This is an attempt at building a fuzzer that uses American Fuzzy Lop's instrumentation, but in Python. Currently it doesn't do much.
As of today, I hadn't rewritten afl-gcc/afl-as, so you need to use American Fuzzy Lop's compiler/assembler wrappers to build an instrumented version of your binaries. Once you do that, you can run them like this (assuming that you're fuzzing GNU bison):
$ ./apr-fuzz /path/to/instrumented/bison /dev/stdin
max=1038 execs/s=457.37 execs=458 crashes=0 hangs=2 elapsed=a second
max=1042 execs/s=511.20 execs=1024 crashes=0 hangs=3 elapsed=2 seconds
max=1042 execs/s=454.00 execs=1364 crashes=0 hangs=7 elapsed=3 seconds
max=1042 execs/s=465.60 execs=1865 crashes=0 hangs=9 elapsed=4 seconds
max=1060 execs/s=488.54 execs=2446 crashes=0 hangs=10 elapsed=5 seconds
(list goes on...)
Yup! That's it. As I said, it currently doesn't do much.
My aim is to create an alternate implementation of AFL that is easier to extend. If you know enough about AFL's fuzzing engine to help me rewrite it, I'll be more than happy to accept a pull request.