JuliaSimCompiler doesn't work for equation system that works fine with default compiler

[ctessum]

Hello,

As suggested in #311, I am experimenting with using the JuliaSimCompiler to allow the use of simulations with larger point counts. I am working with a system similar to this example:

using EarthSciData, EarthSciMLBase, GasChem,
    DomainSets, ModelingToolkit, MethodOfLines,
    Dates, Latexify

@parameters t lev lon lat

start, finish = Dates.datetime2unix.([
    Dates.DateTime(2022, 5, 1),
    Dates.DateTime(2022, 5, 3),
])

domain = DomainInfo(
    partialderivatives_lonlat2xymeters,
    constIC(1.0f0, t ∈ Interval(start, finish)),
    zerogradBC(lon ∈ Interval(-130.0f0, -100.0f0))
)

# Skip unit enforcement for now
ModelingToolkit.check_units(eqs...) = nothing

geos = GEOSFP("0.25x0.3125_NA", t; coord_defaults = Dict(:lat => 34.0, :lev => 1))

model = SuperFast(t) + domain + Advection() + geos

render(latexify(equations(get_mtk(model))))

The equation system looks like this (I cropped the long ones):

Basically there is advection driven by external wind fields and and chemical reactions.

When I try to discretize it with the normal compiler, it works fine:

discretization = MOLFiniteDifference([lon => 6], t, approx_order=2)
@time prob = discretize(get_mtk(model), discretization) # great success

However, with the JuliaSimCompiler, it fails:

using JuliaSimCompiler
@time prob = discretize(get_mtk(model), discretization)

JuliaSimCompiler: generate_system
ERROR: ExtraVariablesSystemException: The system is unbalanced. There are 121 highest order derivative variables and 120 equations.
More variables than equations, here are the potential extra variable(s):
... elided ...
 superfast₊HO2[3]
 superfast₊ISOP[3]
 superfast₊NO[3]
 Dt(superfast₊O3[3], 1, false)
 superfast₊O3[4]
 superfast₊NO2[4]
 superfast₊ISOP[4]
 superfast₊HO2[4]
 superfast₊NO[4]
... elided ...
Stacktrace:
  [1] error_reporting(state::IRState, bad_idxs::Vector{Int64}, n_highest_vars::Int64, iseqs::Bool, orig_inputs::Tuple{})
    @ ModelingToolkit.StructuralTransformations ~/.julia/packages/ModelingToolkit/ZmLra/src/structural_transformation/utils.jl:50
  [2] check_consistency(state::IRState, orig_inputs::Tuple{})
    @ ModelingToolkit.StructuralTransformations ~/.julia/packages/ModelingToolkit/ZmLra/src/structural_transformation/utils.jl:85
  [3] structural_simplify!(state::IRState, io::Nothing; alias_eliminate::Bool, tearing::Bool, check_consistency::Bool, debug::Bool)
    @ JuliaSimCompiler ~/.julia/packages/JuliaSimCompiler/vX0Sq/src/structural_transformation/dae_transformations.jl:87
  [4] structural_simplify! (repeats 2 times)
    @ ~/.julia/packages/JuliaSimCompiler/vX0Sq/src/structural_transformation/dae_transformations.jl:75 [inlined]
  [5] #structural_simplify#84
    @ ~/.julia/packages/JuliaSimCompiler/vX0Sq/src/structural_transformation/dae_transformations.jl:69 [inlined]
  [6] structural_simplify(::IRSystem)
    @ JuliaSimCompiler ~/.julia/packages/JuliaSimCompiler/vX0Sq/src/structural_transformation/dae_transformations.jl:68
  [7] discretize(pdesys::PDESystem, discretization::MOLFiniteDifference{MethodOfLines.CenterAlignedGrid, MethodOfLines.ScalarizedDiscretization}; analytic::Nothing, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ MethodOfLinesJuliaSimCompilerExt ~/.julia/packages/MethodOfLines/YO1pd/ext/MethodOfLinesJuliaSimCompilerExt/MethodOfLinesJuliaSimCompilerExt.jl:62
  [8] discretize(pdesys::PDESystem, discretization::MOLFiniteDifference{MethodOfLines.CenterAlignedGrid, MethodOfLines.ScalarizedDiscretization})
    @ MethodOfLinesJuliaSimCompilerExt ~/.julia/packages/MethodOfLines/YO1pd/ext/MethodOfLinesJuliaSimCompilerExt/MethodOfLinesJuliaSimCompilerExt.jl:57
  [9] top-level scope
    @ ./timing.jl:273 [inlined]
 [10] top-level scope
    @ ./Untitled-1:0

In the error message above Dt(superfast₊O3[3], 1, false) sticks out, mostly because it's a derivative and not a variable. Let me know if you have any suggestions of how to fix this!

[ChrisRackauckas]

Is it 2D? A bug was found in 2D.

[ctessum]

1D:

MOLFiniteDifference([lon => 6], t, approx_order=2)

[YingboMa]

SuperFast isn't defined.

[ctessum]

Thanks for your response! It is defined here: https://github.com/EarthSciML/GasChem.jl/blob/main/src/SuperFast.jl#L33

Let me know if I misunderstood your comment, or if additional information would be useful.

[ctessum]

It should run out of the box if you install all the packages at the top. Let me know if it doesn't

[ctessum]

Hello again! As a follow-up to the above, I notice that https://github.com/SciML/MethodOfLines.jl/commit/186809073a576797944dad92e30cdf93450af699 deletes the extension with the explanation "move extension over". Where has the extension been moved to? Should it still be used the same way? Now, when I try to use JuliaSimCompiler with Julia 1.10 and MethodOfLines 0.10.2 it doesn't seem to actually do anything, i.e. the compile time and resulting system appear to be the same with and without JuliaSimCompiler installed. Thanks for any suggestions you can offer!

[ChrisRackauckas]

It was moved into JuliaSimCompiler itself. I don't think it ended up merging yet though? We have tests on a lot of other cases but the PDE case is a bit behind for various reasons, but repeated components cases are working already.