Most test code is constructing test data, and pulling out the data
to test afterwards. It's worth writing reusable libraries of this from
the beginning! Since haskell lacks introspection, you can't write a
generic diff, so when the data is large you must strip irrelevant data.
E.g. Perform.PitchSignal_test, 'sig' constructor vs. 'unpack'.
Derive.Schema_test.test_compile