GeezerVison

Plot is now scaled to sort-of look like it did before, in the default view.

Note that because the plot is now actually stretched horizontally, but the camera is not, the view angle looks different. Options:

• Change the default view angle too - presumably overriden in matplotlibrc though?
• Change the default pb_aspect back to a cube
• Deal with it

Reasonably straightforward to implement axis=equal from here too

Any progress here?

Unfortunately not. Progress would be noted here.

I was under the impression the ball was not in my court here. Rebasing shouldn't be too awful, but I don't see much point without further review.

@eric-wieser Not an expert in mpl3d and thus I have no idea who would have to take up the torch. I was just stating that the thread still reflects the current status.

@timhoffm: I'll rebase for good measure, it'll allow someone to work from this branch at least if they need cubic plots.

Can you remind me how to regenerate the baseline_images? Is there a mechanism to say "assume the new image is correct for all failed image comparison tests"?

@eric-wieser See Writing an image comparison test.

Copy the output images (in this case result_images/test_lines/test_line_dashes.png) to the correct subdirectory of baseline_images tree in the source directory

I was hoping there was automation for "assume every failed output is an intended change, overwrite the baseline" so that I can diff with git.

I was hoping there was automation for "assume every failed output is an intended change, overwrite the baseline" so that I can diff with git.

Can you just rsync result_images/ and lib/matplotlib/tests/baseline_images?

No, rsync is not the right tool--you need to explicitly copy the appropriate files from result_images, not all of them.

Certainly you dont want to git add all the comparison files.

triage_tests.py has keyboard shortcuts (run with -h to get the help) to accept images and move between them ("A"/"down"/"A"/"down"/etc should get you there).

@anntzer: Great tool, thanks. Rebased with updated tests

I don't understand the circle-ci failures:

/home/circleci/project/lib/mpl_toolkits/mplot3d/axes3d.py:docstring of mpl_toolkits.mplot3d.axes3d.Axes3D.apply_aspect:4: WARNING: more than one target found for cross-reference 'get_anchor': matplotlib.axes.Axes.get_anchor, mpl_toolkits.axes_grid1.axes_divider.AxesDivider.get_anchor, mpl_toolkits.axes_grid1.axes_divider.Divider.get_anchor

It seems that travis is failing the svg tests, but the svg files are no longer being generated on my machine.

Thanks for the fixups @anntzer, they look fine to me

What is the status of this?

I'm not comfortable merging this over @WeatherGod 's requested changes, however I believe they are addressed.

but I think those are all consequences of the 2.0 default value changes, not due to anything in this PR.

I've tried to make that easy to look at by having separate commits for the style and other changes - so hopefully you can confirm whether those changes matter.

Please see https://discourse.matplotlib.org/t/removed-commits-from-master-branch/20868 This merge was "un-done" so that we can re-do it and squash.

Sorry for causing such churn in the repo. I completely forgot to double-check how many image commits there were.

To be clear, is the current status of this "not actually merged"?

Yes. I think you can reopen a PR with the same commits, except that the intermediate images should be stripped out (so yes the intermediate commits won't be testable independently).