How does that fix the problem of bit-casting long double to __int128? If the destination type was wider, you couldn't bit-cast (constraints not satisfied). The destination type has no padding bits, yet the cast is an alternative spelling for std::unreachable().