Dmitry Vyukov

Unfortunately it's not that simple. `sonarRegion` is shared memory region. sync.Mutex does not work across processes. I would also be concerned about performance of mutex approach.

Ideally we have some shared memory protocol that allows to understand what slots are fully-written and what are not.

Are you also overriding runtime.GOMAXPROCS? go-fuzz sets GOMAXPROCS to 1. And I can't reproduce the crashes without overriding GOMAXPROCS. Both background goroutines and overriding GOMAXPROCS is bad for fuzzing? Can...

I've submitted a reproducer for the bug. But it does not reproduce by default, only if runtime.GOMAXPROCS call is uncommented. I don't want to uncomment it, because it's not the...

The race detector can find these bugs even with GOMAXPROCS=1. Also they are usually not considerably affected by the function input. I mean if something starts a goroutine, then it...

> I could maybe build a test case that exhibits the bug. That would be useful. > re race detection, I don't see where go-fuzz turns on race detection? It...

I don't think anything has changed here since the last message. The Open status of the issue is still valid.

Long term it would be good to get rid of the archive as it makes integration with other systems harder. I would inject the version into the generated main function...

As far as I understand AFLs use of XOR is an attempt to get edge coverage on top of the weak binary instrumentation. LibFuzzer uses compiler instrumentation which does instrument...

1. Sensibly benchmarking fuzzing changes is hard. As you noted this can inflate corpus. What is the effect on very large programs? If we inflate corpus from 1000 to 5000...