[Bug]: Axes placement and tight_layout()

[suuuehgi]

Bug summary

Figure.tight_layout() changes relative placement of axes.

I don't know where this is a bug, a feature request or just unavoidable behavior.

(In this simple case, moving .tight_layout() ahead of .add_axes fixes it.)

Code for reproduction

from matplotlib import pyplot as plt
import matplotlib

fig, axs = plt.subplots( nrows=2, sharex=True )

# left, bottom, width, height
box = matplotlib.transforms.Bbox.from_bounds(.5, 0, .1, .1)

fig.add_axes(fig.transFigure.inverted().transform_bbox( axs[0].transAxes.transform_bbox(box)))
fig.add_axes(fig.transFigure.inverted().transform_bbox( axs[1].transAxes.transform_bbox(box)))

fig.tight_layout()

Actual outcome

Expected outcome

Additional information

Related issue: #24225

Operating system

Arch

Matplotlib Version

3.6.1

Matplotlib Backend

module://matplotlib_inline.backend_inline

Python version

Python 3.10.7

Jupyter version

Jupyter Lab v. 3.4.5

Installation

pip

[jklymak]

This is expected behaviour. Have you considered using ax.inset_axes for your use case?

[suuuehgi]

This is expected behaviour. Have you considered using ax.inset_axes for your use case?

Eureka! 🥳 Didn't know that method existed! Thank you!

For anyone stumbling upon this in the future (it is particularly useful for colorbars):

fig, axs = plt.subplots( nrows=2, sharex=True, layout='constrained' )

for ax in axs.flat:
    # left, bottom, width, height
    fig.colorbar( plt.scatter( [], [] ), cax = ax.inset_axes( [.9, .1, .03, .8] ) )

[jklymak]

https://matplotlib.org/stable/gallery/subplots_axes_and_figures/colorbar_placement.html is meant to answer that question. Let us know if that tutorial doesn't make sense. You may also prefer to use layout='constrained' in most cases to tight_layout.

[suuuehgi]

https://matplotlib.org/stable/gallery/subplots_axes_and_figures/colorbar_placement.html is meant to answer that question. Let us know if that tutorial doesn't make sense. You may also prefer to use layout='constrained' in most cases to tight_layout.

Thank you very much for the hint! I did not know about constrained_layout! Afaics the layout argument is not mentioned in the docstring of subplots. I updated my answer above.

Regarding the tutorial: The tutorial is fine but it falls short on colorbars "on top of / within plots" as sketched above.

For matrix plotting, like imshow, pcolormesh and alike, the colorbar naturally goes next to the plot as written within the tutorial.
But for function plotting / scatter-pots, it might be interesting to move them inwards as I did above.