A testing methodology for a dataflow based visual programming language

Dataflow based visual programming languages have become an important topic of research in recent years, yielding a variety of research systems and commercial applications. As with any programming language, visual or textual, dataflow programs may contain faults. Thus, to ensure the coma functioning of dataflow programs, and increase confidence in the quality of these programs, testing is required. Despite this valid observation, we find that the casting criteria found in the literature mainly addressed imperative, declarative, and form-based languages. However, we did not find any discussion that specifically addressed testing criteria for dataflow programs. In this paper, we investigate, from a testing perspective, differences between dataflow and imperative languages. The results reveal opportunities for adapting code-based control-flow testing criteria to test dataflow languages. We show that our proposed testing methodology is well suited for dataflow programs. In particular, the "all-branches" criterion provides important error detection ability, and can be applied to dataflow programs. We implemented a testing system that allows users to visually and empirically investigate the testability of programs written in the visual programming language Prograph. Our empirical results confirm that, analogous to imperative languages, the all-branches criterion cannot detect all the errors in a dataflow program. Thus, to catch those undetected errors, more rigorous testing should be applied.