Another option which registers the new copied item in the stackdict before recursing to the children is

function deepcopy_internal(x::Array{T, N}, stackdict::IdDict) where {T, N}
    if haskey(stackdict, x)
        return stackdict[x]::typeof(x)
    end
    y = stackdict[x] = Array{T, N}(undef, ntuple(Returns(0), Val{N}()))
    setfield!(y, :ref, deepcopy_internal(x.ref, stackdict))
    setfield!(y, :size, x.size)
    y
end

I'm worried about calling setfield! on an array, though. cc. @oscardssmith for weather that is okay in this context and also because #56775 was introduced by https://github.com/JuliaLang/julia/pull/51319/files#diff-471c7f0385825eae442274e0debe4e88fe50ce6898508130c9c14fbb7c29f255R119-R124

If someone could point me towards an appropriate location for the test, that would be very helpful!

I'd put tests for this here:

Another option which registers the new copied item in the stackdict before recursing to the children is

I'm happy to go with whichever is preferred -- I'll wait for @oscardssmith to weigh in before modifying.

Calling setfield! on an array is fine (so long as you don't violate the invariant)

Cool. I'll change the fix.

Fix modified to following @LilithHafner 's suggestion, which I agree is probably cleaner. Test also fixed to actually be a test, rather than just evaluating a Bool.

Is this triggering a Revise.jl internal bug?

It looks like the @eval statement I had (accidentally) left in the code was causing some kind of problem. The latest commit appears to resolve.

It looks to me like the current test failures are unrelated to this PR, but someone who knows more about Julia's CI should probably take a look.