Not sure why the previous author used an arcane incantation rather than just numpy.polyval in _poly_iw(),

Both yours /@repagh's and@swaldherr's (is there a more current GitHub handle?) implementations are very similar. They split the polynomials up into real and imaginary parts after applyingiω, and just store the result in a different variable structure. I just refactored to remove the "duplicate" code.

Coverage increased (+0.2%) to 86.319% when pullingd0f333e on bnavigator:discrete-margins into2d7aad0 on python-control:master.

I think this should go into 0.8.4. It enhances functionality, but does not change the API and it fixes the bug that it returns wrong results with discrete input.