Failure to access discrete-time variable in solution

[baggepinnen]

julia> sol[sys.conversion.y]
ERROR: ArgumentError: Symbol conversion₊y(t) is not present in the system.

MWE:

using ModelingToolkit
using OrdinaryDiffEq
t = ModelingToolkit.t_nounits
k = ShiftIndex(t)


function Float2Bool2(; name)
    @variables begin
        u(t)::Float64
        y(t)::Bool
    end
    equations = [
        y ~ u != 0
    ]
    DiscreteSystem(equations, t; name)
end

function Model2(; name)
    @variables begin
        x(t) = 1
    end
    systems = @named begin
        conversion = Float2Bool2()
    end
    equations = [
        x(k) ~ 0.99x(k-1)
        conversion.u ~ x
    ]
    DiscreteSystem(equations, t; name, systems)
end

@mtkbuild sys = Model2()

prob = DiscreteProblem(sys, [], (0,10))

sol = solve(prob, FunctionMap())
sol[sys.conversion.y]

[AayushSabharwal]

This is known behavior. Due to how structural simplification of discrete systems is handled, observed variables are sihfted forward by one step. Fixing this requires proper compilation of synchronous systems.

julia> observed(sys)
2-element Vector{Equation}:
 Shift(t, 1)(conversion₊u(t)) ~ 0.99x(t)
 Shift(t, 1)(conversion₊y(t)) ~ Shift(t, 1)(conversion₊u(t)) != 0

julia> sol[sys.conversion.y(k+1)]
11-element BitVector:
 1
 1
 1
 1
 1
 1
 1
 1
 1
 1
 1

[AayushSabharwal]

We have had this behavior since MTKv9

[baggepinnen]

This is known behavior.

Do we have an existing issue for it? It may be known, but certainly not desirable so we should track it and eventually fix it

[AayushSabharwal]

I don't think we do. This can serve as the tracking issue, I guess.