Simon Cruanes

As far as I understand, `SHOULD` is not `MUST`. It's better to produce json objects without duplicates, but I don't think that it's invalid.

Ah, that's fair. If you want to go a bit further in bringing sanity to JSON, and you want reasonable APIs, I'd say that objects could use `Map.Make(String)` :grin: ....

Indeed, and I'm surprised now that we even have a special case for "empty" on the wire. The protobuf semantics is that unexpected fields should just be ignored, not that...

I'm not exactly sure, I suspect it's orthogonal. I also have an [implementation](https://github.com/c-cube/ocaml-trace/blob/3c2f804716fed857b546f443cd6b6e2169acb963/src/core/meta_map.ourown.ml) that stores more stuff in the key, _I think_. You can use as much code from there...

Does this mean you can only have one key per type? I'm a bit confused...

That's unfortunate (and a bit weird, afaict literally all calls to `TLS.get_exn` are properly guarded!). Is there a way to get some repro/regression test? :-) Also: try to `opam pin...

I don't know how to run this static initializer _from the main domain_ in a lazy way, alas. If I make it into a some sort of lazy block it...

Blocking signals seems to work for me, can you confirm @amongonz ?

I'm not aware of any signal handling in moonpool itself (besides the new one). Picos does have some signal handling though. I have absolutely no idea where this is coming...

I'm killing moonpool_io, so closing this, sorry.