This paper presents an algorithm that combines modular synthesis for extended finite-state machines (EFSM) with abstraction of variables by symbolic manipulation, in order to compute least restrictive controllable supervisors. Given a modular EFSM system consisting of several components, the proposed algorithm synthesises a separate supervisor for each specification component. To synthesise each supervisor, the algorithm iteratively selects components (plants and variables) from a synchronous composition until a least restrictive controllable solution is obtained. This improves on previous results of the authors where abstraction is only performed by the selection of components and not variables. The paper explains the theory of EFSM synthesis and abstraction and its algorithms. An example of a flexible manufacturing system illustrates how the proposed algorithm works to compute a modular supervisor.