Currently the textureLoad family of function use signed integer for coordinates, for example this is one of the overloads:

vec4<type> textureLoad(texture_1d , i32 coords, i32 level_of_detail)

My understanding is that if any of the components of coords is negative or if level_of_detail is negative, then the access is an out-of-bounds access that will need some form of handling. In a similar spirit to making array indices unsigned because it saves a comparison, what do you think of making any texture coordinate that can cause OOB an unsigned integer?