At the risk of sounding like a broken record... looks like this doesn't happen with mplcairo? can you confirm?

I believe you are correct. Also, I think this affects all artists, not just images, but it really depends on how the artist interacts with the border.

@anntzer coughhumblebragcough 😉

hehe...

As noted in #15970, it was not a full fix, but I couldn't figure out a way to reproduce in plain Matplotlib. Now, I think I have something that still shows the issue. Basically, because the figure background defaults to white, you don't see it, but by changing the background to some other colour, you can see that the image is clipped a bit differently using a clip path over a clip box:

import matplotlib.pyplot as plt
from matplotlib.path import Path
import matplotlib.patches as mpatches
from matplotlib.testing import setup as mpl_setup


plt.rcParams['savefig.facecolor'] = 'b'

data = [[1, 1], [0, 1]]

fig, ax = plt.subplots(1, 4, figsize=(8, 2), sharex=True, sharey=True,
                       facecolor='b')
ax[0].set_xlim(0, 2)
ax[0].set_ylim(0, 2)
ax[0].set_title('Default patch')

for a in ax[2:]:
    a.patch = mpatches.PathPatch(
        Path([[0, 0], [1, 0], [1, 1], [0, 1], [0, 1]]),
        facecolor=plt.rcParams['axes.facecolor'],
        edgecolor='None',
        linewidth=0,
        snap=False,
        transform=a.transAxes)
    a.patch.set_figure(fig)
    a.set_title('Custom patch')

img = []
for a in ax[:3]:
    img.append(a.imshow(data, extent=(0, 2, 0, 2), vmin=-10))
    for s in a.spines.values():
        s.set_bounds(0, 1.25)

img[1].set_clip_box(None)
img[1].set_clip_path(
    Path([[0, 0], [1, 0], [1, 1], [0, 1], [0, 1]]),
    ax[1].transAxes)
ax[1].set_title('Custom clip path')

ax[3].set_title('No image, clip path')
ax[3].set_aspect('equal')
for s in ax[3].spines.values():
    s.set_bounds(0, 1.25)

plt.show()

The first Axes is normal and while the clip position is a bit inconsistent with the spines, they all fall within or under the spine (which being opaque black, covers that up). In the second Axes, where the clip path is set on the image directly, the top extends one pixel extra and the right is clipped a little early (maybe 0.5 pixel.) In the third Axes, the path is set as the background patch, and the image is clipped slightly differently on all four edges. I suspect this is because the patch itself is clipped differently, which is why I put just a custom patch on the fourth Axes and no image. Comparing three and four, that seems to be the case for all sides except the right.

So while the patch changing position a bit might explain a few edge differences, it doesn't explain the second Axes, and I'm going to reopen this.

I was just looking into some snapping related problems and decided to test your cases out here. It looks like the second axes (of the most recent example) is now fixed? Here is the result I get with master:

There are still issues with aliasing in the third and fourth axes in that image though.

If you remove the snap=False from the patch, things look pretty good to me I think. Below is the image with snapping on by default.

Indeed, it is snapping that seems to do this. I am mostly worried about the last case though, as it's the (likely, not tried this change on it yet) problem that manifests in Cartopy.

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!