Jan Midtgaard

Wow, I was just hacking on statistics myself in order to improve the `Out_channel` situation... Here's a branch: https://github.com/ocaml-multicore/multicoretests/tree/generic-stats I considered adding bindings to `Util` as we discussed, but instead...

I also observed this a month back on a CI run after rebasing #304. The symptoms were the same: s390x running trunk/5.2 (bytecode back too then I suspect), also failing...

With a repetition count of 20: https://github.com/ocaml-multicore/multicoretests/tree/thread-createtree-focus I was able to trigger this on 2/20 test runs: https://github.com/ocaml-multicore/multicoretests/runs/14952327534 It is puzzling that it is triggered on s390x and 5.2/trunk only...

Observed this again on s390x 5.2 when merging #376 into main: https://ocaml-multicoretests.ci.dev:8100/job/2023-07-14/221157-ci-ocluster-build-52b7b3 ``` random seed: 362338500 generated error fail pass / total time test name [ ] 0 0 0...

I've given these tests a spin on 5.1 and trunk with this branch and cannot reproduce with a `repeats` count of 50: https://github.com/ocaml-multicore/multicoretests/tree/shym-base-buffer This indicates that the issue is indeed...

Closing this for now, as we abandoned the idea of testing `Base`. Thanks for the PR and for confirming that this could be useful too! :smiley:

This is both a nice improvement and elegantly implemented - many thanks! :smiley: - Could you add a test case to `test/` to guard against someone (read: me) breaking this...

The `Random` module changed between OCaml 4 and 5, and as a result the output distribution changed. Overall that means we cannot expect the same output with the same seed...

Cool! I was thinking about this yesterday too :+1: Logs is a nice and popular library for this kind of logging: https://erratique.ch/software/logs (sans dependencies). AFAIU, it works through callbacks to...

Rebased on latest `main`. Latest changes comment out the unsafe `Bytes.escaped` and include increasing the chance of producing counterexamples to sequential consistency. The PR is still missing a way to...