Will Tebbutt

Yes, this is definitely a Tapir.jl limitation -- we're missing a rule for `Core.Intrinsics.atomic_pointerset`. I need to open an issue about this, and improve some of the error messages in...

@yebai @mhauru is it obvious to you whether Tapir.jl will re-derive its rule with a fresh `Tapir.TapirContext()` at each iteration of e.g. `Gibbs(HMC(0.2, 3, :p; adtype=adbackend), PG(10, :x))`? I'm seeing...

Also, @mhauru I've narrowed down the source of the error above involving `Core.Intrinsics.atomic_pointerset` to the really quite nasty looking `Base._unsetindex!`. Going to try and get a fix out for this...

Yup, that'll do it.

Note, if this is an issue for Tapir.jl, is it also potentially a problem for compiled ReverseDiff.jl?

v0.2.31 of Tapir.jl should fix the problem around `Core.Intrinsics.atomic_pointerset`.

I'm aware that this can happen (I've seen this kind of thing sporadically over the last year), but I don't properly understand why it happens. Re-running CI in order to...

Ahhh, yes, I believe that's exactly what's going on. I can build new unit tests for Tapir.jl from this, and extend whatever functions aren't working properly. Thanks for the pointer...

A version of Tapir.jl which addresses this particular issue should be available in an hour or so.

FWIW I'm also very happy to add an additional integration test suite to Tapir which runs Tapir on examples in Bijectors.jl, particularly if you have a handy list of functions...