Fix stalling when no inputs hit rare branch of interest

[carolemieux]

One one run of xmllint, the rare branches queueing mechanism completely stalled, as if no inputs hit the rare branch of interest. This is strange since every input which hits a new branch is saved to the queue (increment_hit_bits is only called in save_if_interesting). Need to reproduce + investigate.

[carolemieux]

Possible diagnosis: non-deterministic behavior in an already-saved input. Is increment_hit_bits called during calibration?

[hexcoder-]

Hi,
is there any solution/progress on this issue?

I saw some stalls in my tests some time ago also.

If this issue is resolved, we want to evaluate your great extension,
and try to integrate it in AFLplusplus (AFLplusplus/AFLplusplus#18 (comment))

Thanks!

[carolemieux]

Hmmm, I hadn't encountered this issue in a while. I can probably fix this by adding some randomization to the input selection process. Do you have a (sometimes) reproducible example where the stalling occurs, perchance?