Automated Inversion of Attribute Mapping Expressions for Multi-Model Consistency

Because keeping two models synchronized in a multi-view modeling scenario is a tedious and error-prone task, bidirectional model transformation approaches have been developed to automatically ensure model consistency. Many existing approaches focus on the synchronization of structural aspects, i.e. creating and deleting corresponding meta class instances as well as creating, updating and deleting references between meta classes. However, they lack the possibility to express complex mappings between the attributes of these meta classes which go beyond assigning a constant value or the value of one attribute to another attribute. In this thesis an approach is presented that allows synchronizing an attribute of one target meta class with a number of attributes of another corresponding source meta class by writing a single mapping expression using an expressive, Java-based mapping syntax. This mapping expression de nes how the target attribute’s value is computed using numerical, logical and string operations. A computation rule is then automatically generated that de nes how one of the source attributes will be updated given a changed target value and the complete set of source values. This is achieved by de ning so-called inversions for each of the supported operations. We demonstrate that the individual operations can be combined and nested while, whenever possible, obeying round trip laws that have been de ned for solving the related view-update problem. We then present an extensible architecture for implementing these inversions via code generation. Finally, we evaluate the selection of supported operations based on an analysis of the ATL transformation zoo to show that the operations we can invert make up a substantial portion of commonly used attribute mappings.