Thanks @fmder.

Can you let us know which backend your using?:

import matplotlib
print matplotlib.rcParams['backend']

Cheers,

@fmder Is this still a problem on newer version of mpl? (I don't have a mac set up to test this)

I can still reproduce this with today's master. I suspect this is due to some strange limitation of the Quartz rendering library.

@mdehoon: Any thoughts?

I've seen this before with Quartz. I believe the problem is that when CGContextSetLineDash is set the underlying stroke creates many individual lines in the rendering. So, on a very long line, Quartz takes forever to draw the enormous amount of lines created from the dash pattern.

In this case with the exponential data set and the limits set as so, the lines in terms of pixels are insanely big and I bet the time for Quartz to finish would be something completely absurd.

It's indeed CGContextStrokePath that is taking forever to finish.
The arguments that renderer.draw_path receives are:

path = Path([[ 0.00000000e+00 1.00000000e+00]
[ 1.00000000e+00 2.00000000e+00]
[ 2.00000000e+00 4.00000000e+00]
[ 3.00000000e+00 8.00000000e+00]
[ 4.00000000e+00 1.60000000e+01]
...
[ 4.60000000e+01 7.03687442e+13]
[ 4.70000000e+01 1.40737488e+14]
[ 4.80000000e+01 2.81474977e+14]
[ 4.90000000e+01 5.62949953e+14]], None)

transform = Affine2D(array([[ 9.92, 0. , 80. ],
[ 0. , 3.84, 48. ],
[ 0. , 0. , 1. ]]))

This suggests that we are trying to draw the full line instead of just the section that will be visible in the window. Perhaps the calling function (_ draw_dashed in matplotlib/lines.py, called as part of the draw method of Line2D) should only draw the section of the line that will be visible, rather than the whole thing? Currently it seems to rely on clipping in order to show the visible section only.

Closed by #6178.