I think there is some fundamental error in deriving truths from givens and other derived facts that are then interpreted as given truths, and, moreover, with subtly varying semantics. In terms of graph theory: reachability is based on (directed) edges, but it does not establish additional edges.

Deriving Reachability should be done by:

rule deriveLink
when
    Link( $a: a, $b: b )
then
    insertLogical(new Reachable($a,$b));
    System.out.println( "ins reach " + $a + " " + $b );
end

rule deriveReachReach
no-loop true
when
    Reachable( $a: a, $b: b )
    Reachable( a == $b, $c: b != $a )
then
    insertLogical(new Reachable($a,$c));
end

Ideally, I would like to use
   not Reachable(a == $a, b == $c)
instead of the (last resort) no-loop in the second rule, but Drools' truth maintenance is incomplete: it does not let your define the logical dependency on part of the condition (i.e., excluding the CE "not" in this case).

-W


  



On 8 March 2011 05:49, Simon Chen <simonchennj@gmail.com> wrote:
What I had is a very simplified version of how calculating transitive
closure could go wrong...

Let's say we have two rules:
rule 1
when
 link(a,b)
then
 insertLogical(new reachable(a,b))

rule 2
when
 link(a,b) reachable(b,c)
then
 insertLogical(new reachable(a,c))

Let's say, I have link(a,b), link(b,c), link(b,a), link(c,b). So,
we'll have reachable(a,b), reachable(b,c), reachable(a,c), etc. But,
after I retract link(a,b) and link(b,a), guess what, reachable(c,a)
still exists! This doesn't sound right to me.

But in Drools, this is possible, because we have:
reachable(c,a) <- link(c,b), reachable(b,a)
reachable(b,a) <- link(b,c), reachable(c,a)

The problem here is that we actually inserted reachable(b,a) multiple
times: first supported by link(b,a) and rule 1, and secondly by
link(b,c) and reachable(c,a) and rule 2. When reachable(b,a) was
inserted the second time, link(b,c) and reachable(c,a) become the
additional supporting condition - maintained by the truth maintenance
system. So, even if link(b,a) is retracted, reachable(b,a) still
exists further supporting reachable(c,a).

Is it clearer?

Thanks.
-Simon

2011/3/7 Edson Tirelli <ed.tirelli@gmail.com>:
>
>    Simon,
>    The behavior seems correct to me as B is justified by either A or C (or
> both). Of course, from the initial state, A is required for C to first
> exist, but once it starts to exist, your rules say that B and C justify each
> other and so both remain in memory.
>    This is design as intended, but do you think that is wrong?
>    Edson
>
> 2011/3/7 Simon Chen <simonchennj@gmail.com>
>>
>> Hi all,
>>
>> An interesting finding:
>>
>> I have three simple rules:
>> rule "A2B"
>>        when
>>                A()
>>        then
>>                insertLogical(new B());
>> end
>> rule "B2C"
>>        when
>>                B()
>>        then
>>                insertLogical(new C());
>> end
>> rule "C2B"
>>        when
>>                C()
>>        then
>>                insertLogical(new B());
>> end
>>
>> Basically, once we have an A(), we'll logically insert a B(). Once we
>> have a B(), we'll logically insert a C(). Once we have a C(), we'll
>> logically insert a B().
>>
>> So, I first insert an A(), print all the objects. Retract A(), and
>> print all the objects. Here's what I got:
>> com.sample.B@42
>> com.sample.C@43
>> com.sample.A@548997d1
>> after retract!
>> com.sample.B@42
>> com.sample.C@43
>>
>> So, B() and C(), which should be logically depend on A(), somehow are
>> not retracted. The problem I see is the truth maintenance system allow
>> B() and C() to depend on each other, thus not affected by losing A().
>>
>> Is this a bug or my bad usage?
>>
>> Thanks.
>> -Simon
>> _______________________________________________
>> rules-users mailing list
>> rules-users@lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/rules-users
>
>
>
> --
>   Edson Tirelli
>   JBoss Drools Core Development
>   JBoss by Red Hat @ www.jboss.com
>
> _______________________________________________
> rules-users mailing list
> rules-users@lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>
>

_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users