Silently fails with Sundials

[marius311]

It looks like if you use ComponentArrays with Sundials, it "succeeds" but the solution is wrong, eg here it doesn't move from the initial value even though the answer should be e.

julia> using DiffEqBase, Sundials, ComponentArrays

julia> prob = ODEProblem((u,p,t)->u, ComponentArray(x=ones(10)), (0.,1.));

julia> sol = solve(prob, CVODE_BDF(linear_solver=:BCG), reltol=1e-3);

julia> sol(1)
ComponentVector{Float64}(x = [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0])

This is Julia 1.5.2, ComponentArrays v0.8.7, and Sundials v4.2.6.

[jonniedie]

I don’t have access to a computer at the moment and this is a tough one to debug in Codespaces on my iPad, so I’ll have to give it a look tomorrow. This one looks like it might take some digging.

[jonniedie]

Alright, here's a quick patch that will work until I figure out the right way to do this:

Base.convert(::Type{Sundials.NVector}, x::ComponentArray) = Sundials.NVector(getdata(x))

[jonniedie]

Figured it out. Was missing a method to explicitly convert to Array. Which is kinda weird because it was doing the conversion to NVector directly correctly even without either convert method defined.