Sébastien Doeraene

Thanks. Yes, it's most certainly a race condition. We'll have to investigate, as I mentioned. (this was Linux, Temurin 17; other details are probably irrelevant)

We can make the new registration API not use any Scala type. But it won't change the fact that the public API exposes `Option`s, `List`s and `Seq`s (through varargs). We...

#5183 is a variant that relies on a Java-only API in the `javalibintf`.

The `Reflect` API definitely is one of the not-so-great things in the overall design. IMO it has always been a "necessary evil", which we need for testing frameworks to be...

I agree. Indeed, that's a good analysis. The opt-in via subtype across libraries is crucial here.

Coming back to this to try and move it forward. Given that (it seems) we agree that centralization is desirable, it seems either this PR or #5183 would be appropriate...

Yes, that's a long-standing issue with `NaN`s, that is basically unfixable. NaN's basically have unexpected behavior all around. It's a major design issue for any language with NaN's and containers....

This is now ready for review. Also review by @tanishiking. Note that the first commit would become much smaller if this PR gets rebased on top of #4998. But conceptually...

I can split into two PRs. It's a bit annoying because then the second commit cannot remove *all* of the weird helpers (`jsArrayPush`, `jsNew`) as they're still used by the...

@gzm0 This PR now only contains the second (most important) commit. I extracted the less-interesting commit into #5030.