Warning: some parameters could not be resolved for type ...

[JinraeKim]

I'm developing
and encountered a weird warning message.
I wonder if

1. it has a potential error that might cause data loss.
2. there is a way of resolving this warning.

Reproducing this error

Sorry for no MWE. It can be reproduced by running test/sim_and_save.jl and, at a new session, test/load_and_plot.jl in this commit.

brief code

• test/sim_and_save.jl

using FlightSims
using Plots
# using Pipe: @pipe
using JLD2, FileIO


function main()
    dir_log = "data/sim_and_save"
    mkpath(dir_log)
    env = TwoDimensionalNonlinearPolynomialEnv()
    x0 = State(env)()
    # simulation and data processing
    prob, sol = sim(env, x0, apply_inputs(dynamics!(env); u=FlightSims.optimal_input(env)); tf=10.0)
    df = process(env)(prob, sol)
    # df = @pipe sim(env; tf=10.0) |> process(env)(_...)  # shorthand
    # save (for stable data saving, one may save raw data). Here, raw data means e.g. t = 0, 0.01, 0.02, ...
    FlightSims.save(joinpath(dir_log, "test.jld2"), env, prob, sol)
    # to load it, see `main/example_load.jl`
end

• test/load_and_plot.jl

using JLD2
using FlightSims, OrdinaryDiffEq, ComponentArrays  # for data re-construction
using UnPack

using Plots


function _load()
    # run it in a new session. Be careful for loading packages for re-construction.
    saved_data = load("test.jld2")
    # @unpack env, prob, sol = saved_data
    # df = process(env)(prob, sol)
    # plot(df.times, hcat(df.states...)')
    nothing
end

Warning message

julia> _load()
┌ Warning: some parameters could not be resolved for type ODEFunction{true,FlightSims.#5#6{TwoDimensionalNonlinearPolynomialEnv},LinearAlgebra.UniformScaling{Bool},Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,Nothing,typeof(SciMLBase.DEFAULT_OBSERVED),Nothing}; reconstructing
└ @ JLD2 ~/.julia/packages/JLD2/fAkkN/src/data/reconstructing_datatypes.jl:440

[JonasIsensee]

Hi @JinraeKim ,

if I understand correctly, you are trying to store and load structures from OrdinaryDiffEq which is problematic.

JLD2 (currently) does not support serializing (anonymous) functions because this is generally a difficult thing to do.
I've tried working on that but have so far failed. I have asked for help but with little success
See e.g. https://discourse.julialang.org/t/serializing-functions-methods-closures/57061

If you really want to store the full function / problem / solution structures, you could consider using the stdlib
Serialization. Note that the resulting files are generally not portable, though.

Alternatively, you could consider extracting the relevant information from these structures and store those using JLD2.

[JinraeKim]

@JonasIsensee
Thank you for your explanation. Yes, you understood correctly.
How's the outlook of JLD2? Would it support serialising anonymous functions in the near future?

[JonasIsensee]

Hey @JinraeKim ,
I spent my weekend working on this and seem to have finally made progress.
Have a look at #316