make benchmarks standalone
- Dominant language
- Python
- Stars
- 1.2k
- Forks
- 302
- PR merge metrics
- No merged PRs in 30d
Description
Most of our benchmarks rely on system dependencies such as zlib, etc
This has the following problems/limitations:
1. We're basically fuzzing not just the benchmark code, but benchmark code + dependencies. If we want to fuzz the dependencies, we should have a dedicated benchmark fo them. Example for zlib
2. The dependencies are not instrumented by the fuzzer during compilation because it uses system-installed libraries. This means if a fuzzer reaches dependency code, no feedback is received.
3. KLEE cannot handle external dependencies well: it needs all the code (besides libc) in one binary.
For these reasons, we should make benchmarks standalone, ie no dependent on other libraries as much as possible
Contributor guide
Assessment
This issue has not been assessed yet.