely layed out figure:

Expected outcome

This is the expected figure with constrained_layout=False from the example:

Matplotlib version

• Operating system: Linux, Gentoo 2.6
• Matplotlib version: 3.1.1
• Matplotlib backend: module://ipykernel.pylab.backend_inline
• Python version: 3.6.7
• Jupyter version (if applicable): 5.2.3
• Other libraries:

Packages were installed in a conda environment from conda-forge channel.