Possible bug with nth1 predicate in Eclipse-clp?

I'm writing a sudoku solver in Prolog. The sudoku itself is given as a vector of vectors:

P =
    [[1,_,_, _,_,_, _,_,_],
     [_,_,2, 7,4,_, _,_,_],
     [_,_,_, 5,_,_, _,_,4],

     [_,3,_, _,_,_, _,_,_],
     [7,5,_, _,_,_, _,_,_],
     [_,_,_, _,_,9, 6,_,_],

     [_,4,_, _,_,6, _,_,_],
     [_,_,_, _,_,_, _,7,1],
     [_,_,_, _,_,1, _,3,_]].

I'm converting this to a matrix with the following code:

convert_to_matrix(P, Puzzle) :-
dim(Puzzle, [9, 9]),    % create square matrix
( multifor([I, J], 1, 9), % fill array 
  param(P, Puzzle) 
  do
    nth1(I, P, RowI),
    nth1(J, RowI, Elem),
    subscript(Puzzle, [I , J], Elem)
).

This works fine for the first 8 rows and colums, but fails for everytime either I or J is 9. doing nth1(9, P, RowI) makes RowI a brand new variable instead of the 9th vector/row. Doing nth1(9,RowI, Elem) doesn't return the 9th element (for the eight row it should return 1, but returns a new Variable). Is this a bug in Eclipse-clp or am I missing something?

Solution

You are absolutely right, this is indeed a bug in nth1/3, introduced and only present in version 7.0#45! You can replace your code with something like

lol_matrix(Xss, M) :-
    length(Xss, N),
    dim(M, [N,N]),
    ( foreach(Xs,Xss), foreacharg(Row,M) do
        array_list(Row, Xs)
    ).

However, the easiest way is to write your data directly as a matrix (as done in this example code), then you don't need any conversion at all:

P = []([](1,_,_, _,_,_, _,_,_),
       [](_,_,2, 7,4,_, _,_,_),
       ...
       [](_,_,_, _,_,1, _,3,_)).

By the way, ECLiPSe bugs can be reported via [email protected]