Several issues have been raised about the current behaviour of the "Complex" class, located in package "o.a.c.m.complex" (e.g.
The ensuing discussion revealed various, sometimes incompatible, requirements with focus on efficiency or consistency or backwards compatibility or comparison with other math packages (Octave) or faithfulness to standards (C99x).
It is thus proposed to create several classes, each with a clearly defined purpose.
The consensus seems to be that the first task is to implement a new "BasicComplex" class where the computational formulae (for computing real and imaginary part of a complex) will be applied directly without worrying about limiting cases (NaNs and infinities). Doing so will automatically produce a behaviour similar to the Java double primitive. It is also assumed that it will be the most efficient implementation for "normal" use (i.e. not involving limiting cases).
This task would consist in copying most of the code in the existing "Complex" class over to "BasicComplex". And similarly with "ComplexTest". Then, in "BasicComplex", one would remove all variables that refer to NaNs and infinities together with checks and special assignments, and adapt the unit tests along the way.
A new "ExtendedComplex" class would inherit from "BasicComplex". This class would aim at representing the compactified space of the complex numbers (one point-at-infinity).
A new "C99Complex" class would inherit from "BasicComplex". This class would aim at implementing the C99x standard.