use base64.encodestring on python2.7

[tomoemon]

This PR fixes #5508.
I don't know the standard way to solve version compatibility issues in this project. So I chose explicit way.

I've tested a code below with IPython notebook on Python2.7.5 and Python3.4.2.

get_ipython().enable_matplotlib("inline")

import os
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
mpl.style.use('ggplot')

mpl.rcParams['animation.html'] = 'html5'
mpl.rcParams['animation.ffmpeg_path'] = os.path.expanduser("~/bin/ffmpeg")

fig = plt.figure()
x = np.arange(500)
frame_list = []

for a in range(10):
    frame = plt.plot(x, a * x ** 2, lw=2)
    frame_list.append(frame)

from matplotlib import animation
animation.ArtistAnimation(fig, frame_list, interval = 500)

[jenshnielsen]

This break another test in test simplification because it checks for the presence of encodebytes.

I would prefer that the code does not monkeypatch the stdlib modules. I would suggest something like

try:
    from base64 import decodebytes
except:
   from base64 import decodestring as decodebytes

and then change L935 to

vid64 = encodebytes(video.read())

[jenshnielsen]

The pep8 test fails because you need at least 2 spaces before an inline comment

 E261 at least two spaces before inline comment

[tomoemon]

@jenshnielsen Thank you for you comment and sorry for my poor testing. I've fixed my commit, but some tests are failing by other reasons. Do you know why that tests fail?

[jenshnielsen]

Great. The new test failures are just random flukes not related to your changes. I have restarted these tests

[WeatherGod]

Hmm, coveralls isn't happy for some reason. Shall we ignore this for now? I think we can improve test coverage of the animation module by using the new NullWriter as a smoke tester later.

[jenshnielsen]

@WeatherGod Yes ignore that. This code is uncovered. Hence why we did not find it earlier and now it's one line longer so the coverage goes down 😢

[WeatherGod]

backported to v1.5.x as 2baedb0

[WeatherGod]

Hmm, not exactly sure, but I am wondering if 1.5.x is now broken for py2.6? I don't think this PR broke it, but for some reason, one of the usetex unittests is breaking for py2.6...