plotting/_compat Version Comparisons Not Working in Py27

[WillAyd]

I noticed this while setting up a virtual environment using matplotlib 1.4.0 to test #18190. __version__ sometimes returns a string and other times returns a unicode object in Python 2.7. When returning unicode, the comparisons that occur in plotting/_compat.py will raise

>>> from distutils.version import LooseVersion
>>> import matplotlib
>>> matplotlib.__version__
u'1.4.0'
>>> matplotlib.__version__ < LooseVersion("1.5")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "~/miniconda3/envs/mpl1_4/lib/python2.7/distutils/version.py", line 296, in __cmp__
    return cmp(self.version, other.version)
AttributeError: 'unicode' object has no attribute 'version'

In some cases, the versions are converted to str objects in plotting/_compat.py to presumably avoid this error, but it is not done consistently. As perhaps a better approach, we could create LooseVersion objects from all the __version__ properties so that the comparison is done between equal types across the board.

>>> LooseVersion(matplotlib.__version__) < LooseVersion("1.5")
True

Output of pd.show_versions()

Python 2.7.14 | packaged by conda-forge | (default, Nov 4 2017, 10:22:41)
[GCC 4.2.1 Compatible Apple LLVM 6.1.0 (clang-602.0.53)] on darwin
Type "help", "copyright", "credits" or "license" for more information.

import pandas as pd
pd.show_versions()

INSTALLED VERSIONS

commit: 2c903d5
python: 2.7.14.final.0
python-bits: 64
OS: Darwin
OS-release: 17.2.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: None.None

pandas: 0.22.0.dev0+283.g2c903d594.dirty
pytest: 3.3.0
pip: 9.0.1
setuptools: 38.2.3
Cython: 0.27.3
numpy: 1.9.3
scipy: None
pyarrow: None
xarray: None
IPython: None
sphinx: None
patsy: None
dateutil: 2.6.1
pytz: 2017.3
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 1.4.0
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: None
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None

[TomAugspurger]

I've seen this error when a library imports unicode_literals.

As perhaps a better approach, we could create LooseVersion objects from all the version properties so that the comparison is done between equal types across the board.

That's probably best anyway.