Towards a Practical Full Mode Inference System for CLP(H, N)

A full mode analysis system for CLP(H,N) could result from a straightforward combination of the deeniteness analysis (developed at U.P.M.) and the freeness analysis (developed at K.U.Leuven). However, the computational complexity of the freeness abstraction, which is due to an exhaustive enumeration of possible dependencies between variables, impedes this. This paper proposes two orthogonal approaches to reduce the size of the freeness abstraction. Using the deeniteness information the rst approach splits oo the deenite variables and their dependencies without any loss of precision. The second approach retains only basic freeness information { the minimal information { from which a safe approximation of the rest can be reconstructed. The combination of both approaches results in a practical full mode analysis system; to our knowledge this is the rst full mode analysis system for CLP. The optimisations are implemented and timings of the resulting systems show their eeectiveness for a set of benchmark programs.