fix(nonlinearsystem): Fix codegen issue for vector parameters

[SebastianM-C]

Co-authored-by: Aayush Sabharwal [email protected]

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

Add any other context about the problem here.

[ChrisRackauckas]

test? What does this solve?

[SebastianM-C]

This solves a codegen issue for SCCNonlinearProblems, where vector parameters show up in registered functions.
The problem is that with DestructuredArgs, we get codegen that just unpacks the elements of the vector parameter

__mtk_arg_2 = ___mtkparameters___[1]
var"vol1₊rho_0[1]" = __mtk_arg_2[1]
var"vol1₊rho_0[2]" = __mtk_arg_2[2]
var"vol2₊rho_0[1]" = __mtk_arg_2[3]
var"vol2₊rho_0[2]" = __mtk_arg_2[4]
...

but the entire vector variable (e.g. vol1₊rho_0) is not available, so if there is something that needs the entire vector, like a registered function, then we get an undefined variable error.

As for the test, I've been trying to get one based on the existing

ModelingToolkit.jl/test/scc_nonlinear_problem.jl

Lines 165 to 255 in 0a5c1ce

	import ModelingToolkitStandardLibrary.Blocks as B
	import ModelingToolkitStandardLibrary.Mechanical.Translational as T
	import ModelingToolkitStandardLibrary.Hydraulic.IsothermalCompressible as IC
	@testset "Caching of subexpressions of different types" begin
	liquid_pressure(rho, rho_0, bulk) = (rho / rho_0 - 1) * bulk
	gas_pressure(rho, rho_0, p_gas, rho_gas) = rho * ((0 - p_gas) / (rho_0 - rho_gas))
	full_pressure(rho, rho_0, bulk, p_gas, rho_gas) = ifelse(
	rho >= rho_0, liquid_pressure(rho, rho_0, bulk),
	gas_pressure(rho, rho_0, p_gas, rho_gas))
	@component function Volume(;
	#parameters
	area,
	direction = +1,
	x_int,
	name)
	pars = @parameters begin
	area = area
	x_int = x_int
	rho_0 = 1000
	bulk = 1e9
	p_gas = -1000
	rho_gas = 1
	end
	vars = @variables begin
	x(t) = x_int
	dx(t), [guess = 0]
	p(t), [guess = 0]
	f(t), [guess = 0]
	rho(t), [guess = 0]
	m(t), [guess = 0]
	dm(t), [guess = 0]
	end
	systems = @named begin
	port = IC.HydraulicPort()
	flange = T.MechanicalPort()
	end
	eqs = [
	# connectors
	port.p ~ p
	port.dm ~ dm
	flange.v * direction ~ dx
	flange.f * direction ~ -f
	# differentials
	D(x) ~ dx
	D(m) ~ dm
	# physics
	p ~ full_pressure(rho, rho_0, bulk, p_gas, rho_gas)
	f ~ p * area
	m ~ rho * x * area]
	return ODESystem(eqs, t, vars, pars; name, systems)
	end
	systems = @named begin
	fluid = IC.HydraulicFluid(; bulk_modulus = 1e9)
	src1 = IC.Pressure(;)
	src2 = IC.Pressure(;)
	vol1 = Volume(; area = 0.01, direction = +1, x_int = 0.1)
	vol2 = Volume(; area = 0.01, direction = +1, x_int = 0.1)
	mass = T.Mass(; m = 10)
	sin1 = B.Sine(; frequency = 0.5, amplitude = +0.5e5, offset = 10e5)
	sin2 = B.Sine(; frequency = 0.5, amplitude = -0.5e5, offset = 10e5)
	end
	eqs = [connect(fluid, src1.port)
	connect(fluid, src2.port)
	connect(src1.port, vol1.port)
	connect(src2.port, vol2.port)
	connect(vol1.flange, mass.flange, vol2.flange)
	connect(src1.p, sin1.output)
	connect(src2.p, sin2.output)]
	initialization_eqs = [mass.s ~ 0.0
	mass.v ~ 0.0]
	@mtkbuild sys = ODESystem(eqs, t, [], []; systems, initialization_eqs)
	prob = ODEProblem(sys, [], (0, 5))
	sol = solve(prob)
	@test SciMLBase.successful_retcode(sol)
	end

and make rho_0 a 1 element vector and register the full_pressure or liquid_pressure/gas_pressure functions, but I'm running into other issues (differentiation wrt array expression).

[SebastianM-C]

Rebased on master and added test.

[ChrisRackauckas]

Looks like it's failing?

[SebastianM-C]

The failures seem to be due to aliasing, I think we can re-trigger

[SebastianM-C]

Looks like the fix was already merged, this just adds the test

[AayushSabharwal]

I "accidentally fixed" this in #3390, so this PR adds the test

[ChrisRackauckas]

Test fails though? Needs to rebase?