Abstrakt | Value analysis is an important building block in static program analysis. While several approaches have been proposed, evaluating and comparing them is not trivial. Up to this day, a reliable and large benchmark specifically for value analysis is missing. Such a suite must not only provide test cases, but also a ground truth with the correct values to be found. In this paper, we propose ValBench, an extensible value benchmark suite consisting of 372 test cases for Java analysis and 59 test cases for Android analysis tools. Furthermore, we present an evaluation framework that automatically generates a ground truth for these test cases, identifies their respective challenges for program analysis and orchestrates the execution and result collection on the various value analysis tools. We further present an evaluation of 7 existing value analysis tools on ValBench and highlight the challenges faced by these tools as an empirical overview over the state of the art in value analysis. |
---|