Crash deduplication is the process of identifying and grouping identical or similar crash reports to reduce redundancy and improve debugging efficiency.
kotlinx.fuzz currently uses Jazzer as its main fuzz engine. Jazzer already integrates some crash deduplication techniques, however, only the most basic ones.
Mainly, Jazzer can deduplicate only crashes that have exactly similar stack traces. Unfortunately, that does not reduce the number of crash reports significantly and Jazzer still requires a large amount of manual work to sort the results.
In kotlinx.fuzz we use LibCASR, a Rust library that provides API for parsing stacktraces, collecting crash reports, triaging crashes (deduplication and clustering), and estimating severity of crashes. We implemented our custom casr-adapter library that provides JVM bindings for necessary APIs of LibCASR. This allows kotlinx.fuzz significantly reduce the amount of crashes found during the fuzzing campaign and also ensures that only unique crashes are shown to the user.