Feature: Make executors and feedbacks easier to use outside of the fuzzing loop #2511
Conversation
…utside of LibAFLs Fuzzer loop
Lukas-Dresel
changed the title
Lukas-Dresel
changed the title
Lukas-Dresel
changed the title
|
@addisoncrump can you take a look? It might interfere with #2438 somewhat (?) |
ideally the normal run_target would already be reusable / have minimal generic bounds/ have sensible Nop Variants for easy use. |
|
Sorry for the delay -- lots of IRL stuff right now. I will look at this tomorrow. |
|
Sorry for the delay, I think we can merge this. |
|
wait, i'll check this when i refactor (or before) executor. |
|
Just checking in, what's the status of this? Is the idea still to merge this or replace it by other changes? |
| testcase: &mut Testcase<S::Input>, | ||
| ) -> Result<(), Error> | ||
| where | ||
| S: State + UsesInput, |
There was a problem hiding this comment.
Can you try to not use UsesInput here?
You can specify
S: HasCorpus and make the input to be <S::Corpus as Corpus>::Input
| &mut self, | ||
| state: &mut S, | ||
| _manager: &mut EM, | ||
| _input: &<S as UsesInput>::Input, |
| testcase: &mut Testcase<S::Input>, | ||
| ) -> Result<(), Error> | ||
| where | ||
| S: UsesInput + HasCorpus<Input = NautilusInput> + HasMetadata, |
|
Sorry for the late review. |
| testcase: &mut Testcase<S::Input>, | ||
| ) where | ||
| OT: ObserversTuple<S>, | ||
| S: UsesInput, |
|
|
||
| impl<S, T> Feedback<S> for ListFeedback<T> | ||
| where | ||
| S: State + HasNamedMetadata, |
| manager: &mut EM, | ||
| input: &<S as UsesInput>::Input, | ||
| _manager: &mut EM, | ||
| _input: &<S as UsesInput>::Input, |
|
tell me if you want me to merge from main and resolve conflicts |
|
@tokatoka if you find the time let's just finish/merge this for 0.14, lukas has done his part I think :) |
|
Finishing this in #2637, thanks for your help @Lukas-Dresel ! :) |
…zzing loop (extends #2511) (#2637) * feat(libafl_core): make executors and feedbacks more cleanly usable outside of LibAFLs Fuzzer loop * cargo +nightly fmt * updated type constraints * reformatted and final type constraint fixes * made unicode extraction stage useful separately * fix libafl_cc error message * fix state type constraint to be constrained on the method * removed unnecessary observer constraint * renamed unused variables * fix unnecessary error wrapping in helper functions * converted unicode conversion stage into associated function and fixed nautilus changes * more update * Remove extra I * more fmt * bounds? * less bounds * more less bounds * different trait bounds again * more less generics * fix unicode * fix list * remove unneeded bound --------- Co-authored-by: Lukas Dresel <[email protected]> Co-authored-by: Toka <[email protected]>
It would be nice to be able to use LibAFL components more easily outside of LibAFL's fuzzer harnessing (by which I mean the plug-and-play fuzzing loop) to allow for more reuse.
For example, I found myself ripping my hair out over trying to just run the forkserver with an observer to simply collect, aggregate and plot coverage. To do so, I had to provide types for fuzzers, feedbacks, schedulers, etc. even though none of those things were relevant for my use-case.
I ended up making changes like the ones in this PR in my own fork and it made it quite a bit nicer to use, now you can reuse some of the components and are only required to satisfy the directly needed type constraints for these components.