Bug summary

Currently, Line2D doesn't copy its inputs if they are already arrays. Most of the time, in-place modifications to the input arrays do not affect the draw line, because there is a cache that doesn't get invalidated, but in some circumstances, it is possible for these modifications to affect the drawn line.

Instead, Line2D should just copy its inputs. This was rejected in #736 on a memory-saving argument, but note that AxesImage (which would typically have much bigger (2D) inputs than Line2D (which has 1D inputs)) does a copy, which if anything is much worse memory-wise.

Code for reproduction

from pylab import *
t = arange(0, 6, 2)
l, = plot(t, t, ".-")
savefig("/tmp/1.png")
t[:] = range(3)  # in place change
savefig("/tmp/2.png")  # no effect
l.set_drawstyle("steps")  # ... unless we trigger a cache invalidation
savefig("/tmp/3.png")  # in fact, only the x array got updated, not the y

Actual outcome

(1)

(2) (same as (1))

(3) (different, but only x got updated, not y)

Expected outcome

Modifying t a posteriori should not affect the Line2D. Compare e.g. with AxesImage:

im = arange(9).reshape(3, 3)
imshow(im)
savefig("/tmp/4.png")
im[:, :] = im[::-1, ::-1]
savefig("/tmp/5.png")

Both images are identical.

Operating system

linux

Matplotlib Version

3.5b1

Matplotlib Backend

mplcairo

Python version

39

Jupyter version

No response

Other libraries

No response

Installation

source

Conda channel

No response