Up till now, it is the decision of the compiler, whether to evaluate those expressions in compile-time or runtime. This was chosen on purpose. "must be a compile-time constant." would contradict this philosophy. Even __builtin_constant_p keeps the compiler's freedom not to detect compile-time constants, if it chooses not to so or if it would take too many resources. If you want to standardize a different behaviour it would take a lot more of a reasoning than just a simple: It would be useful. You would have to understand the reasoning behind the current behaviour and balance the arguments.