> but you still cannot do things like “attach this surface of object A to that surface of object B)”

Sure you can, but openscad is an imperative language so you need to do it in an imperative manner.

    a_cube_loc = [10, 5, 5];
    a_cube_size = [2, 4, 1];
    b_cube_loc = [a_cube_loc[0], a_cube_loc[1] + a_cube_size[1], a_cube_loc[2] ];
    b_cube_size = [2, 4, 1];

   translate(a_cube_loc) cube(a_cube_size);
   translate(b_cube_loc) cube(b_cube_size);

It would be nice to have more declarative constraints. The solvespace file format is plain text and it almost feels like you could write it by hand, but that would be a lot of manual record keeping. and you would loose all that imperative goodness. Perhaps you could have an imperative layer(say python or lisp or forth) that outputs the declarative layer(solvespace) and then solvespace renders(picture or stl) the declarative layer.