Towards Efficient and Scalabale Omniscient Debugging for Model Transformations

Abstract

Model transformations (MTs) are central artifacts in modeldriven engineering (MDE) that define core operations on models. Like other software artifacts, MTs may possess defects (bugs). Some MDE tools provide support for debugging. In this paper, we describe an omniscient debugging technique. Our technique enhances stepwise execution support for MTs by providing the ability to traverse, in either direction, the execution history of a live debugging session. We also introduce a proof of concept prototype applying the described technique and a preliminary study of the scalability, in terms of memory consumption, and performance, in terms of time to execute.

Omniscient debugging for executable DSLs

“…Corley et al [57] propose omniscient debugging facilities for the cloud-based modeling solution AToMPM, in order to step both forward and backward in model transformations executed in an AToMPM runtime. Yet it does not support languages that are not model transformation languages.…”

Advanced and efficient execution trace management for executable domain-specific modeling languages

“…For Petri nets, the only possible change is within a Place object (line 12), more precisely the amount of tokens it holds (line 14). In that case, a new TokensValue object is created to replace the former one within the copied ExecutionState (lines [15][16][17][18][19]. Finally, this new TokensValue is added to the tokensSequence dimension in the corresponding TracedPlace object (line 20).…”

“…For this, we have adapted our semantic model differencing framework [43] to analyze multidimensional domain-specific execution traces instead of generic clone-based ones. 16 Considered xDSML. Based on this adapted semantic differencing framework, we have performed a case study on the real world xDSML foundational UML (fUML) [55] that we presented in Sect.…”

“…A large amount of domain-specific modeling languages (DSMLs) has been proposed and used to model the behavior of systems [6,25,31,52,55]. Early dynamic verification and validation (V&V) techniques, such as omniscient debugging [9,16], semantic differencing [43], and runtime verification [44], are necessary to ensure that such models are correct. These techniques require models to be executable, which can be achieved by defining the execution semantics of the DSMLs used to define them.…”

“…Sometimes called micro-step[16,34] 8. Sometimes called macro-step[16,34], combo step[24], or compound step[34].…”