Program simplification as a means of approximating undecidable propositions

We describe an approach which mixes testing, slicing, transformation and formal verification to investigate speculative hypotheses concerning a program, formulated during program comprehension activity. Our philosophy is that such hypotheses (which are typically undecidable) can, in some sense, be 'answered' by a partly automated system which returns neither 'true' nor 'false', but a program (the 'test program') which computes the answer. The motivation for this philosophy is the way in which, as we demonstrate, static analysis and manipulation technology can be applied to ensure that the resulting test program is significantly simpler than the original program, thereby simplifying the process of investigating the original hypothesis.