animation fixes #4995
animation fixes #4995
Conversation
Using draw was triggering infinite recursion, using draw_idle does not.
This was introduced by matplotlib#4800. We need to disconnect the initial draw callback in _start() before doing _init_draw(), otherwise, we end up back in _start() when the draw_event fires. Also go ahead and remove the call to _init_draw() in __init__(), since _start() will handle it.
|
I also cherry-picked @dopplershift 's infinite recursion commit onto this PR just to be safe. |
For blitted animations in is essential to return a list of the animated artists. However, is other cases you can get away with out returning such a list. In the case of blitting it is important to make sure that the animation flags on the animated artist are set correctly, t his commit makes sure that we check that we actually got an iterable before trying to iterate over it.
There are race conditions where _blit_clear may be called before the first draw (which is what will grab the backgrounds)
|
@WeatherGod I think I have fixed the source of the not iterable bugs. All of the animation examples run correctly on this branch. |
|
And all of my examples now run too. There is one example of mine that is still showing a weird off-by-one-frame error, but I haven't yet determined if that is a bug on my part or a bug in matplotlib. This brings everything back to a functional state. |
| try: | ||
| a.figure.canvas.restore_region(bg_cache[a]) | ||
| except KeyError: | ||
| # something has happened out of order? |
There was a problem hiding this comment.
Do we really want to be swallowing this exception silently? Should we at least attach one of our logging stuff to it?
Only attempt to force `set_animated(True)` when blitting is enabled.
|
@WeatherGod I backed out the changes you commented on. I am still not happy with the iterable issue. The underlying issue is that if you are using blitting then the animated artists must have |
|
@WeatherGod What is the history of the animated property? |
|
Without blitting, the _post_draw method uses draw_idle; with blitting, _post_draw calls _blit_draw, which calls ax.figure.canvas.blit. This is odd; I would expect blitting to be done at idle time, just like regular drawing. |
|
@mdehoon I agree that is a bit odd, but I think it has to be this way so that the client code can manage the blitted artists + backgrounds. I agree it would be better if we could push the blitting logic onto the other side of the canvas interface, but that is a bigger project that we can do before 1.5. |
|
You'd have to ask others. It predates me.
|
|
@tacaswell Once the blitting is done on the other side of the canvas interface, I can fix the MacOSX backend to handle blitting. I'm OK with waiting until after 1.5. |
|
@WeatherGod If you are happy with this can you merge it? |
|
Things look good here. Merging... |
Using draw was triggering infinite recursion, using draw_idle does not.
This fixes the recursion depth reported in #4967
attn @WeatherGod