FIX: Apply aspect before drawing starts #11753

jklymak · 2018-07-23T17:20:44Z

PR Summary

As noted in #4375 (comment) connection patches between artists with different transforms do not always work on the first draw, though they will on the second draw. This happens because apply_aspect happens at draw time for axes, and not always in the order necessary for these connections to take place properly.

This PR proposes calling apply_aspect on all axes in a figure before they are drawn. This positions them properly before the draw, and makes these kinds of connectors work.

Pro: connector patches (annotate) work.
Con: apply_aspect gets called twice. I've removed the apply_aspect call from ax.draw. So the aspect is applied at the figure level rather than at each axes. This isn't necessary - it can be in both places, but that has extra overhead. However, I'm not 100% certain that apply_aspect can be removed from ax.draw.

import matplotlib.pyplot as plt
import numpy as np

x=np.linspace(0,10,200)
y=np.exp(-x)*np.sin(x)

fig,ax = plt.subplots(figsize=(3.39,3))

ax.plot(x,y)

axins = ax.inset_axes([0.4, 0.5, 0.3, 0.3])
axins.xaxis.set_visible(False)
axins.yaxis.set_visible(False)
axins.set_aspect(0.2)
ax.annotate("",xy=(x[150],y[150]), xycoords=ax.transData,
        xytext=(1, 0), textcoords=axins.transAxes,
        arrowprops=dict(arrowstyle="->"))
# axins.apply_aspect()

plt.show()
fig.savefig('boo.png')

Before

Show:

boo.png:

After

PR Checklist

Has Pytest style unit tests
Code is PEP 8 compliant
New features are documented, with examples if plot related
Documentation is sphinx and numpydoc compliant
Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way

jklymak · 2018-07-23T17:23:53Z

Ping @leejjoon who's comment over at #4375 inspired this.

WeatherGod · 2018-07-23T17:36:45Z

I wonder if there is a good way to create a test for this? As for moving the apply aspect, I think it should be in both places, but perhaps we can add some logic to detect that it has already happened?

…

On Mon, Jul 23, 2018 at 1:23 PM, Jody Klymak ***@***.***> wrote: Ping @leejjoon <https://github.com/leejjoon> who's comment over at #4375 <#4375> inspired this. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#11753 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARy-CYXGBX9o767ju5clwoHX2cRviUmks5uJgargaJpZM4VbYVF> .

jklymak · 2018-07-23T17:45:44Z

Agreed, this definitely needs a test. Happy to keep it in twice. I don't think its really very slow and makes life a lot more consistent...

jklymak · 2018-07-23T20:05:48Z

Actually to pass the tests, we need the apply_aspect in the draw as well.

jklymak · 2018-07-23T21:41:00Z

Failing test is just the macOS small tolerance issue, not this PR...

Note this fixes an annoyance in the zoom-inset example where the indicators didn't line up (grey lines in the plots)....

Before

https://matplotlib.org/devdocs/gallery/subplots_axes_and_figures/zoom_inset_axes.html#sphx-glr-gallery-subplots-axes-and-figures-zoom-inset-axes-py

After

https://12024-1385122-gh.circle-artifacts.com/0/home/circleci/project/doc/build/html/gallery/subplots_axes_and_figures/zoom_inset_axes.html#sphx-glr-gallery-subplots-axes-and-figures-zoom-inset-axes-py

jklymak · 2018-07-23T22:00:32Z

Milestoning as 3.1, as this has been a long-standing problem with linked artists between axes, and I know we want to get 3.0 out. OTOH, it would be nice to see it get into 3.0, and even 2.2.3 because it really does fix any plot that has artists that depend on the proper axes location to draw.

jklymak · 2018-08-15T20:46:31Z

Rebased on master.

I still think this would be good for 3.0, because it really fixes some long-term annoyances...

jklymak · 2018-09-19T19:51:18Z

I don’t think this needs a rebase ;-)

jklymak · 2018-09-22T03:55:09Z

Ooops, I guess it did - rebase complete!

QuLogic · 2018-09-22T06:27:46Z

You have flake8 errors:

./lib/matplotlib/tests/test_axes.py:5793:1: E303 too many blank lines (3)
1     E303 too many blank lines (3)

ImportanceOfBeingErnest · 2018-09-29T21:07:57Z

If you need to apply_aspect for the figure to correctly show up, doesn't that mean that at some point down the road real coordinates are used instead of transforms? I mean if the transform chain isn't broken, it would be enough to draw the axes, inside the axes' draw apply the aspect then draw all children of the axes.

jklymak · 2018-09-29T21:23:15Z

I think the problem is that Connection patches attach two axes but they are the child of only one axes and you can’t control if the other axes has been drawn yet.

ImportanceOfBeingErnest · 2018-09-29T22:28:47Z

Makes sense. Left is what is currently done, right is what would prevent this from happening.

I suppose there is then no other way, except completely refactoring the drawing order - which I guess is out of question.

jklymak · 2018-09-29T23:08:05Z

Refactoring the drawing order wouldn't be too hard, but I think would make the code a lot less clear.

Another idea here is to call apply_aspect when set_aspect is called. That would potentially be less expensive and might still fix the problem. OTOH, I don't think apply_aspect is that expensive in the first place, though the code is long-ish.

Scratch that - the user could change a lot about the axes after its aspect is set that would change its physical size (i.e. the x/y limits) and hence the apply_aspect needs to be done at draw time.

ImportanceOfBeingErnest

Approving since there seems currently no easy alternative than to call apply_aspect this soon.

ImportanceOfBeingErnest · 2018-10-02T17:58:24Z

Not sure in how far this could go into 3.0.x, since it's mainly a bug fix?

jklymak · 2018-10-02T18:01:07Z

Yeah but not a regression (it’s basically always been this way) so we don’t usually backport.

jklymak · 2018-10-05T15:37:10Z

This has two approvals... I'm hesitant to self-merge in case someone else wants to make sure it makes sense...

jklymak · 2018-11-05T18:25:32Z

Rebased. Will merge if passes CI....

jklymak force-pushed the fix-apply-aspect-before-draw branch 2 times, most recently from a2f0415 to f120164 Compare July 23, 2018 17:22

jklymak added the status: work in progress label Jul 23, 2018

jklymak force-pushed the fix-apply-aspect-before-draw branch 2 times, most recently from 4f17959 to b53f410 Compare July 23, 2018 20:05

jklymak removed the status: work in progress label Jul 23, 2018

jklymak force-pushed the fix-apply-aspect-before-draw branch from e7b4b43 to 87cb53c Compare July 23, 2018 21:49

jklymak added this to the v3.1 milestone Jul 23, 2018

jklymak force-pushed the fix-apply-aspect-before-draw branch from 87cb53c to cc44df8 Compare July 24, 2018 00:08

jklymak force-pushed the fix-apply-aspect-before-draw branch from cc44df8 to 887c9d2 Compare August 15, 2018 20:45

dstansby approved these changes Sep 16, 2018

View reviewed changes

QuLogic added the status: needs rebase label Sep 18, 2018

jklymak force-pushed the fix-apply-aspect-before-draw branch from 887c9d2 to cbbaca2 Compare September 22, 2018 03:54

jklymak removed the status: needs rebase label Sep 22, 2018

jklymak mentioned this pull request Sep 29, 2018

Improve selection of inset indicator connectors. #12334

Merged

6 tasks

jklymak force-pushed the fix-apply-aspect-before-draw branch from cbbaca2 to c4a2feb Compare September 29, 2018 23:17

jklymak force-pushed the fix-apply-aspect-before-draw branch from c4a2feb to 7d76522 Compare September 29, 2018 23:17

ImportanceOfBeingErnest approved these changes Oct 2, 2018

View reviewed changes

jklymak added the topic: geometry manager LayoutEngine, Constrained layout, Tight layout label Oct 3, 2018

jklymak added 3 commits November 5, 2018 10:23

FIX: Apply aspect before drawing starts

b39bd9d

FIX: add apply_aspect back to _base draw

c720045

TST: add test for first draw having correct annotation position

a4e7ddd

jklymak force-pushed the fix-apply-aspect-before-draw branch from 7d76522 to a4e7ddd Compare November 5, 2018 18:24

jklymak merged commit b6154b6 into matplotlib:master Nov 5, 2018

ImportanceOfBeingErnest mentioned this pull request Jul 12, 2019

savefig messes up coordinate transformation of annotate #4375

Closed

Uh oh!

FIX: Apply aspect before drawing starts #11753

FIX: Apply aspect before drawing starts #11753

Uh oh!

Conversation

jklymak commented Jul 23, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

Before

Show:

boo.png:

After

PR Checklist

Uh oh!

jklymak commented Jul 23, 2018

Uh oh!

WeatherGod commented Jul 23, 2018 via email

Uh oh!

jklymak commented Jul 23, 2018

Uh oh!

jklymak commented Jul 23, 2018

Uh oh!

jklymak commented Jul 23, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Before

After

Uh oh!

jklymak commented Jul 23, 2018

Uh oh!

jklymak commented Aug 15, 2018

Uh oh!

jklymak commented Sep 19, 2018

Uh oh!

jklymak commented Sep 22, 2018

Uh oh!

QuLogic commented Sep 22, 2018

Uh oh!

ImportanceOfBeingErnest commented Sep 29, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jklymak commented Sep 29, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ImportanceOfBeingErnest commented Sep 29, 2018

Uh oh!

jklymak commented Sep 29, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ImportanceOfBeingErnest left a comment

Choose a reason for hiding this comment

Uh oh!

ImportanceOfBeingErnest commented Oct 2, 2018

Uh oh!

jklymak commented Oct 2, 2018

Uh oh!

jklymak commented Oct 5, 2018

Uh oh!

jklymak commented Nov 5, 2018

Uh oh!

Uh oh!

jklymak commented Jul 23, 2018 •

edited

Loading

jklymak commented Jul 23, 2018 •

edited

Loading

ImportanceOfBeingErnest commented Sep 29, 2018 •

edited

Loading

jklymak commented Sep 29, 2018 •

edited

Loading

jklymak commented Sep 29, 2018 •

edited

Loading