Could you provide a minimal example of what you are trying to do. That makes it easier to understand if this is a bug and how to fix it.

I have seen exceptions like this before if you have strings (in this case probably ylabels) on a multi-Axes figure where the text is wider than two axes.

Without a minimal example there is not much we can do as, (as you point out) the tight layout code is rather complicated.

I'm currently trying to build a minimal example, but what I can already say is that this indeed has to do with ylabels. If I shorten my label by 4 letters, the bug doesn't show up.

Here's an example:

#!/usr/bin/env python

import matplotlib
matplotlib.use("PDF")
import matplotlib.pyplot as plt

import sys

n = int(sys.argv[1])

x = [1, 2, 3, 4, 5, 6, 7, 8]

xmin = min(x)
xmax = max(x)

data1_f = [9.2579827408192e-05,
           0.0003372918804693761,
           0.0007482795416375103,
           0.0008833191108781534,
           0.0006607025669975578,
           0.00032707300226202557,
           0.00018651452667246514,
           3.514019691439356e-05]
data1_r = [2.605700651541691e-05,
           2.89119480988452e-05,
           4.2170270917124425e-05,
           4.998876011255778e-05,
           3.1114424088781854e-05,
           1.3605595433950519e-05,
           5.257642663359189e-06,
           3.5083320041254324e-06]
data2_f = [8.470149678300792e-05,
           0.0003150887959573901,
           0.0007836372284836411,
           0.0010760646463750388,
           0.0008765372009137937,
           0.00048812392173035456,
           0.00029758922448336404,
           5.93581217984265e-05]
data2_r = [1.816789924161253e-05,
           2.4672634481371817e-05,
           4.221322246098976e-05,
           6.554534103019158e-05,
           4.963479723751039e-05,
           3.1058180581352124e-05,
           1.2640321929190978e-05,
           4.019617548011485e-06]

# I want data1 and data2 on separate graphics, but with the same vertical scale.
ymin = - max(max(data1_r), max(data2_r))
ymax = max(max(data1_f), max(data2_f))

y_span = ymax - ymin
margin = y_span * 0.05

plt.axis([xmin - 0.5, xmax + 0.5, ymin - margin, ymin + margin])
axis = plt.gca()
axis.set_autoscale_on("False")
axis.bar(x, data1_f, facecolor="red", align="center")
axis.bar(x, [-v for v in data1_r], facecolor="blue", align="center")
axis.set_ylabel(" ".join(["data1"] * n))
plt.tight_layout()
plt.savefig("data1.pdf")
plt.cla()

plt.axis([xmin - 0.5, xmax + 0.5, ymin - margin, ymin + margin])
axis = plt.gca()
axis.set_autoscale_on("False")
axis.bar(x, data2_f, facecolor="red", align="center")
axis.bar(x, [-v for v in data2_r], facecolor="blue", align="center")
axis.set_ylabel(" ".join(["data2"] * n))
plt.tight_layout()
plt.savefig("data2.pdf")
plt.cla()

If I call this with n = 15, the bug doesn't occur. With n = 16, it does:

./debug.py 15
/usr/lib/python2.7/dist-packages/matplotlib/tight_layout.py:225: UserWarning: tight_layout : falling back to Agg renderer
  warnings.warn("tight_layout : falling back to Agg renderer")
./debug.py 16
/usr/lib/python2.7/dist-packages/matplotlib/tight_layout.py:225: UserWarning: tight_layout : falling back to Agg renderer
  warnings.warn("tight_layout : falling back to Agg renderer")
Traceback (most recent call last):
  File "./debug.py", line 72, in <module>
    plt.tight_layout()
  File "/usr/lib/python2.7/dist-packages/matplotlib/pyplot.py", line 1288, in tight_layout
    fig.tight_layout(pad=pad, h_pad=h_pad, w_pad=w_pad, rect=rect)
  File "/usr/lib/python2.7/dist-packages/matplotlib/figure.py", line 1656, in tight_layout
    self.subplots_adjust(**kwargs)
  File "/usr/lib/python2.7/dist-packages/matplotlib/figure.py", line 1516, in subplots_adjust
    self.subplotpars.update(*args, **kwargs)
  File "/usr/lib/python2.7/dist-packages/matplotlib/figure.py", line 225, in update
    raise ValueError('bottom cannot be >= top')
ValueError: bottom cannot be >= top

fig, ax = plt.subplots()
ax.set_ylabel('a'*50000)
plt.tight_layout()

is a minimal example.

There is a check in tight_layout that the pad can not be negative, but there also needs to be one that it can not be > .5.

Hey guys we've written up a quick fix for this one and put in a pull request. Let us know if it's cool with you.

Thanks!

I've got another example that doesn't involve a label that's too long (but maybe too tall?):

from matplotlib import pyplot

fig, ax = pyplot.subplots(figsize=(8, 3))

ax.set_xscale('logit')
ax.set_xlim(left=0.02, right=0.98)
ax.set_xlabel('new logit scale')
ax.set_yticks([])
fig.tight_layout()

Which produces the following (full traceback from jupyter notebook):

C:\Miniconda3\envs\probscale\lib\site-packages\matplotlib\figure.py in tight_layout(self, renderer, pad, h_pad, w_pad, rect)
   1754                                          rect=rect)
   1755 
-> 1756         self.subplots_adjust(**kwargs)
   1757 
   1758 

C:\Miniconda3\envs\probscale\lib\site-packages\matplotlib\figure.py in subplots_adjust(self, *args, **kwargs)
   1610 
   1611         """
-> 1612         self.subplotpars.update(*args, **kwargs)
   1613         for ax in self.axes:
   1614             if not isinstance(ax, SubplotBase):

C:\Miniconda3\envs\probscale\lib\site-packages\matplotlib\figure.py in update(self, left, bottom, right, top, wspace, hspace)
    228             if self.bottom >= self.top:
    229                 reset()
--> 230                 raise ValueError('bottom cannot be >= top')
    231 
    232     def _update_this(self, s, val):

ValueError: bottom cannot be >= top

Dropping into to the debugger, bottom is being set to inf.

Affects me on version 2.2.2

This probably needs a better error message (its possible there already is one for 3.0, I’ve not checked, though I recall this coming up before). However, if you get this error, it means your decorators are not leaving room for your axes, and its size has shrunk to nothing. Either make your decorators smaller or your figure larger.

The error message is still the same on master.

@jklymak

Either make your decorators smaller or your figure larger

It happens when tight_layout() is called. It is resizing the elements that were OK before it was called. So it's a bug there, it must figure automatically if there's no room and leave all as is.

@maahn That should be what it is doing and it is notifying you it can not satisfy the constraints by raising an exception. You can then catch this and do what ever the right thing for your application is.

Right. Or conversely, how small do you want it to make the axes before it errors?

OK, the error here was in the logit scale as in #11386, closed by #11404

I got the same error as a result of calling import matplotlib.pyplot as plt; plt.tight_layout(pad=1.0). The workaround, which allowed me to get rid of the error, was to reduce the font size on my legend. So, it seems, one can get the error if the legend does not fit on the figure.

matplotlib.__version__: 2.2.2

Using @tacaswell's minimum example now gives

>>> plt.tight_layout()
/Users/dstansby/miniconda3/lib/python3.7/site-packages/matplotlib/tight_layout.py:181: UserWarning: Tight layout not applied. The bottom and top margins cannot be made large enough to accommodate all axes decorations. 
  warnings.warn('Tight layout not applied. '

Which I think is good enough to close this as "fixed with a sensible warning". If I've missed anything, feel free to comment or re-open this issue.

import matplotlib.pyplot as plt
fig, ax = plt.subplots()
ax.set_ylabel('a'*50000)
plt.tight_layout()

Running the code above still results in the stack trace below, in matplotlib 3.0.3, the latest version available for Python 2.

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python2.7/dist-packages/matplotlib/pyplot.py", line 1379, in tight_layout
    fig.tight_layout(pad=pad, h_pad=h_pad, w_pad=w_pad, rect=rect)
  File "/usr/lib/python2.7/dist-packages/matplotlib/figure.py", line 1756, in tight_layout
    self.subplots_adjust(**kwargs)
  File "/usr/lib/python2.7/dist-packages/matplotlib/figure.py", line 1612, in subplots_adjust
    self.subplotpars.update(*args, **kwargs)
  File "/usr/lib/python2.7/dist-packages/matplotlib/figure.py", line 230, in update
    raise ValueError('bottom cannot be >= top')
ValueError: bottom cannot be >= top

3.0.3 is not py2 compatible. Maybe you mean 2.2.3? Doubt this fix will be back ported

Apologies, it appears the version of matplotlib I was using as 1.5.1, not 2.2.4.

Upgrading to that version makes the above code function as expected.