What I don't understand is why that change fixes anything (and/or why it was broken in the first place). equals works fine for j.l.Integer.

I've tentatively put this on a new 2.13.0 milestone, since it seems like something that could mergeable without a further RC. @adriaanm will make the final call

@NthPortal you need to pick to either use equals and hashCode and ignore collaborative equality/hashing (which is faster but some things are "weird"), or you use == and ## (and pay the performance cost). When you're inconsistent these issues crop up (subjectively, worst of both worlds). Notice the use of ##s in HashSet.

oh oops - I didn't notice that one was 1 and one was 1L

Let's ship this in 2.13.0. Thank you, @xuwei-k!