I prefer not to think in terms of undo, that lays itself to a particular use case. Instead I prefer to model it on logical insertions, like logical insertions when a rule is no longer true we retract the inserted fact. A closure is an anonymous code block that will be executed using the given state, think of it as a stateful callback. So in the same way that a logical inssertion will execute an action that will retract the fact it inserted a logical closure will execute the anonymous block using hte previously supplied state - obviosly the onus is on the user to provide state that would only be valid for execution at that time. The main reason for this is rules are very good at telling you when something happens, but not telling you when something stops happeneing  - 'not' CEs are not quite flexible enough for this.

Mark


James C. Owen wrote:
Greetings:

It's been interesting watching this thread develop.  Originally, I wasn't going to comment but then I threw caution to the four corners of the globe and said, "Why not?"  

Originally, rulebased systems were simple IF-THEN clauses in a non-monotonic environment such the engine rules were evaluated as true or false, those that were true were placed on the agenda to be fired, which rule to fire next was then selected via some conflict resolution strategy (such as MEA or LEX), the engine fired that rule, and the process started all over until  there were no more rules to fire.

Along the way, a rule might not be true until some other rule fired that made it true.  Conversely, if some rule fired that made another rule false that the second rule was removed from the agenda table.  A rule is composed of its data and its logic.  Originally, an OR statement meant that another rule had to be written.  Likewise, an ELSE statement meant that DeMorgan's theorem was applied to the LHS of the rule and a new rules was written.

Then came OPSJ that featured OR, ELSE and Conditional ELSE where the ELSE (1) clause would only be evaluated if the first CE was false.  Very selective and very handy. Normally an ELSE fires if ANY of the CEs are false on the LHS.  It seems to me that the idea of "undo" and "rollback" comes from the database world.  If a rule no longer matches that does not mean that one of the other rules might not change the data of another CE in another rule so that it does, in fact, match later.  Refraction - a feature of almost all rulebased systems - will remove a rule (logic+data) once it fires so I would not think that this is not a matter under consideration.

Now, to the part that mystifies me:  If there was a previous "fireAllRules" and that rule did not match does not mean that the rule would not match this time we run a "fireAllrules" unless the data and the rule logic remained unchanged.  if a rule never fires and never will fire, why would it be in the system?  Conversely, if the data changes why would one want to "undo" the run from a previous match in another run?  You wouldn't.  Obviously.  But that seems to be what has been promulgated in the foregoing discussion.  "No longer matching" does not mean that the rule will never match in the preset "fireAllrules" run - you don't know that until all of the rules on the Agenda (and the changes to the Agenda) have been processed completely.  And by that time it's way too late to consider an "undo" for those rules.

SDG
jco
"This above all: to thine own self be true,
And it must follow, as the night the day,
Thou canst not then be false to any man."
Hamlet, Act 1, Scene III

On Feb 19, 2009, at 8:45 AM, Mark Proctor wrote:

Geoffrey De Smet wrote:
What would an "else clause" do?
Imagine a rule with matches on 5 different fact sets.
How many times would the else part match?
- none (because it matches at least one)
- a very lot (because it matches on any fact set that isn't that one of those 5)

The "undo-then" is another concept:
it matches when a rule that matched before (in a previous fireAllRules), now no longer matches.
"undo-then" probably isn't the best name, so better suggestions are welcome, but "else" isn't a good name for it as it's not about "not matching" but about "no longer matching".
As I mentioned, i'd do it as a logical closure using an anonymous code block, in a similar manner that we do for logical insertions.

Mark


Anyway, I 've been thinking and it wouldn't work for all use cases in drools-solver:
rule
 when
   q1 : Queen()
   q2 : Queen()
   eval(q1.getY() - q2.getY() < 10)
 then
   a.add(q1.getY() - q2.getY());
 undo-then
   a.subtract(q1.getY() - q2.getY())
end
The y of a queen changes, so the subtract wouldn't subtract the exact same number that was added.
Any way we could work around that, or is there no avoiding insertLogical?

With kind regards,
Geoffrey De Smet

Mark Proctor schreef:
Greg Barton wrote:
--- On Wed, 2/18/09, Geoffrey De Smet <ge0ffrey.spam@gmail.com> wrote:


The current workarounds [to undo-then] are clunky:
- Writing an negative (opposite) rule isn't efficient:
it means declaring the rule twice effectively. Also the
negative rule is usually using lots of or's and
not's which isn't fast.
   

This would be made easier by...drum roll please...the else clause! :)
 
yes we want OPSJ style else statements, edson has an idea on how to do that, just a matter of time :(
Now, if there was the else clause plus undo-then/closures you could probably write an entire complex ruleset in one rule.  Might as well just use perl, then. :P   
heh, that's true you would have potentially encapsulated 4 possible executions in a single rule.

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


 


------------------------------------------------------------------------

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

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




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


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