This isn't a problem with comparison by subtraction, it's a problem with integer overflow. There's nothing wrong with comparison via subtraction if you account for overflow, which you always need to do in programming.
In mainstream compilers, even on x64, sizeof(long) == sizeof(int). You need long long instead. Of course, in the general case there's exactly zero guarantee that there even exists an integral type wider than int.
On a (slightly) more practical level: there exists some largest integer type. You may someday need to sort it. If you want your sort to always work, you can't afford to overflow.
