Documentation building bug (related to Measurements.jl)

[ArseniyKholod]

While building Documentation 16: Differentiable programming fails.
In part Propagating errors using Measurements.jl we are using following code:

semi = SemidiscretizationHyperbolic(mesh, equations, initial_condition_convergence_test,
                                    solver, uEltype=Measurement{Float64})

And finally LoadError appears (building summary)

Differentiable programming: Error During Test at /home/runner/work/Trixi.jl/Trixi.jl/docs/literate/make.jl:59
Got exception outside of a @test
LoadError: MethodError: no method matching Float64(::Measurement{Float64})

I looked in History of Measurements.jl. They released new version 2.9.0 last week with following:

• New methods value(::Missing) and uncertainty(::Missing) (#127, #128).

• one(::Measurement{T}) now returns one(T), instead of an instance of Measurement{T} (#134, #135). This change should not breaking (and hence the major version was not increased), unless you were relying on the previous incorrect return type of one(::Measurement{T}), which did not follow the precise semantic of the one function.

Then I downgraded version of Measurements.jl to previous one - 2.8.0. And after that error disappeared. I think that Trixi or related packages use function one(uEltype) (explicit or implicit) somewhere.

Like a temporal solution I added following code to 16: Differentiable programming in my PR.

using Pkg #hide #md
Pkg.add(name="Measurements", version="2.8.0") #hide #md

Now Documentation check doesn't fail. But it doesn't seem to be a good solution.

[ArseniyKholod]

According to error's stacktrace (building summary) problem seems to be in OrdinaryDiffEq.jl or
DiffEqBase.jl. Because error appears after calling

sol = solve(ode, BS3(), save_everystep=false);

in Propagating errors using Measurements.jl

[ranocha]

Since the stacktrace you linked above comes from calculate_residuals! (see https://github.com/trixi-framework/Trixi.jl/actions/runs/4545729357/jobs/8013456987#step:6:1116), it looks to me like it's a problem in OrdinaryDiffEq.jl. You you please very that by running a simple ODE with OrdinaryDiffEq.jl and Measurements.jl? If that works with the old version of Measurements.jl (it should) but not the new one, we have tracked the problem better. Then, you could open an issue there.

[ArseniyKholod]

I tested following simple ODE with OrdinaryDiffEq.jl and Measurements.jl:

using OrdinaryDiffEq, Measurements, Plots
u₀ = 1.0  ± 0.1
tspan = (0.0, 1.0)
radioactivedecay(u, p, t) = -5.730 * u
prob = ODEProblem(radioactivedecay, u₀, tspan)
sol = solve(prob, Tsit5())
plot(sol)

With old (2.8.0) and new (2.9.0) version of Measurements it works exactly in the same way

[ArseniyKholod]

Remark to previous comment. Problem actually appears in OrdinaryDiffEq. But was invisible in easiest example with only scalar variable.
I tried ODE's examples with vector-variable u, in this case vector of value and it's derivative:

using OrdinaryDiffEq, Plots, Measurements
const g = 9.81±0.1
L = 1.0±0.1
u₀ = [0±0.0, π / 2±0.1]
tspan = (0.0, 6.3)
function simplependulum(du, u, p, t)
    θ = u[1]
    dθ = u[2]
    du[1] = dθ
    du[2] = -(g / L) * sin(θ)
end
prob = ODEProblem(simplependulum, u₀, tspan)
sol = solve(prob, Tsit5())

Error appears only with latest version [email protected]
ERROR: LoadError: MethodError: no method matching Float64(::Measurement{Float64})
OrdinaryDiffEq might incorrect handle Vector{Measurement{Float64}}.
Stacktrace also comes from calculate_residuals!.

[sloede]

Thanks for the investigation! With this, it might make sense to open an issue with the OrdinaryDiffEq package. If you do, please make sure to

1. describe the problem (as you did here),
2. include the MWE code,
3. include the full error message (including the stacktrace!), and
4. link to this issue here for reference.

It would also be great if you could tag Hendrik and me by adding something like "cc @ranocha @sloede" in your issue, such that we are aware of it.

[ranocha]

I can't reproduce your example. Please tell us the version numbers of everything involved (Julia, Measurements, OrdinaryDiffEq) and the full stacktrace.

[ArseniyKholod]

Thank you for response!
I updated all packages, error doesn't appear any more.
20 hours ago problem was solved in RecursiveArrayTools.jl. Because of the same issue created in
DifferentialEquations.jl.
See PR Handle measurement units #260.
Trixi documentation now builds successful according to your commit(https://github.com/trixi-framework/Trixi.jl/actions/runs/4582247950/jobs/8092347449)