Csmith vs Fuzzilli
Csmith
github.comFuzzilli
github.comSide by side
| Figure | Csmith | Fuzzilli |
|---|---|---|
| Price from | Free | Free |
| Free plan | Yes | Yes |
| Free trial | — | — |
| Open source | — | — |
| Windows | Yes | — |
| macOS | — | Yes |
| Linux | Yes | Yes |
| Self-hosted | — | Yes |
| Plans published | 1 | 1 |
| Spec score · Fuzz Testing Software | 7.8 (#12) | 8.4 (#6) |
| License | The repository includes a COPYING license file. | The repository identifies its license as Apache-2.0. |
| Purpose | Csmith generates random C programs to find compiler bugs using differential testing. | Fuzzilli is a coverage-guided fuzzer for dynamic language interpreters that mutates programs in FuzzIL and translates them to JavaScript. |
Plans
In their own words
Csmith
Csmith is a free, open-source generator of random C programs for finding compiler bugs through differential testing. Its output avoids undefined behavior and includes statistics for each generated program. A documented workflow compiles programs with GCC and Clang, then compares their outputs to identify differences. The tool can also create material for a C test suite when users do not want to write one themselves. Command-line options shown by csmith -h and csmith -hh customize random generation. Csmith runs locally on Linux and Windows. Ubuntu source builds require g++, CMake, and m4, followed by CMake and make; Windows builds are documented with Visual Studio or WSL. The project says it should build on most modern machines with a C++ compiler. Generated programs may loop indefinitely, so execution timeouts are recommended. The repository includes a COPYING license file. Bug reports and suggestions go through GitHub issues, with discussion on a mailing list; maintainers warn replies may be delayed.
Fuzzilli
Fuzzilli is a free, open-source, coverage-guided fuzzer for dynamic language interpreters. It represents test programs in FuzzIL, mutates them, and translates them to JavaScript for execution. Its mutators can change data flow or operation parameters, generate or splice code, and combine corpus programs. Components include a mutation fuzzer, script runner, corpus, environment, minimizer, evaluator, and lifter. A target engine runs repeated scripts in a read-eval-print-reset loop, resetting between cases; multiple instances can synchronize within a process or across machines over TCP. Listed target directories include JavaScriptCore, JerryScript, QuickJS, QtJS, Serenity, SpiderMonkey, V8, XS, Duktape, and njs. Setup requires compiling a supported engine with coverage instrumentation using clang 4.0 or later, then building Fuzzilli with Swift Package Manager. Docker and Google Compute Engine tooling are provided. The project supports Linux, macOS, and self-hosted use and identifies its license as Apache-2.0. Its repository says it is not an officially supported Google product. Contributions require a Contributor License Agreement and are reviewed through GitHub pull requests.