Undefined behavior and C++ optimization

We have just seen one example in the previous section, where, by assuming that every loop in the program eventually terminates, the compiler is able to optimize certain loops and the code containing these loops. The fundamental logic used by the optimizer is always the same: first, we assume that the program does not exhibit UB. Then, we deduce the conditions that must be true in order for this assumption to hold and assume that these conditions are indeed always true. Finally, any optimization that is valid under such assumptions may proceed. The code generated by the optimizer will do something if the assumptions are violated, but we have no way of knowing what it will be (beyond the already mentioned restrictions that it's still the same computer executing some sequence of instructions).

Almost every case of UB documented in the standard can be converted into an example of a possible optimization (whether a particular compiler takes...