Incorrect result with eqrel?

[bertram-gil]

Hi guys,
Consider the following program:

.decl two_set(x : number, y : number) eqrel
.decl union_two_set(x : number, y : number)
.decl unique2(x : number)
.decl equivalent(a: number, b: number) eqrel
.decl out(a: number)
.decl A(a:number, b:number) 
.output A 

two_set(1, 2).
two_set(3, 1).
two_set(6, 10).
unique2(1).
equivalent(1, 2).
equivalent(1, 3).

union_two_set(x, y) :- two_set(x, x), two_set(y, y).
out(x) :- equivalent(x, 1).

A(j, j) :- union_two_set(j, g), unique2(a),  equivalent(a, h).

Running the above program with Souffle gives me the following values for A:

1	1
2	2
3	3
6	6
10	10

If I add a new rule A(a, a) :- out(a), !out(a). to get the following program:

.decl two_set(x : number, y : number) eqrel
.decl union_two_set(x : number, y : number)
.decl unique2(x : number)
.decl equivalent(a: number, b: number) eqrel
.decl out(a: number)
.decl A(a:number, b:number) 
.output A 

two_set(1, 2).
two_set(3, 1).
two_set(6, 10).
unique2(1).
equivalent(1, 2).
equivalent(1, 3).

union_two_set(x, y) :- two_set(x, x), two_set(y, y).
out(x) :- equivalent(x, 1).

A(j, j) :- union_two_set(j, g), unique2(a),  equivalent(a, h).
A(a, a) :- out(a), !out(a).

Running the above program returns an empty result for A. But the result should not change.

Souffle version I am using: 235a3bc

[b-scholz]

With the latest performance changes of eqrel, we destroyed the iterators. A smaller example is the following:

.decl E(a:number, b:number) eqrel
E(1,2).
E(1,3).
 
.decl A()
A() :-
     E(1,_).
 
.decl B()
B() :-
  E(_,1).
.output  A,B

Both nullaries should contain the empty tuple. However, only the nullary B contains the empty tuple but not the nullary A.

[b-scholz]

Can you check whether the latest release has the same issue?

[b-scholz]

The problem is the execution of two subsequent existence checks on an equivalence relation, i.e.,

QUERY
    IF ((ISEMPTY(+disconnected1) AND (NOT ISEMPTY(E))) AND (NUMBER(1),UNDEF) IN E)
     INSERT () INTO +disconnected1
   END QUERY

[b-scholz]

This is only an issue in the interpreter - not in the compiler - right?

[b-scholz]

After further analysis, it seems to be an interpreter issue that undefined values for equivalence relations cause problems and this needs to be fixed in the Generator.