Sundials.jl does not work with ArrayPartition from RecursiveArrayTools.jl

[mauro3]

This errors:

using OrdinaryDiffEq, RecursiveArrayTools

function problem!(du, u, p, t)
    h, ϕ = u.x
    dh, dϕ = du.x
    dh .= 0.5 * h
    dϕ .= 0.1 * ϕ
end

h0 = rand(3, 4)
ϕ0 = rand(3, 4)
u0 = ArrayPartition(h0, ϕ0)

tspan = (0.0, 1e3)
prob = ODEProblem(problem!, u0, tspan)
sol = solve(prob, Tsit5()) # works
sol = solve(prob, CVODE_BDF())

with

ERROR: MethodError: Cannot `convert` an object of type Vector{Float64} to an object of type ArrayPartition{Float64, Tuple{Matrix{Float64}, Matrix{Float64}}}
Closest candidates are:
...

Feel free to close if this is not going to be supported.

[ChrisRackauckas]

Someone could do a kind of "reinterpret to a vector" kind of thing to get it to work in Sundials. Basically have a uvec created for internal to the binary and keep that separate from the one in the f and just u .= uvec if it's not a type that is compatible with the binary. It would have a bit of overhead like that but it's possible. I don't think it's a priority though, especially with the waning use cases where Sundials is actually faster, so I'll keep it open but don't plan to do it.