That optimization fudges the difference between boolean or (comparing truth values) and bitwise or. It happens to work under the safe assumption that value1, value2, and value3 use sign bits (as all two's-complement integers do), but logically it's a disaster.
monocasa isn't saying this should be in source code, but rather the compiler should create machine code that does this. Compilers know the types, so they will know whether they're signed (if they're not signed, then < 0 can be optimized out as impossible). Compilers have no goal of creating assembly that would look good in source code.
