trying to second-guess users as to their expectations is part of a compiler's job, and that's what warnings are often for. but the primary mission of a compiler is to translate the source code to object code; an optimizing compiler adds to that making the object code run fast or be compact or somesuch, still according to the meaning of the program according to the language specification. second-guessing the user with -fdo-what-I-mean :-) enabled by default is usually a recipe for trouble, because those who understand the language and take advantage of its features to get better code end up unhappy, and most users who don't know the darkest corners of the language in such amount of detail still end up unhappily surprised because incorrect expectations are just too varied. it's tough, but and it's easy to argue for "just this one case" while missing the point of all the other "just this one case" that would be just as deserving