Can you also add tests and awhat_new entry?

Hmm there seems to be a problem with removingtwinx andtwiny axes where the lines and ticks remain, only the numbers disappear. Will look into this later.

Ok it's because thetwin* methods also screw with the host axes' visibility settings. I don't know if and how this should be fixed.

Are you sure you want awhats_new entry for this? Seems like more of a minor bug fix than something worth mentioning there...

I like to advertise when we add long standing missing functionality.

It also serves as advertisement that the remove functionality exists on all
of the other artists.

If you don't want to do it i will leave it to your judgement.

On Tue, Aug 11, 2015, 7:01 PM productivememberofsociety666 <
notifications@github.com> wrote:

Are you sure you want a whats_new entry for this? Seems like more of a
minor bug fix than something worth mentioning there...

—
Reply to this email directly or view it on GitHub
#4898 (comment)
.

Could you reduce the tests to only having pngs ?

I'm trying to understand what the point is behind the host axes' visibility settings being changed by thetwin* functions. Here is what it looks like if I just set them all to invisible:

So the removal code works at least, but thetwin* functions seem a bit inconsistent:twinx andtwiny set their lines invisible and show only the ticks and numbers, whereastwin has everything visible.

Regarding the failing build, I don't know why the pictures aren't close. It works on my machine with all Python versions from 2.6 to 3.4. Fonts maybe?

Make sure that you don't have any funny rcparams set (which should not
matter as the testsshould be overriding that).

Also check what version of freetype you have installed against what is on
travis.

On Wed, Aug 12, 2015 at 6:33 PM productivememberofsociety666 <
notifications@github.com> wrote:

Regarding the failing build, I don't know why the pictures aren't close.
It works on my machine with all Python versions from 2.6 to 3.4. Fonts
maybe?

—
Reply to this email directly or view it on GitHub
#4898 (comment)
.

Can you also squash this down to remove the un-used images from the history? Part of the goal is to keep the size of the repository down.

All right I took out an rcparam that set the backend to GTK3 and uploaded a slightly modified image. Now apparently all builds work except that 2.7 one due to another test failing...

Your first few commits were made without setting your author info; you should update them for reasonable attribution.

Build passes now although I didn't change anything. The test that failed and then magically worked wasmathfont_stix_53 intest_mathext, seehttps://travis-ci.org/matplotlib/matplotlib/jobs/75359652.

There is a race condition in the math text rendering tests that
intermittently fails.

On Thu, Aug 13, 2015, 6:35 AM productivememberofsociety666 <
notifications@github.com> wrote:

Build passes now although I didn't change anything. The test that failed
and then magically worked was mathfont_stix_53 in test_mathext, see
https://travis-ci.org/matplotlib/matplotlib/jobs/75359652.

—
Reply to this email directly or view it on GitHub
#4898 (comment)
.

I think the reason that thetwin* functions mess with the visibility is to prevent any artifacts from rendering nominally identical artists (the ticks) on top of each other.

I think that this remove behavior should also restore the visibility settings of the host axes.

Can you also check if the twin functions mutate any other data structures in the axes (I know the normalaxes.Axes.twin* methods do)? That should probably be un-done as well.

I extended the tests to show better what is going on. Here is what things look like with the currenttwin* functionality:

Like I already said, there is inconsistency between thetwinx/twiny methods and the baretwin method (axis line visibility).

In my opinion, it would be better if all methods behaved liketwin, i.e. make corresponding host axes completely invisible. The last commit contains code that does just that and also makes the_remove_method reverse these changes. Here is what things look like then:

As for your question, I haven't seen any other changes being made to the host axes by thetwin* functions.

Awesome 👍

Thanks for wading into this!

I am not going to hold 1.5 for this PR, but have no problem with it sneaking in (it is far less disruptive than other open PRs).

Looks like a great bugfix. Looks like this needs a rebase, then another run through Travis, but then I'm happy to merge.

Rebase done!