Programming in Alma-0, or Imperative and Declarative Programming Reconciled

04/05/2000 ∙ by Krzysztof R. Apt, et al. ∙ 0

In (Apt et al, TOPLAS 1998) we introduced the imperative programming language Alma-0 that supports declarative programming. In this paper we illustrate the hybrid programming style of Alma-0 by means of various examples that complement those presented in (Apt et al, TOPLAS 1998). The presented Alma-0 programs illustrate the versatility of the language and show that "don't know" nondeterminism can be naturally combined with assignment.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Logic programming languages, notably Prolog, rely on two important features: nondeterminism and unification. The form of nondeterminism used is usually called “don’t know” nondeterminism. According to it some path in the computation tree should lead to a correct outcome.

There have been some efforts to incorporate this form of nondeterminism into the imperative programming paradigm. For early references see [Cohen1979]. More recent examples are the languages Icon of [Griswold and Griswold1983]) and SETL of [Schwartz et al.1986].

In [Apt et al.1998] we pursued this approach to programming by proposing another, simple, imperative language Alma-0 that supports this form nondeterminism.

Our rationale was that almost 25 years of experience with logic programming led to an identification of the programming techniques that make it a distinct programming paradigm. The imperative programming constructs that support nondeterminism should support these programming techniques in a natural way.

And indeed, we found that a number of logic programming jewels could be reproduced in Alma-0 even though unification in the language is limited to bare minimum and the language offers no support for symbolic programming.

But we also found that other programs, such as the solution to the Eight Queens problem, could be coded in Alma-0 in a more natural way than the logic programming paradigm permits. Also, some programs, such as the solution to the Knapsack problem, seem to be very natural even though they use both nondeterminism and assignment.

So the hybrid programming style of Alma-0 apparently calls for new programming techniques that need to be better understood and explored. This is the aim of this paper that can be seen as a companion article of [Apt et al.1998].

To this end we provide here a number of Alma-0 programs that show versatility of the language and provide further evidence that the constructs of the language encourage a natural style of programming. In particular, Alma-0 programs without assignment are declarative in the sense that they admit a dual reading as a logic formula.

It should be clarified that in general two types of nondeterminism have been considered in programming languages, “don’t know” nondeterminism and “don’t care” nondeterminism. According to the latter one each path in the computation tree should lead to a correct outcome. This form of nondeterminism is present in the guarded command language of [Dijkstra1975]. It leads to different issues and different considerations.

The paper is organized as follows. In Section 2 we recall the basic elements of Alma-0. In the remainder of the paper we provide selected examples of Alma-0 programs that complement those presented in [Apt et al.1998] and illustrate its use in different contexts. More specifically, in Section 3 we present two versions of a classical graph traversal problem, namely the longest path problem. In Section 4 we show how a typical feature of the logic programming paradigm, namely negation as failure, can be also profitably exploited in Alma-0. Next, in Section 5 we illustrate how executable specifications can be written in Alma-0. In Section 6 we provide a more complex example of Alma-0 programming by describing a solution to a classical scheduling problem. Finally, in Section 7 we draw some conclusions and describe the current status of the Alma project.

2 The language Alma-0

Alma-0 is an extension of a subset of Modula-2 that includes nine new features inspired by the logic programming paradigm. We briefly recall most of them here and refer to [Apt et al.1998] for a detailed presentation.

  • Boolean expressions can be used as statements and vice versa. A boolean expression that is used as a statement and evaluates to FALSE is identified with a failure.

  • Choice points can be created by the two nondeterministic statements ORELSE and SOME. The former is a dual of the statement composition and the latter is a dual of the FOR statement. Upon failure the control returns to the most recent choice point, possibly within a procedure body, and the computation resumes with the next branch in the state in which the previous branch was entered.

  • The created choice points can be erased or iterated over by means of the COMMIT and FORALL statements. COMMIT S END removes the choice points created during a successful execution of S. FORALL S DO T END iterates over all choice points created by S. Each time S succeeds, T is executed.

  • The notion of initialized variable is introduced and the equality test is generalized to an assignment statement in case one side is an uninitialized variable and the other side an expression with known value. The KNOWN relation is introduced to test whether a variable of a simple type is initialized.

  • A new parameter passing mechanism, called call by mixed form, is introduced for variables of simple type. It works as follows: If the actual parameter is a variable, then it is passed by variable. If the actual parameter is an expression that is not a variable, its value is computed and assigned to a new variable (generated by the compiler): it is that is then passed by variable. So in this case the call by mixed form boils down to call by value.

    This parameter mechanism, denoted by MIX, is introduced to allow us to pass both values and uninitialized variables as actual parameters.

To clarify these extensions and Alma-0 programming style consider the following problem from [Gardner1979].

Problem 1

Ten cells numbered inscribe a 10-digit number such that each cell, say , indicates the total number of occurrences of the digit in this number. Find this number.

Here is a simple solution to it in Alma-0.

MODULE tendigit;
VAR i, j, k, l, count, sum: INTEGER;
    a: ARRAY [0..9] OF INTEGER;
BEGIN
  FORALL
    sum := 0;
    FOR i := 0 TO 9 DO
      SOME j := 0 TO 10-sum DO
        a[i] = j;
        sum := sum + j
      END;
    END;
    sum = 10;
    FOR k := 0 TO 9 DO
      count := 0;
      FOR l := 0 TO 9 DO
        IF a[l] = k THEN count := count + 1; a[k] >= count END;
      END;
      a[k] = count
    END
  DO
  FOR i := 0 TO 9 DO WRITE(a[i]) END
  END
END tendigit.

To better understand this program first note that any 10-digit number that is a solution to this problem has the property that the sum of its digits is 10.

Now, the first FOR loop nondeterministically generates 10-digit numbers, written as an array, with this property. This is done by means of a SOME statement. The equality a[i] = j is used here as an assignment, while the equality sum = 10 is used as a test.

The second FOR loop tests whether a candidate array is a possible solution. The testing can be abandoned if for some k the count exceeds the value a[k]. This explains the use of the test a[k] >= count.

The above described code is within the FORALL statement, so all solutions to the problem are generated and each of them is printed. The program yields the unique solution, namely 6210001000.

The still unexplained features of Alma-0 will be discussed later.

3 Graph Traversal

We now illustrate by means of two examples how Alma-0 can be used in a natural way for graph-related problems.

3.1 Knight’s Tour

We begin with the following well-known problem.

Problem 2

Find a knight’s tour on the chess board in which each field is visited is exactly once.

Here is a solution in Alma-0.

MODULE KnightTour;
CONST
  N = 5;
TYPE
  [1..N] = [1..N];
  Board = ARRAY [1..N], [1..N] OF [1..N*N];

PROCEDURE Next(VAR row, col: INTEGER);
VAR i, j: INTEGER;
BEGIN
  EITHER i = 2;  j = 1
  ORELSE i = 1;  j = 2
  ORELSE i = -1; j = 2
  ORELSE i = -2; j = 1
  ORELSE i = -2; j = -1
  ORELSE i = -1; j = -2
  ORELSE i = 1; j = -2
  ORELSE i = 2; j = -1
  END;
  row := row + i;
  col := col + j;
  (1 <= row) AND (row <= N);
  (1 <= col) AND (col <= N)
END Next;

VAR i, j, k: INTEGER;
  x: Board;
BEGIN
  x[1,1] = 1;
  i = 1; j = 1;
  FOR k := 2 TO N*N DO
    Next(i,j);
    x[i,j] = k
  END;
  Print(x)
END KnightTour.

Here the Next procedure nondeterministically generates the coordinates of the next field, given the current one. This is done now by means of an ORELSE statement that explores all eight possibilities in turn.

After a call to Next the (implicitly) incremented value of k is assigned to this new field. Note that this assignment, a[i,j] = k, is performed by means of an equality. This is crucial, as it also prevents that a field is visited again. Indeed, if this is the case then a[i,j] has already a value and the equality fails. In this case the backtracking takes place and the next, if any, candidate field is generated.

3.2 Longest Path

In the Knight’s tour problem the chess board can be viewed as a graph in which the squares are the nodes and the possible knight moves are the arcs. In this way the knight tour problem accounts to finding a simple path of maximal length. The length of this path equals , the number of nodes.

Consider now a more general problem of finding the longest path in an arbitrary directed graph.

Problem 3

Given a directed graph and two nodes find the longest simple path that starts in and ends in .

Recall that this decision problem is NP-complete (see [Garey and Johnson1979, problem ND29, page 213]).

We assume that the graph is represented by its adjacency matrix. We also employ an array for marking the visited nodes and for storing the current longest path. In what follows we use the following type declarations.

  Graph = ARRAY [1..N],[1..N] OF BOOLEAN;
  PathMark = ARRAY [1..N] OF INTEGER;

The basic building block that we use for traversing the graph is the following function Successor that upon backtracking generates all successors of a given node. The function fails if the node has no successor.

PROCEDURE Successor(G: Graph; X: Node): Node;
VAR i: Node;
BEGIN
  SOME i := 1 TO N DO
    G[X,i]
  END;
  RETURN i
END Successor;

The following procedure LongestPath consists of some initializations followed by a FORALL loop that explores all possible paths. Inside the FORALL loop, each path is constructed by an inner loop that searches exhaustively for unvisited successors until it gets to the requested final node.

In contrast to Problem 2, we do not know the length of the longest path in advance. Therefore we use here a WHILE statement rather than a FOR statement for constructing the path. In addition, for each generated path we need to check its length against the currently longest one.

A node X is viewed as unvisited as long as Path[X] = 0. When X is visited, Path[X] gets the value k which represents the position of X in the path.

PROCEDURE LongestPath(G: Graph; InitNode, FinalNode: Node): PathMark;
VAR k, max: INTEGER;
    i: Node;
    Path, LongPath: PathMark;
BEGIN
  FOR i := 1 TO N DO Path[i] := 0 END;
  i := InitNode;
  k := 0;
  max := 0;
  FORALL
    WHILE (Path[i] = 0) AND (i <> FinalNode) DO
      k := k+1;
      Path[i] := k;
      i := Successor(G,i) (* generate a successor
                             nondeterministically *)
    END
  DO
    IF (i = FinalNode) AND (k > max)
    THEN max := k; LongPath := Path END
  END;
  RETURN LongPath
END LongestPath;

The longest path is delivered by means of the return value of the procedure. If no path between InitNode and FinalNode exists, then the variable LongPath remains uninitialized, and thus the value returned is also an uninitialized array, which can be tested within the calling procedure by using the built-in procedure KNOWN.

4 Use of Negation

One of the important notions in logic programming is negation by failure. It is, in a nutshell, a meta-rule that allows us to conclude a negation of a statement from the fact that it cannot be proved (using the resolution method used in logic programming). Negation by failure is a very useful concept that allows us to write some remarkably concise Prolog programs. Also, it supports non-monotonic reasoning. Actually, the negation by failure mechanism provides a computational interpretation of the latter, a feature other main approaches to non-monotonic reasoning lack.

Negation by failure is supported in Alma-0, as well. In fact, as in logic programming, it is the mechanism used to evaluate negated statements. Consequently, we can use it in Alma-0 in the same way as in logic programming and Prolog.

In [Apt et al.1998] we already presented a number of programs that used negation. Here we show an Alma-0 solution to the proverbial Tweety problem, one of the classical benchmarks for non-monotonic reasoning. Let us recall it.

The problem is to reason in the presence of default assumptions. In the natural language they are often expressed by means of the qualification “usually”. In what follows the “usual” situations are identified with those which are not “abnormal”.

We stipulate the following assumptions.

  • The birds which are not abnormal fly (i.e., birds usually fly).

  • The penguins are abnormal.

  • Penguins and eagles are birds.

  • Tweety is a penguin and Toto is an eagle.

The problem is to deduce which of these two birds flies. Here is a solution in Alma-0, where the code for Print is omitted.

MODULE penguin;
TYPE Animal = (Tweety, Toto);

PROCEDURE penguin(MIX x: Animal);
BEGIN
  x = Tweety
END penguin;

PROCEDURE eagle(MIX x: Animal);
BEGIN
  x = Toto
END eagle;

PROCEDURE ab(MIX x: Animal);
BEGIN
  penguin(x)
END ab;

PROCEDURE bird(MIX x: Animal);
BEGIN
  EITHER penguin(x) ORELSE eagle(x) END
END bird;

PROCEDURE fly(MIX x: Animal);
BEGIN
  bird(x);
  NOT ab(x)
END fly;

VAR x: Animal;
BEGIN
  FORALL fly(x)
  DO Print(x)
  END
END penguin.

The use of the MIX parameter mechanism allows us to use each procedure both for testing and for computing, as in Prolog. In particular, the call fly(x) yields to a nondeterministic computing of the value of x using bird(x) and subsequent testing of it using NOT ab(x).

It is instructive to compare this program with the more compact Prolog program (see, e.g., [Apt1997, page 303]):

penguin(tweety).
eagle(toto).
ab(X) :- penguin(X).
bird(X) :- penguin(X).
bird(X) :- eagle(X).
fly(X) :- not ab(X), bird(X).

While logically both programs amount to equivalent formulas we see that it is difficult to compete with Prolog’s conciseness.

Other natural uses of negation in Alma-0 can be found in some other programs in this article.

5 Executable Specifications

The next example shows that in some circumstances Alma-0 yields programs that are more intuitive than those written in Prolog.

In general, specifications can and do serve many different purposes. The issue whether specifications should be executable or not has been for a long time a subject of a heated discussion, see, e.g. [Fuchs1992]. We do not wish to enter this discussion here but we show how Alma-0 supports executable specifications in a very natural way.

As an example, consider the problem of finding the lexicographically next permutation, discussed in [Dijkstra1976].

To specify this problem recall that by definition a sequence is a permutation of if for some function from onto itself we have

This definition directly translates into the following Alma-0 program:

TYPE Sequence = ARRAY [1..N] OF INTEGER;

PROCEDURE Permutation(VAR in, out: Sequence);
VAR pi: Sequence;
    i, j: INTEGER;
BEGIN
  FOR i := 1 TO N DO
    SOME j := 1 TO N DO
      pi[j] = i
    END
  END; (* pi is a function from 1..N onto itself and ...        *)
  FOR i := 1 TO N DO
    out[i] = in[pi[i]]
  END  (* out is obtained by applying pi to the indices of in *)
END Permutation;

The procedure Permutation provides, upon backtracking, all permutations of the given input sequence.

Next, we need to define the lexicographic ordering. Let us recall the definition: the sequence precedes lexicographically the sequence if some in the range exists such that for all in the range we have , and .

In Alma-0 we write these specifications as follows:

PROCEDURE Lex(a,b: Sequence);
VAR i, j: INTEGER;
BEGIN
  SOME i := 1 TO N DO
    FOR j := 1 TO i-1 DO
      a[j] = b[j]
    END;
    a[i] < b[i]
  END
END Lex;

Now is the lexicographically next permutation of if

  • is a permutation of ,

  • precedes lexicographically,

  • no permutation exists that is lexicographically between and .

This leads us to the following procedure Next that uses an auxiliary procedure Between, which checks whether a permutation exists between a and b:

PROCEDURE Between(a,b: Sequence);
VAR c: Sequence;
BEGIN
  Permutation(a,c);
  Lex(a,c);
  Lex(c,b)
END Between;

PROCEDURE Next(VAR a, b: Sequence);
BEGIN
  Permutation(a,b);
  Lex(a,b);
  NOT Between(a,b)
END Next;

This concludes the presentation of the program. Note that it is fully declarative and it does not use any assignment. It is obviously hopelessly inefficient, but still it could be used on the example given in Dijkstra’s book, to compute that 1 4 6 2 9 7 3 5 8 is the lexicographically next permutation of 1 4 6 2 9 5 8 7 3.

It is interesting to see that the above program is invertible in the sense that it can be also used to specify and compute the lexicographically previous permutation. In fact, we can use for this purpose the same procedure Next — it just suffices to pass now the given permutation as the second parameter of the procedure Next. For this purpose both parameters are passed by variable in the procedures Next and Permutation.

In this way we can compute for instance that 1 4 6 2 9 5 8 3 7 is the lexicographically previous permutation of 1 4 6 2 9 5 8 7 3.

6 A Scheduling Application

We now show how Alma-0 can be employed to solve scheduling problems. In particular, we introduce a specific scheduling problem known as the university course timetabling problem and discuss its solution in Alma-0.

6.1 Problem Definition

The course timetabling problem consists in the weekly scheduling for all the lectures of a set of university courses in a given set of classrooms, avoiding the overlaps of lectures having common students. We consider the basic problem (which is still NP-complete). Many variants of this problem have been proposed in the literature. They involve more complex constraints and usually consider an objective function to be minimized (see [Schaerf1999]).

Problem 4

There are courses , and each course consists of required lectures, and periods . For all , all lectures must be assigned to a period in such a way that the following constraints are satisfied:

Conflicts:

There are curricula , which are groups of courses that have common students. Lectures of courses in must be all scheduled at different times, for each .

Availabilities:

There is an availability binary matrix of size . If then lectures of course cannot be scheduled at period .

Rooms:

There are rooms available. At most lectures can be scheduled at period , for each .

6.2 A solution in Alma-0

We now provide a solution of this problem in Alma-0. We start with the constant and type definitions necessary for the program.

CONST
   Courses = 10;  (* p *)
   Periods = 20;  (* q *)
   Rooms = 3;     (* r *)
TYPE
   AvailabilityMatrix = ARRAY [1..Courses],[1..Periods] OF BOOLEAN;
   ConflictMatrix = ARRAY [1..Courses],[1..Courses] OF BOOLEAN;
   RequirementVector = ARRAY [1..Courses] OF INTEGER;
   TimetableMatrix = ARRAY [1..Courses],[1..Periods] OF BOOLEAN;

Conflicts are represented by a matrix of the type ConflictMatrix such that the element of the matrix is true if courses and belong simultaneously to at least one curriculum.

The solution is returned by means of a boolean matrix of the type TimetableMatrix. Each element of the matrix is true if a lecture for the course is given at period and false otherwise.

The procedure Timetabling provides the solution of this problem in Alma-0. It follows faithfully the specification of the problem and it performs an exhaustive backtracking search for a feasible solution.

For each course the procedure looks for a number of periods equal to the number of lectures of the course. The array BusyRooms counts the number of rooms already used for each period, and is used to check the room occupation constraints.

In order to avoid exploring symmetric solutions for the lectures of a course, each lecture is always scheduled later than the previously scheduled lectures of the same course. This is done by using the variable PeriodOfPreviousLecture which keeps track of the period of the most recently scheduled lecture.

PROCEDURE Timetabling(Available: AvailabilityMatrix;
                      Conflict: ConflictMatrix;
                      Requirements: RequirementVector;
                      VAR Timetable: TimetableMatrix);
VAR
   BusyRooms : ARRAY [1..Periods] OF INTEGER;
   C, C1, L, P : INTEGER;
   PeriodOfPreviousLecture : INTEGER;
BEGIN
   FOR P := 1 TO Periods DO
      BusyRooms[P] := 0;
   END;
   FOR C := 1 TO Courses DO
      PeriodOfPreviousLecture := 0;
      FOR L := 1 TO Requirements[C] DO
         SOME P := PeriodOfPreviousLecture+1 TO Periods DO
            Available[C,P];
            BusyRooms[P] < Rooms;
            FOR C1 := 1 TO C-1 DO
               NOT (Conflict[C1,C] AND Timetable[C1,P])
            END;
            Timetable[C,P] := TRUE;
            BusyRooms[P] := BusyRooms[P] + 1;
            PeriodOfPreviousLecture := P;
         END
      END
   END
END Timetabling;

The proposed procedure can solve only relatively small instances of the problem. For larger ones more complex algorithms and heuristic procedures are needed (see

[Schaerf1999]).

6.3 Additional Functionalities

If no solution to the given problem instance exists, it is in general necessary to relax some of the constraints. The following procedure checks whether a solution exists when one single conflict constraint is relaxed. If the solution of the relaxed instance of the problem is found, its solution is returned along with the constraint which has been relaxed. This constraint is returned by means of two courses c1 and c2 which are no more considered in conflict.

PROCEDURE RelaxedTimetabling(Available: AvailabilityMatrix;
                             VAR Conflict: ConflictMatrix;
                             Requirements: RequirementVector;
                             VAR Timetable: TimetableMatrix;
                             MIX c1, c2: INTEGER);
VAR
   i, j: INTEGER;
BEGIN
   EITHER
     Timetabling(Available, Conflict, Requirements, Timetable)
   ORELSE
     SOME i := 1 TO Courses-1 DO
       SOME j := i+1 TO Courses DO
         Conflict[i,j];
         c1 = i; c2 = j;
         Conflict[i,j] := FALSE;
         Timetabling(Available, Conflict, Requirements, Timetable)
       END
     END
   END
END RelaxedTimetabling;

Finally, the following procedure produces all relaxed and non-relaxed solutions of the problem. The simple code for the procedures Initialize and PrintSolution is omitted.

PROCEDURE CreateTimetable;
VAR
  Available: AvailabilityMatrix;
  Conflict: ConflictMatrix;
  Requirements: RequirementVector;
  Timetable: TimetableMatrix;
  NbrSolutions: INTEGER;
  c1, c2: INTEGER;
BEGIN
  Initialize(Available,Conflict,Requirements,Timetable);
  NbrSolutions := 0;
  FORALL
    RelaxedTimetabling(Available,Conflict,Requirements,Timetable,c1,c2)
  DO
    NbrSolutions := NbrSolutions + 1;
    WRITELN(’Solution number ’,NbrSolutions);
    PrintSolution(Available,Timetable);
    IF KNOWN(c1)
    THEN WRITELN(’Conflict between course ’, c1,’ and ’,c2,’ relaxed’)
    ELSE WRITELN(’No constraint relaxed for this solution’);
    END
  END;
  IF NbrSolutions > 0
  THEN WRITELN(’Number of solutions : ’,NbrSolutions)
  ELSE WRITELN(’No solution found.’);
  END;
  WRITELN
END CreateTimetable;

Note the use of the built-in procedure KNOWN that checks whether the variable c1 is initialized or not. This test allows us to check whether a constraint has been relaxed.

Finally, note that c1 and c2 are passed by MIX. This way, not only a variable but also a constant can be supplied as an actual parameter. For example, the following call searches for a solution in which the possible relaxation involves course :

   RelaxedTimetabling(Available,Conflict,Requirements,Timetable,1,c);

Here c is an uninitialized variable.

7 Conclusions

In this paper we presented a number of programs written in Alma-0. They were chosen with the purpose of illustrating the versatility of the resulting programming style. The solution to some other classical problems, such as - search, STRIPS planning, knapsack, and Eight Queens, have been already provided in [Apt et al.1998].

These programs show that imperative and logic programming can be combined in a natural and effective way. The resulting programs are in most cases shorter and more readable than their counterparts written in imperative or logic programming style.

Let us review now the work carried out on Alma-0. The implementation of the language Alma-0 is based on an abstract machine, called AAA, that combines the features of a RISC architecture and the WAM abstract machine. In the current version the AAA instructions are translated into C code. The implementation is described in [Apt et al.1998] and explained in full detail in [Partington1997]. The Alma-0 compiler is available via the Web at http://www.cwi.nl/alma.

An executable operational specification of a large fragment of Alma-0 is provided using the ASF+SDF Meta-Environment of [Klint1993]. This is described in [Apt et al.1998] and comprehensively explained in [Brunekreef1998].

An extension of Alma-0 that integrates constraints into the language is the subject of an ongoing research. Various issues related to such integration are highlighted in [Apt and Schaerf1999]. In particular, the role of logical and customary variables, the interaction between the program and the constraint store, the local and global unknowns, and the parameter passing mechanisms are considered there.

Finally, in [Apt and Bezem1999] a computational interpretation of first-order logic based on a constructive interpretation of satisfiability w.r.t. a fixed but arbitrary interpretation is studied. This work provides logical underpinnings for a fragment of Alma-0 that does not include assignment and allows us to reason about Alma-0 programs written in this fragment.

References

  • [Apt and Bezem1999] K. R. Apt and M. A. Bezem. Formulas as programs. In K.R. Apt, V.W. Marek, M. Truszczyński, and D.S. Warren, editors, The Logic Programming Paradigm: A 25 Year Perspective, pages 75–107, 1999. Available via http://xxx.lanl.gov/archive/cs/.
  • [Apt and Schaerf1999] K. R. Apt and A. Schaerf. The Alma project, or how first-order logic can help us in imperative programming. In E.-R. Olderog and B. Steffen, editors, Correct System Design, Lecture Notes in Computer Science 1710, pages 89–113, 1999.
  • [Apt et al.1998] K. R. Apt, J. Brunekreef, V. Partington, and A. Schaerf. Alma-0: An imperative language that supports declarative programming. ACM Toplas, 20(5):1014–1066, 1998.
  • [Apt1997] K. R. Apt. From Logic Programming to Prolog. Prentice-Hall, London, U.K., 1997.
  • [Brunekreef1998] J. Brunekreef. Annotated algebraic specification of the syntax and semantics of the programming language Alma-0. Technical Report P9803, Programming Research Group, University of Amsterdam, The Netherlands, 1998. Available online at
    http://www.wins.uva.nl/research/prog/reports/reports.html.
  • [Cohen1979] J. Cohen. Non-Deterministic algorithms. ACM Computing Surveys, 11(2):79–94, 1979.
  • [Dijkstra1975] E. W. Dijkstra. Guarded commands, nondeterminacy and formal derivation of programs. Communications of the ACM, 18:453–457, 1975.
  • [Dijkstra1976] E. W. Dijkstra. A Discipline of Programming. Prentice-Hall, Englewood Cliffs, N.J., 1976.
  • [Fuchs1992] N. Fuchs. Specifications are (preferably) executable. IEE Software Engineering Journal, 7(5):323–334, 1992.
  • [Gardner1979] M. Gardner. Mathematical Circus. Penguin, Harmondsworth, 1979.
  • [Garey and Johnson1979] M. R. Garey and D. S. Johnson. Computers and Intractability—A guide to NP-completeness. W.H. Freeman and Company, San Francisco, 1979.
  • [Griswold and Griswold1983] R. E. Griswold and M. T. Griswold. The Icon Programming Language. Prentice-Hall, Englewood Cliffs, New Jersey, USA, 1983.
  • [Klint1993] P. Klint. A meta–environment for generating programming environments. ACM Transactions on Software Engineering and Methodology, 2(2):176–201, 1993.
  • [Partington1997] V. Partington. Implementation of an imperative programming language with backtracking. Technical Report P9712, Department of Mathematics, Computer Science, Physics & Astronomy, University of Amsterdam, The Netherlands, 1997. Available online at
    http://www.wins.uva.nl/research/prog/reports/reports.html.
  • [Schaerf1999] A. Schaerf. A survey of automated timetabling. Artificial Intelligence Review, 13(2):87–127, 1999.
  • [Schwartz et al.1986] J. T. Schwartz, R. B. K. Dewar, E. Dubinsky, and E. Schonberg. Programming with Sets — An Introduction to SETL. Springer-Verlag, New York, 1986.