That's not a behavior you can count on — clang 3.4 will optimize doSecure() down to "ret" and main() down to "xorl %eax, %eax" + "ret" in both cases, for exampl

That's not a behavior you can count on — clang 3.4 will optimize doSecure() down to "ret" and main() down to "xorl %eax, %eax" + "ret" in both cases, for example. Also, gcc not optimizing out the malloc + memset in the heap case seems like a missed optimization that the gcc devs might fix in the future.