Opening an issue on this to track progress and get insight from other people.

I have just been playing with tests on my own system, and I have discovered that the tests don't run fully independently of each other...

So far, I have only discovered problems with different backends getting used, for example in test_backend_qt4.py we have a few tests that have :

plt.switch_backend('Qt4Agg')

test_backend_pgf.py seems to have the solution to this with a generic decorator, and it feels odd that such a beautifully generic decorator appears in a very specific class and doesn't get used elsewhere.

Anyway I shall make a PR to correct this backend thing and hope we have no other inter-dependancies, but in the meantime, some questions:

1. Why don't we reset everything at the beginning of each test anyway, we have a setup() function that should do this?
2. Do we have an easy way to print debug information for tests? I.e. specific rcParams? This will help debugging future issues like this (touch wood we don't get any).