One successful case of using rr to track down a subtle bug in LuaJIT's JIT compiler

Yichun Zhang yichun at openresty.com
Fri Feb 8 22:46:19 UTC 2019


Hi folks,

Not sure if this is the right mailing list for such things (if not,
please let me know). rocallahan once asked to me share some good
experiences of using rr to debug LuaJIT bugs, and below is such a
case:

    https://www.freelists.org/post/luajit/Patch-to-fix-a-Assertion-nslots-250-failed-assertion-failure-in-LuaJIT-v21

It's not really a blog post but a mailing list email. Still I think it
has enough details to make this case interesting for some others in
the rr user community (if any of you needs more details, please let me
know).

Many thanks for creating and improving rr! In the past, debugging such
bugs can take a notoriously long period of time and a lot of hacks to
gather needed information in the execution process. This is because
it's very hard to prepare minimal examples for such bugs and even for
big ones it is not always reproducible due to inherent nondeterminism
in the JIT compiler's runtime heuristics. Now it becomes fun rr
exercises for just tens of minutes as long as we can reproduce it for
at least once :)

Best regards,
Yichun


More information about the rr-dev mailing list