Efficient ATL Incremental Transformations
Résumé
Incrementally executing model transformations offers several benefits such as updating target models in-place (instead of creating a new copy), as well as generally propagating changes faster (compared with complete re-execution). Active operations have been shown to of- fer performant OCL-based model transformation incrementality with use- ful properties like fine-grained change propagation, and the preservation of collection ordering. However, active operations have so far only been available as a Java library. This compels users to program at a relatively low level of abstraction, where most technical details are still present. Writing transformations at this level of abstraction is a tedious and error prone work. Using languages like Xtend alleviates some but not all issues. In order to provide active operation users with a more user-friendly front-end, we have worked on compiling ATL code to Java code using the active operations library. Our compiler can handle a significant subset of ATL, and we show that the code it generates provides similar performance to hand-written Java or Xtend code. Furthermore, this compiler also enables new possibilities like defining derived properties by leveraging the ATL refining mode.