(I wrote this for #7, but didn't post before you closed it ;)

I wrote the test that's failing on Travis, though it doesn't fail without this mod; compare recent
https://travis-ci.org/roryyorke/Slycot/builds/153037035
which is jgoppert/master with Python 3.4 and 3.5 added, and
https://travis-ci.org/roryyorke/Slycot/builds/153038553
which is the same thing with your #7 merged.

(The Python 3.2 failure appears to be unrelated; I reckon we should drop 3.2.)

On the interface:

1. the help string doesn't appear to match the argument declaration:

def sb03od(n,m,A,Q,B,dico,fact='N',trans='N',ldwork=None):
    """  U,scale,w = sb03od(dico,n,n,A,Q,B,[fact,trans,ldwork])

2. Can n and m not be inferred from A and B? (though I see this is not done in other Slycot routines, e.g., sb10ad)

I think it would be worth adding simple unit tests, one for the continuous and one for the discrete case.

Oh, and I can't reproduce that failure on my system with conda builds with Python 3.4.

I've updated the help string and I suppose n and m could be inferred from A and B, but I decided to stay consistent with other routines. I don't know why tests are failing on code that I didn't modify.