ax.fill broken for log scale and values below zero

When using ax.fill_between in log scale plots with y-values below zero, the result looks broken and areas are filled that should not be filled

Code for reproduction

import numpy as np
import matplotlib.pyplot as plt
xdata=np.arange(0,10,0.01)
ydata = np.exp(-xdata)
edata = 0.2*(10-xdata)*np.cos(5*xdata)*np.exp(-xdata)

fig, ax = plt.subplots()
plt.yscale('log')

ax.fill_between(xdata, ydata - edata, ydata + edata)

plt.savefig("test.png")

Actual outcome

Expected outcome

The correct example has been generated with matplotlib-2.0.2-1

Probably related: #8623
How to deal with invalid values in log scale plots is certainly a matter of definition. However, it should never happen that areas are filled that should not be filled, as it is the case in the example above.

Matplotlib version

• Operating system: arch linux
• Matplotlib version: 2.1.0-1
• Matplotlib backend: Qt5Agg
• Python version: 3.6.2
• Other libraries: numpy version: 1.13.3-1

Everything installed using the package manager of arch linux.