Vincent Laviron

> As an aside, I’m confused as to why the CI failed. Why did the compiler suddenly start issuing a bunch of warnings that it didn’t issue before? The `configure`...

> 2\. For variants it is awkward because `[@reuse tm]` only makes sense if `tm` also starts with the `Fmapp` constructor, and probably don't want a constructor cehck at runtime....

> @lthls, have you looked at whether you could (easily) find other instances of this issue in matching.ml? I found two suspicious uses of `List.fold_right2`, but one binds with kind...

> What do we gain in return? Consistency. I think it is valuable to be able to guarantee that all the following pieces of code evaluate in the same order:...

I don't remember a policy of closing PRs without consensus, so I'll take the liberty of re-opening it. Note that without this PR, `Sys.opaque_identity` is "broken" because it doesn't preserve...

> I'm not sure if the compiler will properly optimize `{ data : (Int32.t, Value.int32) Hashtbl.t option ref`. It could remove one layer of indirection by "merging" the `option ref`...

> do you think it could be possible to implement an efficient "nullable ref" type as @¢hambart did in [nullable array](https://github.com/chambart/ocaml-nullable-array)? Or is there something fundamentally different between references and...

I can reproduce the problem with the setup above, on trunk. However `gdb` doesn't work in this setup so I'm a bit stuck. I did manage to find out that...

With some additional printf debugging I managed to pinpoint to this line: https://github.com/ocaml/ocaml/blob/1d7d91556fd3a59a086bfd73b7a4c75cfb0ccc7f/otherlibs/runtime_events/runtime_events_consumer.c#L395

A wild guess: is it possible that 0x424da040 comes from `mmap`ed memory, and the processor can't perform atomic accesses on such memory ?