Section: New Results

Model Transformation

Participants : Frédéric Jouault [ contact ] , Jean Bézivin.

As a legacy of our previous work there are a number of assets that we may use [50] , [47] , [48] , [49] , [51] , [52] [10] . The ATL virtual machine and the ATL development environments are stabilized. There is a large user community and an initial library of transformations. Part of these assets are being industrialized with the help of a local company (Obeo) but kept in the open-source mode. The resources freed in this process are invested in pursuing research on model transformation.

In 2009, AtlanMod studied four specific model transformation issues:

• Performance of model transformation applied to large models. This work notably lead to the publication of a case study at the 5th International Workshop on Graph-Based Tools (GraBaTs'2009). This case study [33] focuses on the implementation of program comprehension tasks using model transformation tools. In this domain, large models obtained by parsing source code, and consisting of several millions of elements need to be processed. AtlanMod also submitted a solution [32] to this case study, as well as a solution to a synthesis case study around BPEL and BPMN [31] in the same workshop.

• Complex model navigation in model transformation relies in some cases on OCL expressions, which are more verbose than necessary (e.g., to express a JOIN, a transitive closure). This leads to complex code that is difficult to write, maintain, and optimize. In order to address this issue, we have worked on the HLN [23] (Higher-Level Navigation) language. This language defines explicit language constructs for some advanced navigation problems.

• Higher-Order Transformations (HOTs) enable complex transformation scenarios in which a transformation may for instance generate other transformations. It is possible to write such transformations in ATL because every ATL transformation is a model. After having used HOTs during several years, we have started a systematic study [26] of their applications.

• Transformation chaining constraints need to be well understood before composing transformations. Otherwise, invalid combinations may yield to errors. However, these constraints are not always obvious (e.g., they may depend on source and target metamodel subsets). In [14] , we have studied the automatic discovery of some chaining constraints.

Additionally, AtlanMod studied the application of ATL in system validation chains [13] .