Efficient Model Transformations by Combining Pattern Matching Strategies

Recent advances in graph pattern matching techniques have demonstrated at various tool contests that graph transformation tools can scale up to handle very large models in model transformation problems. In case of local-search based techniques, pattern matching is driven by a search plan, which provides an optimal ordering for traversing and matching nodes and edges of a graph pattern. In case of incremental pattern matching, matches of a pattern are explicitly stored and incrementally maintained upon model manipulation, which frequently provides significant speed-up but with increased memory consumption. In the current paper, we present a hybrid pattern matching approach, which is able to combine local-search and incremental techniques on a per-pattern basis. Based upon experimental evaluation, we identify scenarios when such combination is highly beneficial, and provide guidelines for transformation designers for optimal selection of pattern matching strategy.