Comparing Model-Metamodel and Transformation-Metamodel Co-evolution Louis M. Rose 1 , Anne Etien 2 , David M´ endez 2,3 , Dimitrios S. Kolovos 1 , Richard F. Paige 1 and Fiona A.C. Polack 1 1 Department of Computer Science, University of York, UK. {louis, dkolovos, paige, fiona}@cs.york.ac.uk 2 INRIA Lille Nord Europe, LIFL CNRS UMR 8022, Universit´ e Lille 1, France. Anne.Etien@lifl.fr 3 TICSw Research Group, Universidad de los Andes, Colombia. df.mendez73@uniandes.edu.co Abstract. Changes to a metamodel can affect the definition and well- formedness of other development artefacts, such as models and trans- formations. Co-evolution, in which two or more artefacts are changed together, is one technique for managing change. This paper compares model-metamodel and transformation-metamodel co-evolution, and high- lights challenges for future work on co-evolution. 1 Introduction MDE aims to better automate engineering tasks, but introduces additional chal- lenges for software evolution due to interdependencies between engineering arte- facts. Models and transformations, for example, are interdependent with meta- models, because their definition and well-formedness depends on one or more metamodels. Interdependent artefacts complicate software evolution activities, because the evolution of one artefact may require others to be changed in re- sponse. The interdependency between a model and its metamodel is termed confor- mance [1], and can be described by a set of constraints between models and metamodels [10]. M´ endez et al. [9] identify a similar interdependency, domain conformance, between the definition of a transformation and its metamodels. Co-evolution is the process of evolving several artefacts together to main- tain an interdependency. This paper compares model-metamodel co-evolution and transformation-metamodel co-evolution, and synthesises challenges for their future research. 2 Existing Approaches Existing approaches to managing model-metamodel and transformation-meta- model co-evolution can be categorised as master-slave approaches [7]. In general,