Presentation is loading. Please wait.

Presentation is loading. Please wait.

Declarative Programming in Java using JSetL E. PanegaiG. Rossi Dipartimento di Matematica Università di Parma Roma, 21-22 Giugno 2005 Convegno Italiano.

Similar presentations


Presentation on theme: "Declarative Programming in Java using JSetL E. PanegaiG. Rossi Dipartimento di Matematica Università di Parma Roma, 21-22 Giugno 2005 Convegno Italiano."— Presentation transcript:

1 Declarative Programming in Java using JSetL E. PanegaiG. Rossi Dipartimento di Matematica Università di Parma Roma, 21-22 Giugno 2005 Convegno Italiano di Logica Computazionale

2 2 Outline of the talk Aim: Supporting Declarative Programming (DP) in the context of conventional programming languages; Extending Java maintaining its features; A number of examples in DP style; Conclusions and future work. CILC 05

3 3 Introduction Real-world software development is still done using object-oriented programming languages, such as C++ and Java. Complex problems have always complex solutions? Can we declare a solution? CILC 05

4 4 Introduction Main features of DP: –Relations: defining what, and not how, should be computed. –Assignment: not destructive, single value assignment. –Data Structures: explict representation of data structures, such as lists and sets. –Execution Order and Control: do not matter! It’s part of the language. CILC 05

5 5 Introduction The aim is to develop DP features. Two approaches: –Developing new (OO?) languages; –Extending existing languages. CILC 05

6 6 Introduction Extending Java: JSetL Collection of facilities to support DP that allow the language to be used as a highly declarative modelling tool: –Logical variables; –Unification; –Constraint solving; –Partially specified data structures; –Non determinism. CILC 05

7 7 Logical Variables Heart of declarative computational model: –Single assignment; –Values can be of any type; –In JSetL logical variables are Lvar Lvar x = new Lvar(); –Defining and handling dynamic data structures. CILC 05

8 8 Logical Variables Example: Translating the name of the days of the week from Italian to English and vice- versa. public static void weekTranslate (Lvar g, Lvar d) { if (Solver.boolSolve(g.eq("lunedì").and(d.eq("monday")))) return; if (Solver.boolSolve(g.eq("martedì").and(d.eq("tuesday")))) return; … } The use of logical variables and unification allows the same methods to be used for different purposes. CILC 05

9 9 Logical Variables weekTranslate() defines a binary relation. With : Lvar lunedì = new Lvar(“”, “lunedì” ); Lvar monday = new Lvar(“”, “Monday”); there are three legal invocations of the method: 1.weekTranslate(lunedì, answer); 2.weekTranslate(answer, monday); 3.weekTranslate(lunedì, monday); No assumption is made about which are input and which are output parameters. CILC 05

10 10 Data Structures and Partially Specified Values Two distinct kinds of data structures: lists and sets (collections without order and repetitions). Data structures that contain uninitialized logical variables in place of single elements or as a part of the data structure itself represent partially specified values. CILC 05

11 11 Data Structures and Partially Specified Values Check whether a list L contains at least two elements. public static void at_least_two(Lst L) throws Failure { Lvar x = new Lvar(); Lvar y = new Lvar(); Lst R = new Lst(); Solver.solve(L.eq(R.ins1(y).ins1(x))); // L = [x, y | R] return; } CILC 05

12 12 Data Structures and Partially Specified Values Deterministic concatenation: Given two lists L1 and L2 computes the new list L3 as the concatenation of L1 and L2. public static void concat(Lst L1, Lst L2, Lst L3) throws Failure { if (Solver.boolSolve(L1.eq(Lst.empty))) { Solver.solve(L2.eq(L3)); } else { Lvar x = new Lvar(); Lst R = new Lst(); Lst L3new = new Lst(); Solver.solve(L1.eq(R.ins1(x)).and(L3.eq(L3new.ins1(x)))); // L1 = [x | R]  L3 = [x | L3new] concat(R, L2, L3new)); } return; } CILC 05

13 13 Data Structures and Partially Specified Values Check whether all elements of a set S are pairs, i.e., they have the form [x1, x2], for any x1 and x2. public static void all_pairs(Set s) throws Failure { Lvar x1 = new Lvar(); Lvar x2 = new Lvar(); Lvar[] Y = {x1, x2}; Lvar x = new Lvar(); Solver.solve(forall(x, s, Y, x.eq(Lst.empty.ins1(x1).insn(x2)))); return; } CILC 05

14 14 Nondeterministic Operations The ability to state solutions in a nondeterministic way provides a very high- level control abstraction. Set unification and many other set operations are inherently and naturally nondeterministic. CILC 05

15 15 Nondeterministic Operations Compute the maximum of a set of integers s. public static Lvar max(Set s) throws Failure { Lvar x = new Lvar(); Lvar y = new Lvar(); Solver.solve(x.in(s)); Solver.solve(forall(y, s, y.leq(x))); return x; } Declarative reading: an element x of s is the maximum of s, if for each element y of s it holds that y ≤ x. CILC 05

16 16 Nondeterministic Operations Compute all permutations (one at a time) of the set {1, 2, 3, 4}. Set I = new Set("I", I_elems); // I = {1, 2, 3, 4} Set S = Set.mkset(4); // S = {Lvar_1, Lvar_2, Lvar_3, Lvar_4} Solver.solve(S.eq(I)); // {Lvar_1, Lvar_2, Lvar_3, Lvar_4}.eq({1,2,3,4}) System.out.print("Do you want another solution(y/n)?"); … while( input.readLine().equals("y") ) { if( Solver.nextSolution() ){ S.output(); System.out.print("Do you want another solution(y/n)?"); } else { … } } CILC 05

17 17 Nondeterministic Operations public static void tsp(Set edges, Lvar source, Lvar startNode) throws Failure { Lvar nextNode = new Lvar(); Lst newArc = new Lst(Lst.empty.ins1(startNode).insn(nextNode)); Solver.solve(newArc.in(edges).and(nextNode.nin(visited))); visited = visited.ins(nextNode); path = path.insn(nextNode); if (path.size() < n) tsp(edges, source, nextNode); else tsplast(edges, source, nextNode); return; } Traveling salesman problem: Showing the use of nondeterminism in connection with sets. CILC 05

18 18 Constraints DP is often associated with constraint programming. The ability to deal with constraints allows to compute with partially specified values. In JSetL basic set-theoretical operations, as well as equalities, inequalities and integer comparison expressions, are dealt with as constraints. A solver nondeterministically searches for a solution that satisfies all constraints introduced. Constraint solving in JSetL is basically the same developed for CLP( SET ). CILC 05

19 19 Constraints The ability to deal with constraints allows the programmer to compute with partially specified values and ignore the order of execution of statements. x.read(); Solver.add(x.neq(0)); Solver.solve(); is the same as executing: Solver.add(x.neq(0)); x.read(); Solver.solve(); CILC 05

20 20 Constraints Backtracking is confined to constraint solving: Solver.solve(x.in(s)); // x uninitialized Lvar, s = {0,1} x.output(); Solver.solve(x.neq(0)); It’s possible to write parts of a program to be sensible to backtracking defining them as new constraints: public static void printVar(Lvar x) { x.output(); return; } Solver.solve(NewConstraints.printVar(x)); CILC 05

21 21 Intensional Sets With a DP style is possible to represent a set by its intensional definition. Compute the set of all pairs [x,y] such that x and y belong to a given set s and x  y. public static void allPairs(Set S) throws Failure { Lvar X = new Lvar(); Lvar y = new Lvar(); Lst pair = new Lst(Lst.empty.ins1(x).insn(y)); Set Pairs = Solver.setof(pair, x.in(s).and(y.in(s)).and(x.neq(y))); // collect all [x,y] satisfying x  s  y  s  x  y Pairs.output(); return; } CILC 05

22 22 Intensional Sets Compute the set of all prime numbers smaller or equal to a given integer n. public static Set primes(Lvar n) throws Failure { Lvar x = new Lvar(); Solver.add(x.in(Set.interv(2,n))); // x in 2..n // check whether x is prime Lvar y = new Lvar(); Lvar m = new Lvar(); Solver.add(m.eq(x.sub(1))); // m = x - 1 Solver.forall(y, Set.interv(2,m), (x.mod(y)).neq(0)); // compute the set of all prime numbers in 2..n return Solver.setof(x); } CILC 05

23 23 Nondeterminism Between Statements JSetL supplies facilities to build nondeterministic procedures (using backtracking) as new constraints. This allows the user to define completely “invertible” methods (no distinction between input and output parameters). CILC 05

24 24 Nondeterminism Between Statements A fully nondeterministic version of the list concatenation, that can be used also to compute all possible lists l1 and l2 from l3. public static void concat(Lst l1, Lst l2, Lst l3) { eihter { Solver.add(l1.eq(Lst.empty)); Solver.add(l2.eq(l3)); } orelse { Lvar x = new Lvar(); Lst l1new = new Lst(); Lst l3new = new Lst(); Solver.add(l1.eq(l1new.ins(x))); // l1 = [x | l1new] Solver.add(l3.eq(l3new.ins(x))); Solver.add(concat(l1new, l2, l3new)); } return; } CILC 05

25 25 Nondeterminism Between Statements Nondeterministic method to find a path between two nodes in a directed labeled graph G. public static void path(Set G, Lvar source, Lvar dest) throws Failure { either { Lst finalEdge = Lst.empty.ins1(source).insn(dest)); Solver.add(finalEdge.in(G)); } orelse { Lvar x = new Lvar(); Lst intermediateEdge = Lst.empty.ins1(source).insn(x)); Solver.add(intermediateEdge.in(G)); Solver.add(path(G, x, dest); } return; } CILC 05

26 26 Conclusions and Future Work We have shown how facilities provided by the JSetL library can be used to write programs that exhibit a quite good declarative reading. Computation efficiency has not been a primary requirement in the development of JSetL, however: –various improvements to the current implementation are feasible. –Possible to integrate an efficient FD constraint solver without changing the library interface. CILC 05

27 27 Conclusions and Future Work Interesting new possibilities: –more solvers can coexist and cooperate; –solvers can be different threads; –constructors of the solver class can be parametric w.r.t. constraint solving policies. Download at: http://www.math.unipr.it/~gianfr/JSetL/ CILC 05


Download ppt "Declarative Programming in Java using JSetL E. PanegaiG. Rossi Dipartimento di Matematica Università di Parma Roma, 21-22 Giugno 2005 Convegno Italiano."

Similar presentations


Ads by Google