I (not the person you are replying to) do understand that's how compilers interact with UB. However, a wealth of experience has shown us that the assumption "UB doesn't occur" is completely false. It is, in my opinion, quite irresponsible for compiler writers to continue to use a known-false assumption when building the optimizer. I don't really care how much speed it costs, we need to stop building software on a shaky foundation like that.