Finding Unstable Code via Compiler-Driven Differential Testing

Unstable code refers to code that has inconsistent or unstable runtime semantics due to undefined behavior (UB) in the program. Compilers exploit UB by assuming that UB never occurs, which allows them to generate efficient but potentially semantically inconsistent binaries. Practitioners have put great research and engineering effort into designing dynamic tools such as sanitizers for frequently occurring UBs. However, it remains a big challenge how to detect UBs that are beyond the reach of current techniques.

In this paper, we introduce compiler-driven differential testing (CompDiff), a simple yet effective approach for finding unstable code in C/C++ programs. CompDiff relies on the fact that when compiling unstable code, different compiler implementations may produce semantically inconsistent binaries. Our main approach is to examine the outputs of different binaries on the same input. Discrepancies in outputs may signify the existence of unstable code. To detect unstable code in real-world programs, we also integrate Com-pDiff into AFL++, the most widely-used and actively-maintained general-purpose fuzzer.

Despite its simplicity, CompDiff is effective in practice: on the Juliet benchmark programs, CompDiff uniquely detected 1,409 bugs compared to sanitizers; on 23 popular open-source C/C++ projects, CompDiff-AFL++ uncovered 78 new bugs, 52 of which have been fixed by developers and 36 cannot be detected by sanitizers. Our evaluation also reveals the fact that CompDiff is not designed to replace current UB detectors but to complement them.