numpy-stl: mesh.from_file throws Assertion Error if matplotlib figure is open when speedups=True

I run into troubles loading .stl files with stl.Mesh.from_file when there is some instance of matplotlib or Qt running in the same process. That is

import stl
import matplotlib.pyplot as plt

mesh=stl.Mesh.from_file(pathToStlFile)

works great. However, if before initiate a matplotlib figure, that is

plt.figure()
mesh=stl.Mesh.from_file(pathToStlFile)

I get the following Assertion Error:

 File "/usr/local/lib/python2.7/dist-packages/stl/stl.py", line 316, in from_file
    fh, mode=mode, speedups=speedups)
  File "/usr/local/lib/python2.7/dist-packages/stl/stl.py", line 85, in load
    name, data = cls._load_binary(fh, header, check_size=True)
  File "/usr/local/lib/python2.7/dist-packages/stl/stl.py", line 102, in _load_binary
    count, MAX_COUNT)
AssertionError: File too large, got 909389366 triangles which exceeds the maximum of 100000000

Looking through the code in BaseStl and reading through this issue https://github.com/WoLpH/numpy-stl/issues/37, I realized that this is due to the fallback if _load_ascii throws a runtime error such as

 File "/usr/local/lib/python2.7/dist-packages/stl/stl.py", line 212, in _load_ascii
    return _speedups.ascii_read(fh, header)
  File "stl/_speedups.pyx", line 132, in stl._speedups.ascii_read (stl/_speedups.c:2613)
RuntimeError: (0, "Can't read vertex (4:vertex 176.46 176.46 0)")

So it seems that somehow the speedup is not working properly (some Cython clash with Qt/matplotlib?). Turns out that

plt.figure()
mesh=stl.Mesh.from_file(pathToStlFile,speedup=False)

works great. This took me a while to figure out and I thought it might be interesting.