With suggest.el, I've learnt generative programming suits interpreted PLs. The problem is exponential, so extra perf doesn't really help.