Andreas Rossberg

Sentinel values are bad, especially as input. And making the count artificially signed just for that is even worse. If we are keen on making the "all" case special then...

The signedness / sentinel value kludge is also my biggest complaint, which is why I suggested `wake_all`. But in fact, I don't think that should be necessary either. I'm confused...

@lars-t-hansen, what @conrad-watt said. In particular, the current design does *not* cover all cases in any practical sense, because it makes wake-all unusable. When you actually expect more than 2^32...

> So this means having another bit tag to distinguish unboxed variants from references. Ah, but only if you assume that variants may be unboxed. That was not the idea...

@RossTate: > This means casting from anyref by enumerating each of the cases would likely be impractical Before you continue asserting stuff, can you explain why an OCaml implementation would...

So let me double-check whether you agree that 1. in OCaml this is solely relevant for implementing a couple of (fairly rarely used) runtime primitives, and not for its actual...

I don't follow what you are arguing. That an OCaml port should be able to remove the field? That may save space, but may require cross-cutting changes to the compiler,...

> IIUC, @RossTate would prefer not to have to make this assumption I have a hard time believing that that's truly what @RossTate is suggesting. It hardly seems realistic to...

By “casts”, are you referring to the runtime checks that Java needs on array assignment? Do you imagine eliminating the checks by compiling this loop with code duplication, one copy...

I honestly have no idea. I did not close anything, GitHub just randomly closed some issues, not just this one. Maybe it somehow magically inferred something from the merge commit...