The Henshin transformation meta-model defines the concepts used for the specification of model transformations in Henshin.
Model transformations are specified in terms of modules that comprise a set of units, being either transformation rules - atomic units - or composite units that allow to define a control flow.
Basic building blocks: Rules
For conciseness, the graphical editor merges LHS and RHS into an integrated representation: Nodes and edges are annotated with a «preserve», «create», or «delete» tag depending on their containment in LHS or RHS graphs.
Dangling condition: A common beginner-level problem is that rules are deemed unexecutable by the Henshin interpreter despite looking correct at the surface. This behavior is often related to the dangling condition: A rule execution may not leave behind dangling edges, being edges with a missing source or target. It is possible to turn off the gluing condition check using the checkDangling Boolean flag. Setting the flag to false will lead to the interpreter deleting the dangling edges.
Injective matching: The injectiveMatching flag specifies whether the match-finder is supposed to find matches that assign two or more LHS nodes to the same model element. Per default, matching is set injective (that is, not supposed to assign nodes to the same model element).
Edges: A few remarks on edges: First, rules do not maintain mappings of edges between different rule graphs: Edges are considered identical if their types and their source and target nodes are identical. Second, if a reference has an EOpposite reference, it is not required to specify an edge for the opposite reference: By its default behavior, EMF ensures that changes involving one of the EOpposites will be propagated onto the other. Third, in ordered references, the positions of list entries can be modified using the index attribute. An example is shown in this article.
Advanced concepts: Application conditions and rule nesting
Application conditions are graph patterns that restrict the LHS of a given rule. They may require the presence of additional elements or relationships not included in the LHS. In this case, they are referred to as positive application conditions (PACs). Furthermore, they may forbid the presence of elements or relationships. In this case, they are referred to as negative application conditions (NACs). Consequently, the graphical editor displays PAC elements with a «require» and NAC elements with a «forbid» tag. Application conditions can be arbitarily nested using the propositional operators Or, And, Xor, and Not. Whether an application condition graph represents a NAC or a PAC is determined by whether it is contained in a Not formula.
Since LHS and PACs specify graph patterns required to be present, you may ask now if PACs just add a syntactic sugar on top of the LHS. This is not the case: PACs (and NACs) are used to constrain the matching of the LHS. However, they are not part of the computed match. A match only contains mappings for LHS nodes. This is particularly important in scenarios where you want to apply a rule for each computed match (compare rule-nesting, described below). For an example, please refer to Movies.
Identity of nodes in different graphs, including LHS, PAC, and NAC graphs, can be specified using mappings. The graphical editor maintains these mappings implicitly: For each LHS node, it creates a mapped node inside the NAC or PAC.
Examples: Bank Accounts, Ecore2RDB, Java2StateMachine, Ecore2Genmodel, Grid & Comb Pattern, Gossiping Girls, Probabilistic Broadcast, Movies
Rule-nesting is a powerful concept providing a for-each operator for rules. In nested rules, the outer rule is referred to as kernel rule and the inner rule as multi rule. During execution of a nested rule, the kernel rule is matched and executed once. Afterwards, the match is used as a starting point to match the multi-rule as often as possible and execute it for each match. Multi mappings allow to specify identity between kernel and multi-rule nodes. In the graphical editor, multi-rule nodes are indicated by a layered representation and an asterisk (*). Multiple layers of rule-nesting are allowed. The graphical editor shows the path constituted by the nested rules in the title bar of the respective nodes. An extensive example for multiple-layer rule-nesting is found in Ecore2RDB.
Examples: Bank Accounts, Ecore2RDB, Gossiping Girls, Probabilistic Broadcast
Further Reading: Biermann, Ermel, Taentzer. "Formal foundation of consistent EMF model transformations by algebraic graph transformation." Software & Systems Modeling 11.2 (2012): 227-250. Chapter 5. SpringerLink
Control flow: Units
In Henshin, control flow is specified using units. Units have a fixed number of sub-units, allowing for arbitrary nesting. The following unit types are available:
Sequential units have an arbitrary number of sub-units that are executed in the given order. The Boolean flag strict determines the behavior if one of the sub-units cannot be executed: In strict mode, the execution stops, otherwise, the next sub-unit is executed. The Boolean flag rollback determines whether in the case of a stopped execution previous executions are reverted.
Examples: Ecore2RDB, Java2StateMachine, Ecore2Genmodel, Grid & Comb Pattern, Movies
Priority Units have an arbitrary number of sub-units that are checked in the given order for executability. One sub-unit - the first one found to be executable - is executed.
Independent Units have an arbitrary number of sub-units that are checked in nondeterministic order for executability. One sub-unit - the first one found to be executable - is executed.
Iterated Units have one sub-unit. The sub-unit is executed as often as specified in the iterations property.
Examples: Grid & Comb Pattern
ConditionalUnits have either two or three sub-units: if, then, (and else). If a match for the if unit can be found, the then unit is executed. Otherwise, if present, the else unit is executed. Note that any changes specified in the if unit are ignored; it is used for matching only.
Execution-time variability: Parameters
Parameters allow to shape the behavior of units, including rules, at execution time with variable information. A parameter belongs to exactly one unit. Each parameter has a name, a description, and a parameter kind. The parameter kind specifies the time when the parameter is bound to a concrete value, and whether the parameter is intended to be accessed after the unit has been applied. There are four parameter kinds (in, out, inout, var) and an additional legacy parameter kind (unknown):
- in parameters need to be set externally before the unit application; they cannot be read once the unit has been applied,
- out parameters are set automatically during the unit application; they are intended to be read once the unit has been applied,
- inout parameters need to be set externally before the unit application; they are intended to be read once the unit has been applied,
- var parameters (variables) are set automatilly during the unit application; they are not indended to be unit once the unit has been applied,
- unknown parameters either can be set externally before or automatically during the unit application; they may be read once the unit has been applied.
Parameters can be set externally by the user using the API or the graphical interpreter wizard. In addition, they can be passed in from another unit using a parameter mapping. Out and var parameters specified for LHS elements (that is, deleted or preserved elements) can be set by the matchfinder to the values found in the respective objects. This behavior can be used to propagate values between LHS and RHS elements.
Parameters can optionally specify a type. The type is an arbitrary EClassifier, that is, either an EDataType such as EString, EInt or EBoolean, or any EClass.
Declared parameters can be used at any place in the unit where a string value is expected: In rules, this is the case for node names, attribute values, and attribute conditions. In iterated units, this is the case for the iterations condition. Parameters are accessed by name reference.
Examples: Bank Accounts, Ecore2RDB, Ecore2Genmodel, Grid & Comb Pattern, Gossiping Girls, Probabilistic Broadcast
Parameter mappings: In order to pass parameter values between units, it is required to define parameter mappings. A parameter mapping assigns a source parameter to a target parameter.
Examples: Ecore2RDB, Ecore2Genmodel, Grid & Comb Pattern