ConnectionPatch hidden by plots

[yha]

A ConnectionPatch is sometimes not shown on one of the plots, depending on the position of the axesA and axesB subplots.

A reproducing example, with only minor changes to the ConnnectionPatch example from the docs:

from matplotlib.patches import ConnectionPatch
import matplotlib.pyplot as plt

fig, (ax1, ax2) = plt.subplots(1,2)

xy = (0.3, 0.2)
coordsA = "data"
coordsB = "data"
con = ConnectionPatch(xyA=xy, xyB=xy, coordsA=coordsA, coordsB=coordsB,
                      axesA=ax1, axesB=ax2,
                      arrowstyle="->", shrinkB=5)
ax1.add_artist(con)

# This works fine:
#con = ConnectionPatch(xyA=xy, xyB=xy, coordsA=coordsA, coordsB=coordsB,
#                      axesA=ax2, axesB=ax1,
#                      arrowstyle="->", shrinkB=5)
#ax2.add_artist(con)

ax1.set_xlim(0, 1)
ax1.set_ylim(0, 1)
ax2.set_xlim(0, .5)
ax2.set_ylim(0, .5)

plt.show()

This produces:

Tested on matplotlib 2.0.2 (from Anaconda 4.4.0) on Debian.

[tacaswell]

This is actually the expected behavior, the second axes (ax2) is drawn after ax1 and hence over-draws it. To fix this you can:

• set the zorder (which controls the draw order) of either axes to make ax2 draw second ax1.set_zorder(-1) or ax2.set_zorder(1)
• add the connector to the second axes (as the example did)
• make the background of the second axes undrawn (ax2.set_facecolor('none'))

[tacaswell]

The details of z-order at both the Axes and Figure level should be clearly documented.

[anntzer]

Could an artist such as ConnectionPatch be added to the figure rather than to the axes (and if it can, is there a mechanism to let such artists be drawn last)? This is where it really belongs after all.

[yha]

I think it is also surprising (and does not seem to be explicitly documented) that the artist must be added to the axes passed as axesA rather than axesB, otherwise the ConnectionPatch is not drawn at all.
Perhaps a better interface would be something like
ax.add_connection( xy, coords, target, xyTarget, coordsTarget, <...>)
or a similar method on the figure.

[tacaswell]

It probably should live on the Figure (and there is even an artists list on Figure), but the tooling around that is sparse (and we probably should not document fig.artists.append(...) even if that would work).

[yrevar]

It'd be a nice to have such a feature. Is there any progress on this? Did anyone find any workaround @anntzer @tacaswell @yha ?

[jklymak]

Fig.add_artist doesn’t work?

[yrevar]

Thanks @jklymak , do you mean fig.add_artist(con) instead of ax1.add_artist(con)?

[yrevar]

I just tried that syntax, it fails with error below.

Error in callback <function install_repl_displayhook.<locals>.post_execute at 0x11319b840> (for post_execute):
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/pyplot.py in post_execute()
    107             def post_execute():
    108                 if matplotlib.is_interactive():
--> 109                     draw_all()
    110 
    111             # IPython >= 2

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/_pylab_helpers.py in draw_all(cls, force)
    130         for f_mgr in cls.get_all_fig_managers():
    131             if force or f_mgr.canvas.figure.stale:
--> 132                 f_mgr.canvas.draw_idle()
    133 
    134 atexit.register(Gcf.destroy_all)

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/backend_bases.py in draw_idle(self, *args, **kwargs)
   1897         if not self._is_idle_drawing:
   1898             with self._idle_draw_cntx():
-> 1899                 self.draw(*args, **kwargs)
   1900 
   1901     def draw_cursor(self, event):

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/backends/backend_agg.py in draw(self)
    400         toolbar = self.toolbar
    401         try:
--> 402             self.figure.draw(self.renderer)
    403             # A GUI class may be need to update a window using this draw, so
    404             # don't forget to call the superclass.

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/artist.py in draw_wrapper(artist, renderer, *args, **kwargs)
     48                 renderer.start_filter()
     49 
---> 50             return draw(artist, renderer, *args, **kwargs)
     51         finally:
     52             if artist.get_agg_filter() is not None:

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/figure.py in draw(self, renderer)
   1647 
   1648             mimage._draw_list_compositing_images(
-> 1649                 renderer, self, artists, self.suppressComposite)
   1650 
   1651             renderer.close_group('figure')

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/image.py in _draw_list_compositing_images(renderer, parent, artists, suppress_composite)
    136     if not_composite or not has_images:
    137         for a in artists:
--> 138             a.draw(renderer)
    139     else:
    140         # Composite any adjacent images together

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/patches.py in draw(self, renderer)
   4610         if renderer is not None:
   4611             self._renderer = renderer
-> 4612         if not self.get_visible() or not self._check_xy(renderer):
   4613             return
   4614         FancyArrowPatch.draw(self, renderer)

~/anaconda3/envs/irl/lib/python3.7/site-packages/matplotlib/patches.py in _check_xy(self, renderer)
   4592             x, y = self.xy1
   4593             xy_pixel = self._get_xy(x, y, self.coords1, self.axesA)
-> 4594             if not self.axes.contains_point(xy_pixel):
   4595                 return False
   4596 

AttributeError: 'NoneType' object has no attribute 'contains_point'

[ImportanceOfBeingErnest]

Yeah, it's not working yet. But that would be the plan to have it working soon.

[yrevar]

Oh okay, that'd be cool!
I'm trying to draw trajectories over axis, it should be possible but would be too cumbersome to handle cases like below with connectionpatch's ordering constraints on the axes. Might need to keep flipping axes as per the ordering constraints maybe.

[ImportanceOfBeingErnest]

Actually, I think the logic in

matplotlib/lib/matplotlib/patches.py

Line 4449 in 1f003b0

def _check_xy(self, renderer):

is wrong anyways. So @yrevar, your image would be a good test case; if you can condense it to a minimal, self-contained example it would help us.

[yrevar]

@ImportanceOfBeingErnest Sure, I've pasted a notebook here: https://gist.github.com/yrevar/765cf3456af4119c9fcabf8667dadb4f

[yrevar]

I just found a workaround!
I was using two connection patches to deal with it earlier when I was using lines instead of arrows. It just struck me that the same concept could also be used for drawing arrows, in the reverse direction (i.e., ax2->ax1) we can use "reversed arrow"!!
The limitations is that shrinkB and shrinkA needs to be equal or we'll see weird overlapping arrow heads, but it's okay for my use case.

[ImportanceOfBeingErnest]

Thanks for the example @yrevar
It would be working as it is just by replacing fig.add_artist(con) with #15020. Meanwhile, the workaround you found seems reasonable.