xeno74 wrote: ↑Sun Nov 21, 2021 8:09 am
Yes, I have. Could you please test the first test kernels carefully? If you don’t test the kernels carefully then the next test kernels will have
all wrong bisect results.
Yes I will... I have already been testing rb4 two days, no problems seen so far. But to be sure we are on the right track, I retested today rb2 which was originally the first 'bad' test kernel, and now I got the problem visible as soon as I opened a stream in VLC. So, rb2 is definitely 'bad'! Now I'm retesting rb3 to check if it is really 'good' as it seemed to be in earlier tests.
It came tro my mind.... Would it be possible to test this kind of issue also with 'fifty-fifty' method, so that you would always have a pair of kernels to test side-by-side, one of which forcibly includes the problematic component? In theory the continuous bisect method is faster, but when the problem is not automatically seen, fifty-fifty method would give faster the correct result as you can switch to test the other kernel of the pair if the issue is not immediately seen with the other one.