Jan Midtgaard

> This could, of course, be an entirely unrelated issue, but then I wonder why it would only fail on this test precisely -- maybe the sheer luck of picking...

I've now completed a round of `git bisect` CI golf, and the finger points at: - bdd8d96 - Make the GC compact again (#12193) with the latest run available here:...

I accidentally kicked off a run with an even smaller heap (`"s=2048"`). Among a couple of assertion failures, this triggered the following which confirms my suspecion of a memory corruption...

I've tried to run under gdb batch mode in the CI: https://github.com/ocaml-multicore/multicoretests/actions/runs/9388010576/job/25852264722 For the 6 assertion failures this doesn't add much new info: ``` Starting 49-th run [Thread debugging using...

I've dug some more into this issue. Experiments reveal that this still can trigger without `split_from_parent` and omitting either `Get` or `Set` commands entirely. This indicates that there is an...

Using tmate I've also managed to log into the GitHub action runner machines, reproduce crashes there, and observe backtraces. A backtrace for an assertion failure run with each thread annotated...

I've finally managed to reproduce this locally by using `musl`. It turns out, this is not restricted to 32-bit but reproduces with a `musl` bytecode compiler. I've shared repro steps...

Fixed in https://github.com/ocaml/ocaml/pull/13553

Thanks for this overview! :pray:

Found another occurrence of this causing a live/deadlock: https://github.com/ocaml-multicore/multicoretests/actions/runs/5242663626/jobs/9466351902 ``` random seed: 320533040 generated error fail pass / total time test name [ ] 0 0 0 0 / 100...