Improving Incremental and Bidirectional Evaluation with an Explicit Propagation Graph

Abstract

Active operations enable bidirectional incremental evaluation of OCL-like expressions on collections: changing the source (resp. the result) of an expression causes corresponding updates in the result (resp. the source). The current evaluation approach of active operations is based on the observer pattern. Previous work showed how they can be used for model transformation, and that they can scale to processing large models while maintaining collection ordering. However, observation makes the directed acyclic propagation graph implicit, and imposes a depth-first traversal. This sometimes results in unwanted transitory states, which uselessly increase the amount of computations required for propagating some changes.

To address this issue, we propose in this paper to make the propagation graph explicit. This enables separation of the propagation graph from traversal strategies (e.g., breadth-first instead of depth-first). We show how this approach gets rid of unwanted transitory states, and discuss some of its other advantages like enabling more efficient graph visualization and analysis, as well as more compact propagation graph representations. Additionally, incremental algorithms of active operations do not need to be changed, but can actually be better encapsulated, which decreases maintenance cost of the incremental framework.