Conda on travis - #4114

Sure. I consider this a little lower priority than originally thought, since it does not decrease testing times. @jenshnielsen made the valid point that we could run the Qt tests if we did this, but that will require further refinement.

On the other hand, pip has been being really flaky on travis recently and our use of --use-mirrors is apparently deprecated so we will have to deal with that eventually.

Closing this as the .travis.yaml file has diverged significantly. This is still probably worth doing (it will get us the numpy compile time back) but the docs are now the bottle neck anyway.

There are numpy wheels here if you want to reduce compile time.

@matthew-brett I know I have asked this N times before, but can you point me to directions on how to use them?

BTW It looks like travis already has Numpy preinstalled on there venvs and we are only installing if from source in the python26/numpy16 env Compare https://travis-ci.org/matplotlib/matplotlib/jobs/62790222 to https://travis-ci.org/matplotlib/matplotlib/jobs/62790223

Here's a PR to use travis wheels for installation:

#4435