One of the most common usages of tight_layout is to expand an axes to fill available space, does this break that?

I am some what surprised that this did not fail tests.

I spent a while digging into this and do not yet understand it well enough to tell what is going wrong, however I am very sure that this is not the right fix.

By adding lots of print statements to test script

import numpy as np
import matplotlib.pyplot as plt
fig = plt.figure()
seen_axes = set()
for i in range(16):
    print(len(fig.axes))
    ax = plt.subplot(4, 4, i + 1)
    seen_axes.add(ax)
    im = ax.imshow(np.random.normal(size=100).reshape([10, 10]))
    for a in fig.axes:
        print(a.bbox.bounds)
    print(len(fig.axes))
    plt.tight_layout()
    for a in fig.axes:
        print(a.bbox.bounds)
    plt.title(i)
    print(len(fig.axes))
    print('---')

plt.show()
print(ax.bbox.bounds)

You can see that something goes funny in the 2nd to last call to tight_layout where the axes get wider.

It really seems like this code could be replaced by half as much code, but still can not tell if this code is confusing because it is confusing, or confusing because it is doing something hard.

'power-cycled' to trigger CI against current master.

OK, this seems very stale. Closing, but feel free to re-open if needed.