w.r.t boost HEAD and GCC 4.2.1 in function find_nothrow (line 298) the following fragment appear: if (approximate_matches.size() > 1) boost::throw_exception( ambiguous_option(name, approximate_matches)); else return found.get(); GCC issues a warning about control flow reaching the end of a non-void function. boost::throw_exception is visible to the compiler and it could determine that it _always_ throws. But GCC doesn't. For GCC (and probably the Intel compiler) it would be possible to attach an __attribute__((noreturn)) to the declaration of throw_exception, but this wouldn't be portable. Would it be acceptable to return a fake pointer after the call to throw_exception? We know we'd never get there and the warning would be turned off, a big plus for those using boost in a local directory and insisting on compiling with -Wall and -Werror. Best regards, Maurizio