pep8 :(

/home/travis/build/matplotlib/matplotlib/lib/matplotlib/lines.py:37:1: E302 expected 2 blank lines, found 1

I assume these are unrelated, but the scatter markers don't seem quite so centred and one of the axes in test_axes/errorbar_mixed.png is in the wrong colour.

The markers looking slightly off center is due to 37927b2. It makes the markers look more round, but it also no longer centers them around 0. I'll see if there's a better middle ground in a separate PR.

Good point about the update of the color in test_axes/errorbar_mixed.png. The test is explicitly asking for green there -- I'll have to get to the bottom of why that isn't being obeyed.

The test is explicitly asking for green there -- I'll have to get to the bottom of why that isn't being obeyed.

I think this was just due to a mistake on my part. It's been corrected here.

Do we want to revert 37927b2 ? It did fix a problem, but seems to have created another one which is just as bad.

Do we want to revert 37927b2 ? It did fix a problem, but seems to have created another one which is just as bad.

I'm wondering if it can be addressed by snapping only the end points and not the control points of the Bézier curve. I plan to experiment with that today when I get a chance. If that doesn't yield better results then, yes, maybe reverting is the best option.

EDIT: Actually -- it's already doing what I was planning to try. (git blame == me).

        if (m_snap && agg::is_vertex(code)) {
            *x = floor(*x + 0.5) + m_snap_value;
            *y = floor(*y + 0.5) + m_snap_value;
        }

See #5603 for a possible solution to the off-centered marker issue.

Let's hold on this one until #5603 is merged.

I kicked to restart on current master, but @mdboom needs to re-regenerate the images anyway.

I've redone the baseline images here after #5603. Once Travis passes, I'd say this is good to go.

The superscript changes in errorbar_mixed.png are not for the better but I assume they are unrelated?

I think that might be fixed in master already. I'll rebase and see what happens locally.

I think the new superscript placement is fine -- I realize it's not as high, but with respect to all of the other combinations that sub/super can find themselves in, I think it's the best choice for the DejaVu font -- anything else has negative consequences in other dimensions. See #4873.

Rebased to test this in zero-tolerance mode.

backported as ebf81db

Backport to v2.x is actually via 4f40a9d.