"size_t can store a pointer, it is better to use another unsinged integer type uintptr_t for that purpose " So he is actually recommending the unsigned version. Although there are some people (like the Google guidelines for C++) that recommend not using unsigned types because of overflows (which is arguable but it is a point)

I thought using signed indexes in loops allow GCC to perform some optimizations because signed overflow is undefined.

ptrdiff_t is the wrong type for exactly this reason. OP's article is... not good.