Thanks for this @michaelaye. Apologies for the messy advice on the mailing list. Did you manage to get the workaround functional?

Hopefully this can get some visibility before the 1.2 release.

Thanks again.

Not really (see feedback on the mailing list.) Your example that way fails to set the extent as well. So I will add it here as another way to reproduce.

The following code with the setting of an extent at the end fails to work for me as well. As a side, it also does not share the x and y axis.

import matplotlib.pyplot as plt
from numpy import arange, array

arr = arange(10000).reshape(100,100)
l = [arr,arr,arr,arr]
narr = array(l)

axes = []
fig = plt.figure()
for i in range(4):
    if i == 0:
        axes.append(fig.add_subplot(2, 2, i))
    if i > 0:
        axes.append(fig.add_subplot(2, 2, i, sharex=axes[0], sharey=axes[0]))

for ax, im in zip(axes, narr):
    ax.imshow(im, extent=[0,100,0,100])

plt.show()

The sharing is working for me, but the extent isn't set right in that example either. Actually, that example is diagnostically helpful. Cheers for posting it.

shared_axes and aspect=1 does not play along well . If both parameters are set, matplotlib changes the adjustable parameter to "datalim" (default is "box"). And when the aspect is applied during the drawing time, it changes the data-limits (not box size).

If you have both x- and -y axes are shared and all the shared axes have aspect=1, you can change the adjustable to "box-forced" which will work as expected.

for ax,im in zip(axes.flatten(),narr):
    ax.set_adjustable("box-forced")
    ax.imshow(im)

@michaelaye - did @leejjoon 's suggestion work for you?

Punting this to (at least) 1.4.x as this is an issue with too many constraints on the limits not playing nice with each other/having unclear precedence/hysteresis.

Sorry for not having answered above. @leejjoon 's workaround does work for me, even showing the sharing of the axes work. I don't remember why we had such a inflated example though. These days I would just do:

fig, axes = subplots(nrows=2, ncols=2, sharex=True, sharey=True)
arr = arange(10000).reshape(100,100)
for ax in axes.flatten():
    ax.set_adjustable('box-forced')
    ax.imshow(arr)
show()

Was there no subplots available in Oct 2012? Time flies...

Edit: Ah, my bad. it was the distinction between subplots and add_subplot that was the essence of this bug in the first place.

Closing as the set_adjustable fixes the problem and the default behavior looks like a feature not a bug.