Effective Bug Detection with Unused Definitions

Unused definitions are values assigned to variables but not used. Since unused definitions a re u sually c onsidered redundant code causing no severe consequences except for wasting CPU cycles, system developers usually treat them as mild warnings and simply remove them. In this paper, we reevaluate the effect of unused definitions and discover that some unused definitions could indicate non-trivial bugs like security issues or data corruption, which calls for more attention from developers.

Although there are existing techniques to detect unused definitions, it is still challenging to detect critical bugs from unused definitions because only a small proportion of unused definitions are real bugs. In this paper, we present a static analysis framework ValueCheck to address the challenges of detecting bugs from unused definitions. First, we make a unique observation that the unused definitions on the boundary of developers' interactions are prone to be bugs. Second, we summarize syntactic and semantic patterns where unused definitions a re i ntentionally written, which should not be considered bugs. Third, to distill bugs from unused definitions, we adopt the code familiarity metrics from the software engineering field to rank the detected bugs, which enables developers to prioritize their focus.

We evaluate ValueCheck with large system software and libraries including Linux, MySQL, OpenSSL, and NFSganesha. ValueCheck helps detect 210 unknown bugs from these applications. 154 bugs are confirmed by developers. Compared to state-of-the-art tools, ValueCheck demonstrates to effectively detect bugs with low false positives.

• Software and its engineering → Software verification and validation.