- Lots of debugging
- Changed the way the filter keeps information about the scheduler to use for probability/reward queries.
| This was done by keeping a special action at the first position of the action list.
| Which was not exactly consistent with the idea behind the filter actions.
| Now the filter keeps this information as an enum value in a member variable.
- All but one tests are green. So we almost reestablished full functionality.
|- The last test that still fails is SparseMdpPrctlModelCheckerTest.Dice where the second to last model check returns the wrong result.
Next up: Debug. Then introduce the full range of filter actions.
Former-commit-id: fd311966cc
- Fixed similar undefined behavior for the MarkovAutomaton Csl modelchecker.
Next up: Make necessary changes to the formula parsers.
Former-commit-id: e8765fe58b
- General function of the filters: The filter as an abstraction layer between the control flow and the formula/modelchecker.
|- It has a formula as child but is not a formula itself.
|- It invokes the modelchecking process on the child formula and manipulates the result.
|- For the purpose of result manipulation it keeps a list of filter actions.
|- Each action manipulates the result in a certain way. For example: It returns only the results for states 25 to 140.
|- Furthermore the printing of the result to standard out and the log is no longer done by the modelchecker but by the filter.
|- That way the tasks of each class becomes more clear: Modelchecker to compute the results, filter to prepare the computed results for write out.
- Battled with a major design problem: How to integrate the optimizing operator (aka. min or max probs for non-det. models) into the filter scheme.
|- It is now integrated as a separate filter action, which does not touch the results but hold the flag determining whether to maximize or to minimize.
|- This action must be the innermost filter action (i.e. the first list entry) to have any effect.
|- This is combined with a special fuction of the modelchecker that manipulates the mutable minimizationStack calculates the modelchecking result and resets the stack to its original state.
|- This way the information whether to min, to max or not to try is managed by the filter and propagated as needed.
|- Remark: Fixed a major risk of undefined behavior in the SparseMdpPrctlModelChecker.
|- If the formula to be checked did not have a NoBoundFormula as root then the minimumOperatorStack would be empty and minimumOperatorStack.top() would result in undefined behavior.
|- Added tests whether the stack is empty before trying to read out the possibly non existant top element.
Next up: Implement similar filters for LTL and CSL and try to get it compiled.
Former-commit-id: 577998e027
TimQu
dehnert
Sebastian Junges
David_Korzeniewski
masawei
PBerger