Henshin/Critical Pair Analysis
Henshin's critical pair analysis (CPA) feature enables the detection of all potential conflicts and dependencies of a set of rules. The result of the analysis is a set of critical pairs, each reporting on a potential conflict or dependency between two rules. A critical pair describes a minimal conflict or dependency situation, consisting of the first rule and the second rule as well as a minimal model affected by the conflict or dependency.
As Henshin transformations are formally based on graph transformations, the critical pair analysis is also based on algebraic graph transformations.
Kinds of conflicts and dependencies
Two kinds of relations among transformation rules can be detected: conflicts and dependencies. In the case of a conflict, the application of the first rule renders the second rule unapplicable on the produced model. In the case of a dependency, the application of the first rule enables the second rule to be applied on the produced model.
A conflict generally describes a situation in which the two involved rules are both applicable on a minimal model. Applying the first of these rules on the minimal model leads to a changed result model. While the second rule could have been applied on the model in the beginning, it is no more applicable on the result model. In what follows, possible kinds of conflicts are described.
- Delete-use-conflict: The first rule deletes at least one element (an EClass or EReference) of the minimal model, such that the second rule is no more applicable.
- Produce-forbid-conflict: The first rule creates at least one element (EClass or EReference) which is forbidden by the second rule. The second rule becomes inapplicable on the resulting model, accordingly.
- Change-use-conflict: A change-use-conflict describes a similar situation to the delete-use-case, but with the focus on the values of EAttributes: the application of the first rule changes the value of an EAttribute. The second transformation is then no more applicable since it is only applicable when the old value is present.
- Change-forbid-conflict: Change-forbid-conflicts describe the situation where the application of the first rule changes the value of an EAttribute in such a way that it becomes equal to a forbidden value specified in the second rule.
A dependency generally describes a situation in which the application of the second transformation rule becomes enabled by applying the first transformation rule. The minimal model is the result of the application of the first rule and it allows the second rule to be applied. Various situations can be encountered, which are described by the following different kinds of dependencies.
- Produce-use-dependency: A produce-use-dependency describes the situation where the application of the first rule creates a new element (EClass or EReference) which is required by the second rule. Based on the creation of the element the second rule becomes applicable. Note that it is irrelevant how the second rule deals with the created element, that is, if the element is preserved or deleted. The minimal model contains the created element which is required to apply the second rule.
- Delete-forbid-dependency: A delete-forbid-dependency describes the situation where the first rule deletes an element (EClass or EReference) which is forbidden by the second rule. The second rule becomes applicable since its negative application condition (NAC), which forbids the presence of the element, is now fulfilled. The minimal model is the instance after the application of the first rule, such that the second rule becomes applicable on it.
- Change-use-dependency: The change-use-dependency is similar to the produce-use-dependency with the focus on the value of an EAttribute. The dependency describes the situation where the second rule becomes applicable after the application of the first rule. The root cause is the change of the value of an EAttribute such that it fulfills the pattern of the second rule.
- Change-forbid-dependency: The change-forbid-dependency describes the situation where the first rule changes an EAttribute value that is forbidden by the second rule. With the new value, the second rule becomes applicable. The minimal model is the result after applying the first rule.
- Basic model transformations involving the actions preserve, create and delete of EClasses and EReferences as well as the change of the values of EAttributes
- Positive application conditions (PAC) and negative application conditions (NAC), limited to the usual situation of AND concatination. In consequence this means application conditions containing OR and XOR ModelElements are not supported.
- Ecore modeling concepts of inheritance and abstract eClasses and eOpposite references.
- Model transformations for more than one meta-model, for example outplace model transformations involving more than one Ecore instance. Affected rules are identified by more than one import in their containing module.
- Multi-rules (formally, rules with amalgamtion).
- Transformation units.
Applying the Analysis
The analysis can be invoked either using a wizard or programmatically using an API.
The wizard enables the quick and easy usage of the critical pair analysis, notably if the set of rules is currently being developed.
To invoke the wizard, please switch to the Java perspective in Eclipse. Right-click on at least one selected *.henshin file in the Package Explorer and choose Henshin→Calculate Critical Pairs. The selection of multiple files is supported, but the user needs to ensure that all included rules are specified for the same meta-model.
To execute the analysis, the rules have to be selected and the user needs to decide if conflicts, dependencies, or both kinds of relationships are to be considered. The required time for the analysis depends on the number of rules and the size of each rule. A large number and/or size of selected rules can lead to considerable execution times!
After selecting at least one rule and at least one kind of relationship, the analysis can be started using the Finish button. Alternatively, the Next button leads to the second page Option Setting for advanced configuration.
After the analysis finished, all critical pairs as produced by the user-specified configuration are listed in the CPA Result View. The critical pairs are grouped by the rule combinations of which they consist. This means that all analysed conflicts and dependencies of a combination of two rules are listed together. To view them individually, the entry for the rule combination can be expanded. Each single critical pair can be further inspected. Double-clicking or pressing the enter button opens the critical pair in the Critical Pair Editor. Each critical pair consists of the first rule on the left side, the second rule on the right side, and the minimal model in between. Additionally, the results of the analysis are saved in the workspace within a directory tree. To open a critical pair from the workspace, right-click on a *.henshinCp file in the Package Explorer and select Henshin→Open Critical Pair.
The CPA API allows to use the CPA programmatically. As a prerequisite, all considered transformation rules have to be loaded upfront. A description of this process is given in the Henshin Interpreter page.
Depending on the desired result, either one or two lists of the rules to be analysed have to be set up. To obtain all pair-wise conflicts or dependencies within a set of rules, a single list is sufficient. If only the dependencies or conflicts for a special set of rule pairs is needed, a list of the first rules (r1) and a list of the second rules (r2) is required as input. Even though the same output could be produced by analysing all rules and filtering the result afterwards, consider that the superfluous analysis might be very time consuming.
Another required initialisation parameter is a CPAOptions object. This object is easily obtained using its default constructor and can be adapted using its Setter and Getter methods.
The user mainly interacts with the CPA feature using the CpaByAGG class, which is instantiated using its default constructor. Providing the aforementioned list(s) of rules and CPAOptions object as input, the analysis can be initialized with the suitable init(…)-method. Please note that the init(…)-method might throw an UnsupportedRuleException if one of the rules exhibits an unsupported feature. The method check(List<Rule> rules) can be used to avoid this exception by checking the rules upfront.
After successful initialisation, the methods runConflictAnalysis() or runDependencyAnalysis() can be used to start the analysis. Both methods optionally take a ProgressMonitor object as input that can monitor the analysis progress and give a rough estimate for the required time to finish the analysis.
For a usage example of this feature, see the Simple Class Modeling Refactoring.