Inconsistent handling of $cse expressions in models involving interpolate() and state variable change

[casella]

We have a fairy involved model that generates invalid C code due to a missing definition of a $cse variable which is then used in an expression.

We were not able to exactly reproduce the issue, but we managed to reproduce one which seems to be closely related and should be fixed anyway. This is the MWE:

model TestJac
  Real P(stateSelect = StateSelect.prefer), T(stateSelect = StateSelect.prefer);
  Real M(stateSelect = StateSelect.avoid), E(stateSelect = StateSelect.avoid);
  Real rho;
  Real V = 2 + sin(time);
  Real X[2];
  Real MMmix;
  Real c = Modelica.Math.Vectors.interpolate({1, 2, 3}, {1, 3, 8}, T, 1);
  constant Real MM[2] = {1, 2};
  parameter Real R = 1;
equation
  der(M) = 0;
  der(E) = 0;
  der(M*X) = {0, 0};
  M = rho*V;
  E = M*c*T;
  P/rho = R/MMmix*T;
  MMmix = 1/(X[1]/MM[1] + X[2]/MM[2]); 
end TestJac;

If we compile this model and then analyze it with the declarative debugger, we see some weird things happening:

• the interpolate function is called directly in the torn part ((c, _) := Modelica.Math.Vectors.interpolate({1.0, 2.0, 3.0}, {1.0, 3.0, 8.0}, T, 1);), instead of being first called and its result stored in a $cse variable, as it should happen thanks to the WrapFunctionCalls method.
• $cse1 appears in the variables window, it is used to compute some piece of jacobian, but is never actually computed by solving any equation. This is weird, you cannot compute stuff later on using $cse1 if it hasn't been set to some value previously.
• $cse2 shows up in a torn equation and in a jacobian equation, but is also apparently never computed
• $cse3 looks fine

What we observe is that cse3 involves differentiating a built-in function sin(time), which is not a big deal, while cse1 and cse2 involve the interpolate function, which is a true algorithm that require extra steps for differentiation. Maybe that holds the key for the solution of this issue.

[casella]

Keeping @AndreaBartolini in the loop

[casella]

@phannebohm, @kabdelhak, I tried this with the NB and I got this:

[1] 17:46:54 Translation Error
Internal error NBSlice.fillDependencyArray failed because number of flattened indices 1 for dependency $DER.M could not be divided by the body size 2 without rest.

[2] 17:46:55 Translation Error
Internal error NBAdjacency.Matrix.createPseudo failed for:
[ARRY] (2) $DER.M * X .+ M * $DER.X = {0.0, 0.0} ($RES_SIM_4)

[3] 17:46:55 Translation Error
Internal error NBAdjacency.Matrix.create failed to create adjacency matrix for system:
System Variables (10/11)
**************************
(1)       [ALGB] (1) Real T (StateSelect = prefer)
(2)       [ALGB] (1) Real rho
(3)       [ALGB] (1) Real MMmix
(4)       [DER-] (1) Real $DER.M
(5)       [ALGB] (1) Real P (StateSelect = prefer)
(6)       [DER-] (2) Real[2] $DER.X
(7)       [DISC] (1) Integer $FUN_2
(8)       [DER-] (1) Real $DER.E
(9)       [ALGB] (1) Real V = 2.0 + sin(time)
(10)      [ALGB] (1) Real c = Modelica.Math.Vectors.interpolate({1.0, 2.0, 3.0}, {1.0, 3.0, 8.0}, T, 1)[1] 
System Equations (9/11)
*************************
(1)       [SCAL] (1) MMmix = 1/(X[1] + 0.5 * X[2]) ($RES_SIM_0)
(2)       [SCAL] (1) P / rho = (R / MMmix) * T ($RES_SIM_1)
(3)       [SCAL] (1) E = c * M * T ($RES_SIM_2)
(4)       [TUPL] (2) (c, $FUN_2) = Modelica.Math.Vectors.interpolate({1.0, 2.0, 3.0}, {1.0, 3.0, 8.0}, T, 1) ($RES_$AUX_10)
(5)       [SCAL] (1) M = rho * V ($RES_SIM_3)
(6)       [ARRY] (2) $DER.M * X .+ M * $DER.X = {0.0, 0.0} ($RES_SIM_4)
(7)       [SCAL] (1) $DER.E = 0.0 ($RES_SIM_5)
(8)       [SCAL] (1) -(2.0 - V) = sin(time) ($RES_$AUX_9)
(9)       [SCAL] (1) $DER.M = 0.0 ($RES_SIM_6) 

so the NB is not there, either. I would be grateful if you could have a look at this issue with the OB, this is a blocker for us on an important project. Looks like some weird bug in WrapFunctionCalls and/or causalization.

Thanks!