1. Combining Generative and Graph Transformation Techniques for Model Transformation: An Effective Alliance?
Shane Sendall
Software Modeling & Verification Lab.
Computer Science Department
University of Geneva
Switzerland

2. Shane Sendall, CUI, University of Geneva
Overview
Model Transformation
Overview of Gmorph
Graph Transformation/Generative Programming
Closing Remarks
Don’t treat language features in a systematic nor thorough way
Defer any discussion of GP and GT until example: I assume some familiarity
2 December 2018
Shane Sendall, CUI, University of Geneva

3. Model Transformation—What
A model transformation applies one or more rules according to the presence of certain patterns of elements in the set of source models, resulting in a set of target models.
The application of a rule results in any number of the following:
the creation of a new model fragment
the deletion of an existing model fragment
the modification of an existing model fragment
(where model fragment is any thing from a model to a model element)
In the context of MDA, model transformation can be categorized as: language translation and model evolution
2 December 2018
Shane Sendall, CUI, University of Geneva

4. Genermorphous (Gmorph)
Gmorph is a language (still under development) for specifying and executing model transformations
Goal:
push imperative approach toward some advantages of declarative approaches, such as, pattern matching, rule application, etc.
Context:
an experimental language that is not “fully featured”; it does not address all aspects that are useful for mainstream model transformation, e.g., reuse mechanisms, well-formedness checks, concurrent application of rules, etc.
2 December 2018
Shane Sendall, CUI, University of Geneva

5. Genermorphous (Gmorph)
A hybrid approach that offers a (partially) visual transformation language, which address both model evolution and translation categories:
Rule Specification language (mixture of Graph Transformation Rules and Generative Programming Principles)
Rule Composition language (small imperative language: choice, sequence, iteration)
Relation to QVT: results from Gmorph could be fed back into a QVT language
2 December 2018
Shane Sendall, CUI, University of Geneva

6. Shane Sendall, CUI, University of Geneva
Graph Transformation
 Graph Rewriting
LHS (graph) = left-hand side of rule = pattern
RHS (graph) = right-hand side of rule = pattern
(Interface = intersection of LHS & RHS)
Steps taken in applying a rule
match, remove, glue, embed C E D A C B D
RHS
LHS
2 December 2018
Shane Sendall, CUI, University of Geneva

7. GT: relevance to Gmorph
LHS and RHS
Give a ‘before’ and ‘after’ view of rule application
For model evolution transformations: rule is a graph rewrite
2 December 2018
Shane Sendall, CUI, University of Geneva

8. Generative Programming
A generative programming consists of three parts: a problem space, a solution space, and the configuration knowledge.
Problem space defines the appropriate domain-specific concepts and features.
Solution space defines the target model elements that can be generated and all possible variations.
Configuration knowledge specifies illegal feature combinations, default settings, default dependencies, construction rules, and optimization rules.
2 December 2018
Shane Sendall, CUI, University of Geneva

9. GP: relevance to Gmorph
Ability to handle variations in source model element matching and target model generation (language translation)
Parameterization and “open-world” model of matching
Default and configuration information are part of specification
2 December 2018
Shane Sendall, CUI, University of Geneva

10. Closing Remarks and Future Directions
Issues
Pattern matching at instance level
Optional and multiple matches
Transitive closure and sequence of matches
Ability to easily define mappings between instances:
Current approach: 1-to-1, 1-to-many, many-to-1 by traversal of LHS
Interesting alternative: an attribute grammar-like approach, which implicitly traverses LHS
2 December 2018
Shane Sendall, CUI, University of Geneva

11. Closing Remarks and Future Directions
Issues
Reversing transformations and model updating (round-tripping) => (persistent) relations
Separation between rule language and rule composition language => to be fine-tuned with usage
A concrete syntax that doesn’t distort the concrete syntax of source model too much (work back from metamodel instance view)
2 December 2018
Shane Sendall, CUI, University of Geneva

12. Shane Sendall, CUI, University of Geneva
2 December 2018
Shane Sendall, CUI, University of Geneva

13. Shane Sendall, CUI, University of Geneva
Example using Gmorph
Problem: Transform a Java package and its contained abstract classes to a UML package and UML classes that correspond to these Java classes.
2 December 2018
Shane Sendall, CUI, University of Geneva

14. Shane Sendall, CUI, University of Geneva
LHS of Gmorph Rule Spec
Metamodel: JavaNB
javaP: JavaPackage
/javaPackage
X, 0..*
/classes
0..*
/JC: JavaClass
/declaredBy
/JF: Field
/features
Additional Selection Conditions:
context JC ensure:
self.isInterface = false and JC classes must not be interfaces
self.modifier = ModifierKind::abstract JC classes must be abstract
Model element instance level
2 December 2018
Shane Sendall, CUI, University of Geneva

15. Shane Sendall, CUI, University of Geneva
RHS of Gmorph Rule Spec
Metamodel: UML
umlP: Package
/ownedMember
A, 0..*
/owningPackage
0..*
/UC: Class
/ownedAttribute
/AT: Attribute
/class
Model element instance level
2 December 2018
Shane Sendall, CUI, University of Geneva

16. Shane Sendall, CUI, University of Geneva
RHS of Gmorph Rule Spec
Metamodel: UML
<umlP>
<UC> {A, 0..*}
<AT> {0..*}
Model element instance level
2 December 2018
Shane Sendall, CUI, University of Geneva

17. LHS/RHS of Gmorph Rule Spec
Metamodel: JavaNB
Metamodel: UML
umlP: Package
javaP: JavaPackage
/javaPackage
/ownedMember
X, 0..*
A, 0..*
/classes
0..*
/owningPackage
0..*
/JC: JavaClass
/declaredBy
/JF: Field
/UC: Class
/ownedAttribute
/AT: Attribute
/features
/class
Additional Selection Conditions:
context JC ensure:
self.isInterface = false and JC classes must not be interfaces
self.modifier = ModifierKind::abstract JC classes must be abstract
Model element instance level
2 December 2018
Shane Sendall, CUI, University of Geneva

18. Shane Sendall, CUI, University of Geneva
Gmorph Rule Script
Transform Rule:
JavaNB-to-UML_PackageMapping (in javaP: JavaNB::Package, out umlP: UML::Package)
Instance Mapping:
javaP --> umlP;
mapeach x in X onto a in A {
x.JC --> a.UC;
mapeach srcField in x.JC.features onto tgtAttr in a.ownedAttribute {
srcField --> tgtAttr; }
Type Mapping:
UML::Package <-> JavaNB::Package { }
UML::Class <-> JavaNB::JavaClass {
LHS.name <-> RHS.simpleName |
LHS.isRoot <- RHS.superClassName.size() = 0 | …
2 December 2018
Shane Sendall, CUI, University of Geneva

19. Gmorph Rule Script (cont’d)
-- Type Mapping (cont’d)
UML::Attribute <-> JavaNB::Field {
LHS.name <-> RHS.name |
LHS.scope <-> RHS.scope |
LHS.visibility <-> RHS.visibility |
false -> RHS.isVolatile |
LHS.isDerived <- false | … }
UML::VisibilityKind <-> JavaNB::VisibilityKind {
LHS.vk_public <-> RHS.public |
LHS.vk_protected <-> RHS.protected |
LHS.vk_private <-> RHS.private |
LHS.vk_package <-> RHS.package
2 December 2018
Shane Sendall, CUI, University of Geneva

20. Closing Remarks and Future Directions
Advantages of Gmorph
cognitive and practical advantages for concrete syntax approach (WYSIWYG) and explicit LHS and RHS (the before and after transformation view)
Patterns described as object structures works well with “imperative” style traversal
2 December 2018
Shane Sendall, CUI, University of Geneva

21. Shane Sendall, CUI, University of Geneva
Other Approaches
IBM/DSTC QVT (logic language)
Reversibility (due to unification)
Usability, Performance?
QVTP (relation-based approach)
Nice expression of 1-to-1 language translation
Awkward for complex model evolution trans.?
Sun/Compuware QVT (logic language?)
Reversibility, modularity
Feasibility of using OCL to fully describe LHS/RHS?
UMLX
Graphical notation based on Graph Transformation
Feasibility of encoding complex trans. graphically?
Language translation transformations?
2 December 2018
Shane Sendall, CUI, University of Geneva

22. Shane Sendall, CUI, University of Geneva
Matching Features
Additional Features
Optional matches for one or more p-elements
p-element/group of p-elements are optionally matched (matching is mandatory by default)
Maximum possible number of p-element matches are made
Multiple matches for one or more p-elements
Collection of tuples (tuple of p-elements scoped by matching container)
Transitive closure matches for one or more p-elements
Collection of tuples
Fixpoint iteration
Sequence of matches for one or more p-elements
Sequence of tuples
Ordered according to specified criteria
2 December 2018
Shane Sendall, CUI, University of Geneva

23. OMG’s Model Driven Architecture Initiative
Three levels:
Platform Independent Model(s)
Platform Specific Model(s)
Code & Deployment/Configuration Information
The big issues in automation support for MDA:
Consistency and synchronization between levels (vertical and horizontal)
Support in construction tasks
MOF 2.0 Query/View/Transformation Standardization
The important point is parameterized transformations and synchronization between and within levels
2 December 2018
Shane Sendall, CUI, University of Geneva

24. Model Transformation—Why
Categories:
Model Translation
inter-model mapping, synthesis, splitting, and updating
Model Evolution
source model is target model (same/copy)
Examples in MDA:
Model Synchronization, Refinement, and Forward and Reverse Engineering
PIM
We can determine two main categories of transformation in MDA
PSM
PSM
2 December 2018
Shane Sendall, CUI, University of Geneva

25. Model Transformation—Why
Examples (cont’d)
Synthesis of views
E.g., bringing together the protocol view with the operation view to form the behavior model
Generation of Views
E.g., generating a scenario from a statemachine
Application of Software Patterns and Refactoring
E.g., applying the Singleton design pattern to a design model
Protocol View
UML State Diagram
UML Sequence Diagram
Behavior Model
Operation
View
2 December 2018
Shane Sendall, CUI, University of Geneva

26. Model Transformation—How
Declarative approaches
Logic languages
Graph Rewriting languages
Relation/Mapping languages
Imperative, direct model manipulation approaches
<your favorite mainstream language> + API to model repository of tool
XML-based approaches
XSLT (and co.)/XMI.difference (XMI -> XMI)
Generative approaches
Template/Frame languages
Transformation generator languages
Hybrid approaches
Mix declarative and imperative styles
2 December 2018
Shane Sendall, CUI, University of Geneva

27. Shane Sendall, CUI, University of Geneva
Graph Transformation
 Graph Rewriting
LHS (graph) = left-hand side of rule = pattern
RHS (graph) = right-hand side of rule = pattern
(Interface = intersection of LHS & RHS)
Steps taken in applying a rule
match, remove, glue, embed C E D A C B D
RHS
LHS
2 December 2018
Shane Sendall, CUI, University of Geneva

28. Shane Sendall, CUI, University of Geneva
Graph Transformation
Issues
Dangling Edges
Commutativity of rules
Triggering Conditions
Embedding rules
Efficient algorithms for LHS matching
Reversibility of rules (RHS  LHS)
2 December 2018
Shane Sendall, CUI, University of Geneva

29. Shane Sendall, CUI, University of Geneva
JavaNB Metamodel
2 December 2018
Shane Sendall, CUI, University of Geneva