.... well fine that does make more sense that the epicycles I added to make it only drawone extra time.

There is a bunch more logic that should be ripped out and I suspect the failing tests just need to have explicitdraw calls added to force updates.

just to be clear, the tone in my last comment should be mild embarrassment that I did not see that solution.

Only concern is that it is a mild back compatibility break and if people are using a) thepyplot api b)plt.ion c) a non interactive backend this could break things if they do

plt.plot(range(50))plt.gca().get_xlim()# this result changes under this behavior

Only concern is that it is a mild back compatibility break and if people are using a) the pyplot api b) plt.ion c) a non interactive backend this could break things if they do

That's a good point, and exactly the kind of thing I was hoping you'd help find. I suppose what we ought to do isif ion() ... do what we currently do in draw_idle ... else: pass.

Also, I would wonder if animations would break while in an AGG backend? I
can't remember if draw() or draw_idle() is used in animations.

On Tue, Jan 5, 2016 at 10:22 AM, Michael Droettboom <
notifications@github.com> wrote:

Only concern is that it is a mild back compatibility break and if people
are using a) the pyplot api b) plt.ion c) a non interactive backend this
could break things if they do

That's a good point, and exactly the kind of thing I was hoping you'd help
find. I suppose what we ought to do is if ion() ... do what we currently
do in draw_idle ... else: pass.

—
Reply to this email directly or view it on GitHub
#5800 (comment)
.

I've updated this sodraw_idle is only a no-op ifinteractive isFalse.

@WeatherGod: I've confirmed that animation works with this change, both when saving to a file and "live" in TkAgg, Qt4Agg and NbAgg.

@tacaswell: Not sure how to milestone this. It's a pretty important performance improvement, but also hassome potential for accidental consequences given how "core" it is.

I have been thinking about this a bit more recently and coming around todraw_idle should be a no-op for non-interactive backends.

If it is better to make this a no-op here and require that the sub-classes that want to provide itdo or to leave this as-is and make the backends thatdo not want to provide it to over-ride with a no-op is not clear to me yet.

This is very stale. I can't tell if this is a good idea or not...

As far as I can tell, there's no opposition here, only unsure about any broader ramifications. This seems generally like a good idea to me...

The change in backend_bases looks good, and long overdue, though I haven't actually tried it and might be missing something.
I think@jkseppan's suggestion to addplt.ioff() tocleanup is good. Then the last test would not need a special internal function, just aplt.ion().

We have made other changes to the tests that I think actually eliminate most of the changes to the tests.

(bleeding) ~/s/p/matplotlib dont-double-draw↑·18365↓·3|⚑18 Just abit out of date.

I think that this PR pre-dates the ability to push to contributor branches, I'll open a new one.

Closed in favor of#17923