[jboss-jira] [JBoss JIRA] Commented: (JBRULES-449) Logical dependency bug, justifier tuple contains retracted handles, logically depend object stays in working memory

Logical dependency bug, justifier tuple contains retracted handles, logically depend object stays in working memory ------------------------------------------------------------------------------------------------------------------- Key: JBRULES-449 URL: http://jira.jboss.com/jira/browse/JBRULES-449 Project: JBoss Rules Issue Type: Feature Request Security Level: Public(Everyone can see) Components: Reteoo Affects Versions: 3.0.4 Environment: jdk 1.4.2_10 Reporter: Juergen none Assigned To: Mark Proctor Attachments: from-unit-test-retracted-handles-as-justifiers-bug.jpg, retracted-handles-as-justifiers-bug.jpg, test_LogicalAssertionsAccumulatorPattern.drl Forget comlicated description, I leave it for history purposes, look at unit test attached. ============ A rather complicated, intertwined set of rules leads to a logically asserted fact, that remains in working memory, even after its justifiers have been retracted. Sadly I have no unit-test, but I can outline the rules and execution sequence as reported by a listener (a ... assert, al ... assert logical, r ... retract) Its a buggy (BBB asserted twice) implementation of a accumulating set of rules I posted in the user list some time ago. FFF(A), FFF(B) are of same class but not equal, @... is systemHashCode, so different instances, but possible equal (as mentioned) classes without @ have only one instance in working memory a XXX R0: XXX -> al DDD a AAA (external assert) Sequence of rules fired R1: AAA -> al GGG R2: DDD, GGG -> al CCC R3: GGG -> al EEE R4: EEE, DDD, CCC -> a BBB@15737ca, al FFF@b2d75c(A) (BBB@15737ca contributes to hash/equals of FFF@b2d75c) R5: BBB, CCC, FFF(A), not FFF(B) -> mod CCC (hashcodes changes, but all rules are still fulfilled as before), a FFF@b274de(B) R5 again (modify triggered): EEE, DDD, CCC -> a BBB@16d8ba (but equals(BBB@15737ca) is true), (+ my guess, object created and assert called but no listener info: al FFF@2aefe3(A) (equals FFF@b2d75c, backing its logical dependency) (BBB@16d8ba contributes to Hash/equals)) --- r AAA (external retract) -> (according to listener): r AAA r GGG r EEE r CCC breakpoint, screenshot --> justifier list is invalid, contains retracted handles PoolCondition would be DDD, the only valid handle in the (EEE, DDD, CCC) tuple r XXX (retract of XXX has no consequence, just the last handle, still valid in the screenshot, becomes invalid too) -> r XXX r DDD workingMemory.getObjects() lists FFF@b2d75c(A) and FFF@b274de(B) for Rulebase-properties: - RuleBaseConfiguration.PROPERTY_ASSERT_BEHAVIOR -> RuleBaseConfiguration.WM_BEHAVIOR_IDENTITY - empty Major lead could be that after commenting out "mod CCC", FFF(A) is retracted (+ R5 not fired a second time) + problem with changed hashcode of CCC ? + problem with second instances of BBB and FFF(A) being asserted + FFF(B) being assert normally, not logically