I think you're looking at this backwards from clifanatic's point. The more a higher-level compiler can reason about the code, the more optimizations it can perform, the less the compiler output resembles the source code input I gave it.