The solution I adopted is to make SwitchState not a subclass of TrackState but of ElementState, which means that the field(s) to be inherited from TrackState have to be duplicated in SwitchState. (Luckily, here it is only one field.) Adding a type-identifying field goes against the (my) idea of classes for identifying categories and using them as an implicit constraint in rules, making them more readable. That's one of the benefits of having an OO RBS, isn't it? Anyway, I just wanted to draw people's attention to this hidden trap; working around it is no problem once you know it is there. -W On 12 December 2010 05:06, Greg Barton <greg_barton(a)yahoo.com&gt; wrote:

The benefits of an OO RBMS don't matter when you're incompletely constraining the type as you were in your original rules.  The constraint you want is "the id is the same AND they're of corresponding types" which you couldn't get by just specifying the supertypes as you were.   As for your testing comment, I agree.  The best I've ever done is to set up a random dataset I know should pass a test, including random setup conditions, and run the test until I'm reasonably confident the code works properly.  That doesn't work if the input data is too "safe," but that's always a problem with unit tests. --- On Sun, 12/12/10, Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; wrote: From: Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; Subject: Re: [rules-dev] A remarkable flop To: "Rules Dev List" <rules-dev(a)lists.jboss.org&gt; Date: Sunday, December 12, 2010, 2:48 AM Ah, yes, and I'd actually be interested to learn why the indeterminism of rule firings not only depends on two or more rules matching the fact set but also on the order and/or number of facts being inserted. Notice that with three SwitchState facts inserted, id's 1, 2 and 5 match and fire the "correct" (i.e., expected) rule. Adding a fourth fact with id 4 suddenly fires the match of no. 5 with the "wrong" (i.e., unexpected) rule. Actually, this means that you cannot rely on tests of rule sets, as you could if a potentially multiple match between two rules would always first activate A or always first B. -W On 12 December 2010 09:28, Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; wrote: The solution I adopted is to make SwitchState not a subclass of TrackState but of ElementState, which means that the field(s) to be inherited from TrackState have to be duplicated in SwitchState. (Luckily, here it is only one field.) Adding a type-identifying field goes against the (my) idea of classes for identifying categories and using them as an implicit constraint in rules, making them more readable. That's one of the benefits of having an OO RBS, isn't it? Anyway, I just wanted to draw people's attention to this hidden trap; working around it is no problem once you know it is there. -W On 12 December 2010 05:06, Greg Barton <greg_barton(a)yahoo.com&gt; wrote: Well, seeing as you're already maintaining two parallel type hierarchies, and it's a mismatch between the two that's causing the problem, you might as well use that in the rules.    rule "Xyz state update"      when        $s: XyzState( $id: id, $type: type ... )        $e: Xyz( id == $id, type == $type )    then      //...update $e, retract $s    end    rule "Xyz state for unknown element"    when       $s: XyzState( $id: id, $type: type )      not Xyz( id == $id, type == $type )  then       //... diagnostic, retract $s   end That oughter do it, and it avoids salience. --- On Sat, 12/11/10, Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; wrote: From: Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; Subject: [rules-dev] A remarkable flop To: "Rules Dev List" <rules-dev(a)lists.jboss.org&gt; Date: Saturday, December 11, 2010, 10:23 AM Given classes Switch and Track, and the "natural" class hierarchy representing state changes for these two elements:    ElementState    TrackState extends ElementState ("free/occupied")    SwitchState extends TrackState (adds "left/right/moving") Now we have rules for updating Switch and Track elements, like this one:    rule "Xyz state update"     when        $s: XyzState( $id: id, ... )        $e: Xyz( id == $id )    then      //...update $e, retract $s    end and, since we want to catch bad ids, also    rule "Xyz state for unknown element"    when       $s: XyzState( $id: id )       not Xyz( id == $id )   then       //... diagnostic, retract $s   end Best practice, wouldn't you say? Testing by inserting a few of SwitchState objects works fine:    updated: Switch 5 RIGHT occupied    updated: Switch 2 RIGHT    updated: Switch 1 RIGHT occupied  so everything is allright, wouldn't you say? Add another SwitchState for Switch "4" to the test, and suddenly:   updated: Switch 4 RIGHT   track state for unknown element 5   updated: Switch 2 RIGHT   updated: Switch 1 RIGHT occupied What's this?! After some headscratching I realized that the negative rule for the SwitchState's superclass TrackState produces another activation, since, for any Switch element with an id x there clearly isn't a Track element with id x! This activation occasionally precedes the activation for the match in the "update" rule. This is annoying. Of course, negative salience for the "not" rules fixes this, but who would have thought that you need it with conditions (seemingly!) describing disjoint situations. And the negative salience is counter-intuitive, since normally you'd perform the check "no such element" before permitting any update action. Remarkable. Wolfgang     -----Inline Attachment Follows----- _______________________________________________ rules-dev mailing list rules-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-dev _______________________________________________ rules-dev mailing list rules-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-dev -----Inline Attachment Follows----- _______________________________________________ rules-dev mailing list rules-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-dev