Hmm, after 4 hours running, I managed to make it crash under valgrind โ€ฆ not sure if that will be enough to pin down the bug โ€ฆ maybe the next step is to try to reproduce using rr โ€ฆ
2
I have not properly tried `rr` before - it sounds promising
1
Hmm, so I have tried `rr record` and it has wedged hard - the computer is working OK but `rr` and the child process it is tracing are unkillable :-(
1
Need a newer kernel for `rr`!
1
I had to power cycle the box to unwedge it - it got stuck while shutting down! but now I have a rather less ancient kernel. let's see if `rr` works better now...
1
much better :-) and `rr` appears to be a lot faster than `valgrind`
2
1
Replying to @fanf
my favourite thing about rr is reverse-continue combined with watch. Want to know how a variable became corrupt? Watch it and back up until it happened. Amazing boost to debug productivity.

Apr 6, 2018 ยท 1:36 PM UTC

1
1
Replying to @dsilverstone
Now if I can get this thing to crash under `rr` i might be on to something :-)