Michael Reif

Software Technology Group @ TU Darmstadt

Hermes: assessment and creation of effective test corpora

by Michael Reif, Michael Eichberg, Ben Hermann, and Mira Mezini

Abstract: An integral part of developing a new analysis is to validate the correctness of its implementation and to demonstrate its usefulness when applied to real-world code. As a foundation for addressing both challenges developers typically use custom or well-established collections of Java projects. The hope is that the collected projects are representative for the analysis’ target domain and therefore ensure a sound evaluation. But, without proper means to understand how and to which degree the features relevant to an analysis are found in the projects, the evaluation necessarily remains inconclusive. Additionally, it is likely that the collection contains many projects which are – w.r.t. the developed analysis – basically identical and therefore do not help the overall evaluation/testing of the analysis, but still cost evaluation time. To overcome these limitations we propose Hermes, a framework that enables the systematic assessment of given corpora and the creation of new corpora of Java projects. To show the usefulness of Hermes, we used it to comprehend the nature of the projects belonging to the Qualitas Corpus (QC) and then used it to compute a minimal subset of all QC projects useful for generic data- and control-flow analyses. This subset enables effective and efficient integration test suites.

Resources

BibTeX

@misc {REHM17,
  title = {{Hermes: assessment and creation of effective test corpora}},
  author = {Reif, Michael and Eichberg, Michael and Hermann, Ben and Mezini, Mira},
  booktitle = {{Proceedings of the 6th ACM SIGPLAN International Workshop on State Of the Art in Program Analysis}},
  series = {SOAP 2017},
  pages = {43-48},
  year = {2017},
  doi = {10.1145/3088515.3088523},
  url = {http://dx.doi.org/10.1145/3088515.3088523},
}