Do we want to rename parallel to mpi in the packaging, to reduce confusion (as there is now two other parallel methods)?
Also, do we want to keep the current "tests", given 1) unrelieable, and 2) only cover the C serial version (the Fortran and parallel versions appear to be commented out, presumably because they keep failing)? Wouldn't be better to write a simple set of tests which solve (for example) the simple harmonic oscillator, and which run against the different features sundials has (c.f.
https://github.com/bmcage/odes/tree/master/scikits/odes/tests)? I can do that when I get back home.