Coverage remained the same at 77.329% when pulling a7e0bba on testing into 5ab74cf on master.

Changes here increase runtime for CI builds by approximately 90 seconds (as a sum across all configurations: Python 2.7, 3.3, 3.4, 3.5).

There is at least one negative consequence of changes here. Tests that are independent of Slycot are performed twice without potential benefit. However, the marginal gain in computation time and load is small.

Looks good. I assume you saw all the deprecation warnings from scipy.

FWIW, an alternative is to use Travis CI's matrix capability; I imagine "with Slycot" and "without Slycot" configurations. Using a matrix setup opens other possibilities:

1. a "no-conda" config---I don't know how useful this is, but I gather from express dependencies in setup.py Slycot#3 that this could be desired.
2. Mac OS testing. I haven't explored this (don't have a Mac), but I think Travis CI does support it.

This PR looks good to me. Will merge in 24 hours if there are no further comments.

@slivingston: leaving it to you to delete the branch (just in case there is some reason to keep it).

@roryyorke and @murrayrm thanks to both of you for reviews.

re #133 (comment):

1. Yes, I noticed the deprecation warnings, but I think that it is outside the scope of this pull request. My plan is to add more tests while addressing issue Organize unit tests to follow the NumPy/SciPy guidelines #95 before making changes to avoid deprecated external API.

2. Before opening this PR, I tried testing on Mac OS as part of a matrix setup on Travis CI, e.g., https://travis-ci.org/python-control/python-control/builds/201547383. Ultimately I did not succeed. Because Slycot is the main difficulty in terms of cross-platform support, I decided to devote more effort to configuring Mac OS on Travis CI for Slycot instead of control.