animation blitting is no longer working

[jrevans]

This is mentioned in #4897.

I currently do not have a solution for this one, but I do have a script that demonstrates the problem. I am using the current devel branch and a Qt5Agg backend.

The following is in reference to the attached script.

The line 'bg = fig.canvas.copy_from_bbox( fig.bbox )' will in fact save the current Agg buffer state as appropriate. The later calls to 'fig.canvas.restore_region( bg, fig.bbox )' however do not seem to actually restore the Agg buffers to the saved state, because in the subsequent call to 'fig.canvas.blit( fig.bbox )' the Agg buffers are still in the same state as before the call to 'restore_region'. The result is the red ellipse gets drawn multiple times to the Agg backend.

At the end of the script a new call to the canvas 'draw' method causes the entire figure to redraw (and the Agg backend), and the plot then is updated to look as expected.

======================================================================
#   blit.py
======================================================================
#!/bin/env python

import pylab as p
from matplotlib.patches import Ellipse
import math, time

p.ion()
fig, ax = p.subplots()
ax.set_xlim( [ -3, 3 ] )
ax.set_ylim( [ -3, 3 ] )

e1 = Ellipse( [ 0, 0 ], 1, 2, 0 )
ax.add_patch( e1 )

e2 = Ellipse( [ 2, 0 ], 1, 3, 0 )
ax.add_patch( e2 )

raw_input( "Press return to start blit" )

bg = fig.canvas.copy_from_bbox( fig.bbox )

e0 = Ellipse( [1,0], 1, 1, 0, facecolor="red" )
ax.add_patch( e0 )

t0 = time.time()
num = 0
while True:
   dt = time.time() - t0
   if dt > 1:
      print "%d frames in %g sec (%g frames/sec)" % ( num, dt, num/dt )
      break

   fig.canvas.restore_region( bg, fig.bbox )

   x = num / 10.0
   e0.center = [ math.cos( x ), math.sin( x ) ]

   ax.draw_artist( e0 )
   fig.canvas.blit( fig.bbox )

   num += 1


print "================="
print "Qt5 bug: red ellipse is not removed each time."
raw_input( "Press return to remove red ellipse" )

e0.remove()
fig.canvas.restore_region( bg, fig.bbox )
fig.canvas.blit( fig.bbox )

print "================="
print "The red ellipse should be gone - if it's still there, try to resize"
print "the screen.  In some cases an extra 'draw' happens here which "
print "causes the entire plot to redraw and look correct.  In Qt5agg, "
print "this shows the bug..."
print
raw_input( "Press return to exit" )

[tacaswell]

I ended up modifying your script a bit (mostly style + python3 compatibility as I only have qt5 + python3 set up to test with) and can not reproduce this issue with either pyqt4 or pyqt5 in python3 @ 3520e94

from __future__ import print_function
import matplotlib.pyplot as plt
from matplotlib.patches import Ellipse
import math
import time

plt.ion()
fig, ax = plt.subplots()
ax.set_xlim([-3, 3])
ax.set_ylim([-3, 3])

e1 = Ellipse([0, 0], 1, 2, 0)
ax.add_patch(e1)

e2 = Ellipse([2, 0], 1, 3, 0)
ax.add_patch(e2)

input("Press return to start blit")

bg = fig.canvas.copy_from_bbox(fig.bbox)

e0 = Ellipse([1, 0], 1, 1, 0, facecolor="red")
ax.add_patch(e0)

t0 = time.time()
num = 0
while True:
    dt = time.time() - t0
    if dt > 1:
        print("%d frames in %g sec (%g frames/sec)" % (num, dt, num/dt))
        break

    fig.canvas.restore_region(bg, fig.bbox)

    x = num / 10.0
    e0.center = [math.cos(x), math.sin(x)]

    ax.draw_artist(e0)
    fig.canvas.blit(fig.bbox)

    num += 1


print("=================")
print("Qt5 bug: red ellipse is not removed each time.")
input("Press return to remove red ellipse")

e0.remove()
fig.canvas.restore_region(bg, fig.bbox)
fig.canvas.blit(fig.bbox)

print("=================")
print("The red ellipse should be gone - if it's still there, try to resize")
print("the screen.  In some cases an extra 'draw' happens here which ")
print("causes the entire plot to redraw and look correct.  In Qt5agg, ")
print("this shows the bug...")

input("Press return to exit")

[mdehoon]

With the current master (latest commit 512b5d7), the animation doesn't run at all with qt5agg. With tkagg, the animation runs, but the red ellipse gets drawn multiple times as described by @jrevans .

[tacaswell]

ah, I now see what is going on, in 512b5d7 I am getting a bunch of extra draws so it looks like the animation is working, but is running much slower than it should be. I do see the overlapped circles with tkagg and if I merge the two qt PRs into master.

[tacaswell]

The problem seems to be this code path https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/backends/backend_agg.py#L347

If you change

fig.canvas.restore_region(bg)

then it works as expected.

This looks like fall out from the cxx conversion.

[tacaswell]

@mdehoon @jrevans Thanks! Can you check if #4951 fixes this for you?

Sorry I was being cranky.

[mdehoon]

@tacaswell #4951 fixes the problem for tkagg.