Recent fix introduced serious performance degradation in Datalog computation

[bertram-gil]

I think the commit 430b4ea that fixed a recent bug in datalog computation (#4870), has introduced a serious performance issue.
Consider the following program:

Z 64

in1(A:Z, B:Z) input
in2(A:Z, B:Z) input
KK(A:Z, B:Z)
OS(A:Z)
out(A:Z) printtuples

in1(23, 7).
in1(46, 41).
in1(7, 42).
in1(36, 15).
in2(45, 39).
in2(36, 14).
in2(17, 38).

KK(y,j) :- in2(YZD,j), in2(y,UyW).
OS(v) :- KK(v,Q), KK(A,Q), KK(B,Q).  
out(y) :- OS(y), OS(y). 

When i run this program with 7089610, I get the result in 5 milliseconds:

Tuples in out: 
        (A=45(7))
        (A=36(5))
        (A=17(10))
Time: 5ms
Parsing: 1ms, other: 3ms

However, if i run this program with the latest commit (c49d39a), I don't get a result even after 3 minutes of waiting.