Bug report

Bug summary

When enough NaNs are in the data, tricontour and tricontourf would result in a segmentation fault.

Code for reproduction

import numpy as np
import matplotlib.pyplot as plt

x,y=np.meshgrid(np.arange(10),np.arange(10))
z=x**2.+y**2.
z[x<y]=np.nan
fig,ax=plt.subplots()
ax.tricontourf(x.ravel(),y.ravel(),z.ravel())
#ax.tripcolor(x.ravel(),y.ravel(),z.ravel(),vmax=100,vmin=0)
plt.show()

Actual outcome

Segmentation fault (core dumped)

Expected outcome
tripcolor gets a nice picture, no segmentation fault.
Here is the picture if I use ax.contourf(x,y,z) instead:

Matplotlib version

• Operating system: Ubuntu 17.04
• Matplotlib version: 2.1.1+887.g2296775da.dirty
• Matplotlib backend (print(matplotlib.get_backend())):Qt5Agg
• Python version: 2.7.13 and 3.5.3
• Jupyter version (if applicable):
• Other libraries:

I've tried both the deb package installed matplotlib, and the one from source. Also observed on older RHEL7 machine.

When running through gdb, it point to:

Thread 1 "python3" received signal SIGSEGV, Segmentation fault.
TriContourGenerator::follow_interior (this=this@entry=0x555556666c40, contour_line=..., tri_edge=..., end_on_boundary=end_on_boundary@entry=false, level=@0x7fffffffcd10: -1e-13, 
    on_upper=on_upper@entry=false) at lib/matplotlib/tri/_tri.cpp:916
916	        if (!end_on_boundary && _interior_visited[visited_index])