Repository navigation
Description
Bug report
Producing 4 subplots with a rectangular figure (height > width) and setting the subplots to ax.set_aspect('equal') causes fig.tight_layout(pad=0., h_pad=0., w_pad=0.) to introduce a huge vertical gap between the subplots on the first call.
Each subsequent call to tight_layout reduces the gap slowly, until the set arguments pad and h_pad are satisfied.
Code for reproduction
import matplotlib.pyplot as plt
import numpy as np
randarr = np.random.rand(200, 4)
txtbox = 'this is some\ntext with multiple\nlines and absolutely\nno meaning'
fig, ((a1, a2), (a3, a4)) = plt.subplots(2, 2, figsize=(16/2.54, 25/2.54))
a1.scatter(randarr[:, 0] * 3, randarr[:, 1] * 3, label='a\nb')
a2.scatter(randarr[:, 1] * 2, randarr[:, 2] * 2, label='b\nc')
a3.scatter(randarr[:, 2] * 2, randarr[:, 3] * 2, label='c\nd')
a4.scatter(randarr[:, 3] / 3, randarr[:, 0] / 3, label='d\ne')
_ = [_ax.set_aspect('equal') for _ax in (a1, a2, a3, a4)]
fig.tight_layout(pad=0., h_pad=1., w_pad=1.)
To reduce the vertical gap until padding is satisfied:
for _ in range(10):
fig.tight_layout(pad=0., h_pad=1., w_pad=1.)
And it seems that for plots with a high complexity, calls to fig.canvas.draw are required between the calls to tight_layout. But I cannot reproduce this reliably:
for _ in range(10):
fig.tight_layout(pad=0., h_pad=1., w_pad=1.)
fig.canvas.draw()
Actual outcome
The outcome of the code without repeated calls to tight_layout:

Expected outcome
The in my opinion correct output can be produced with:
import matplotlib.pyplot as plt
import numpy as np
randarr = np.random.rand(200, 4)
txtbox = 'this is some\ntext with multiple\nlines and absolutely\nno meaning'
fig, ((a1, a2), (a3, a4)) = plt.subplots(2, 2, figsize=(16/2.54, 25/2.54))
a1.scatter(randarr[:, 0] * 3, randarr[:, 1] * 3, label='a\nb')
a2.scatter(randarr[:, 1] * 2, randarr[:, 2] * 2, label='b\nc')
a3.scatter(randarr[:, 2] * 2, randarr[:, 3] * 2, label='c\nd')
a4.scatter(randarr[:, 3] / 3, randarr[:, 0] / 3, label='d\ne')
_ = [_ax.set_aspect('equal') for _ax in (a1, a2, a3, a4)]
for _ in range(10):
fig.tight_layout(pad=0., h_pad=1., w_pad=1.)
Matplotlib version
- Operating system: Windows 10 64bit
- Matplotlib version: 3.3.1
- Matplotlib backend: Qt5Agg
- Python version: 3.7.7
- Other libraries: Numpy v 1.19.1
All packages are installed from the default conda channel.

Activity
jklymak commentedon Aug 21, 2020
You are asking for two constraints that can't be solved at the same time: 1) that the figure is a certain size and the axes fill the figure, 2) that the axes be a fixed aspect ratio. Given that you have set the aspect ratio to equal, there has to be extra space unless your figure size is tuned to match. So where to put that space?
tight_layouttends to put it between the axes and centre the axes in their allotted space. You would like them to contract and be close together, but that is harder than it sounds to do automatically and generally. You may wish to look at #17246 to see how hard, and that PR still has problems.So yes this would be nice to fix, but unfortunately its not easy. You may want to look at image_grid (https://matplotlib.org/gallery/axes_grid1/simple_axesgrid.html) which makes a ready-made set of axes that do this automatically, but I don't think they will obey tight_layout.
JoElfner commentedon Aug 22, 2020
Thanks! Yeah, I understand that this is a really hard thing to implement. Any other way to achieve it? My usecase for this is: I need my figures to span exactly the text width of an A4 paper size, which typically is either 16cm or 17cm, while the vertical spacing should be minimized as much as possible. Is there any way to define the figure width while leaving the height flexible?
ImageGridseems to be settingsharex=Trueandsharey=True, even thoughshare_all=Falseis set. ForGridin contrast, sharing axes can be disabled, but the same vertical spacing problem occurs as with subplots.jklymak commentedon Aug 22, 2020
I'd start by making the aspect ratio of the figure closer to one-to-one (for a 2x2 grid of axes). That won't be perfect depending on tick labels and other labels, but will be closer. After that you can adjust by hand.
When #17246 gets completed, the idea will be to call that, and then call
fig.savefig('boo', bbox_inches='tight')so that the figure size gets cut down to the right size. OTOH, that may not get you exactly 16 cm; in that case the are pdf trim tools you could use after saving....JoElfner commentedon Aug 24, 2020
Yep, this was a fairly extreme example to show what I wanted to show. :)
I'm looking forward to #17246. Many thanks for working on it!
github-actions commentedon Aug 18, 2023
This issue has been marked "inactive" because it has been 365 days since the last comment. If this issue is still present in recent Matplotlib releases, or the feature request is still wanted, please leave a comment and this label will be removed. If there are no updates in another 30 days, this issue will be automatically closed, but you are free to re-open or create a new issue if needed. We value issue reports, and this procedure is meant to help us resurface and prioritize issues that have not been addressed yet, not make them disappear. Thanks for your help!
QuLogic commentedon Aug 19, 2023
I think this seems to work alright with the compressed layout engine (pass
layout='compressed'tosubplots) instead of tight layout?jklymak commentedon Aug 19, 2023
Should work. Lets close.