The out-of-thin-air (OOTA) properties of C11 and C++11 memory_order_relaxed accesses have resulted in considerable consternation over the years, despite the fact that no one has managed to construct a real system that exhibits OOTA behaviors. And yet attempts to create memory models that exclude OOTA behaviors have produced results that have either been non-algorithmic, overly complex, or unloved by compiler implementers, and mostly all three.

This rant argues that compilers must respect semantic dependencies. By this it is not meant that I will personally scream at compiler implementers until they buckle down and do what I ask, tempting though such a course of action might be. It rather instead means that it is not possible to build a correct C or C++ implementation that exhibits OOTA behaviors in real-world conditions, courtesy of a couple of fundamental laws of physics, longstanding hardware practices, and existing formal (atemporal) memory models.