Stem is slow and will crash if I try to close the window

[goldfita]

Bug summary

Stem will crash if you try to close the window while it's processing. It's also very slow for what I would consider a relatively small number of data points. I'm using the code from Julia, but I was informed the problem is Matplotlib. See original discusion.

Code for reproduction

using PyPlot
stem(randn(32768))

Actual outcome

This will work if you wait about half a minute. If you interfere by trying to close the window, the application crashes (Julia in this case).

Matplotlib version

I don't know. I'm using the most recent version of PyPlot and PyCall in Julia 0.5. Conda installed Python 2.7.

[tacaswell]

Can you please provide python code that generate the issue? As the julia folks point at us, it is not clear that this is not a bug on the julia side.

[goldfita]

I tried just now. I'm afraid I'm too much of a beginner with both Julia and Matplotlib to figure this out. However, poster jheinen on the Julia forum claims to have reproduced this problem in pure Python. I suggest asking him for his code. I'm sorry I can't be of any more help. I don't want to get in the middle of any finger pointing between these two communities.

[tacaswell]

It is not finger pointing, just sorting out what the actual bug is.

attn @jheinen

[jheinen]

This is how you can reproduce the bug (in Python)

import matplotlib.pyplot as plt
import numpy as np
plt.stem(np.random.randn(10000))

Fails on macOS X, can't check other backends before Monday ...

[anntzer]

Indeed, this is likely due to the fact that stem generates one Line2D per stem, rather than grouping everything in a LineCollection, which would probably improve the performance quite a bit.

[anntzer]

(but LineCollection currently cannot take a custom marker :-()

[tacaswell]

At a minimum, we could save some time by not going through plot

[dstansby]

Having a look at this and think I'm making progress

[terriyu]

I also confirmed this bug in Python 3 on Ubuntu 16.04. I tried to plot a stem plot with 130,000 points and the computer crunched on it a long time without displaying anything.

[grr37213]

Since I encountered this problem by plotting around 30000 data points today I found this issue.
I can only link to this workaround I found, which solved my problem through manually calculating the line, coordinates and plotting a line collection, instead of using stem itself.
https://stackoverflow.com/questions/48301044/optimization-of-matplotlib-stem-plot
Maybe it helps someone.

[racinmat]

Will this be merged to some 3.0.x version, or do we need to wait for stable 3.1.x version if we want this fixed in pip?

[dstansby]

I'm afraid this is only going to be in 3.1.x onwards.

[timhoffm]

But 3.1.0 should hopefully come out soon 😄 .

[tacaswell]

Hopefully today or tomorrow.