Issue with parameter bindings involving Complex equations in the NF

[Francesco Casella]

Please consider the following test model

model TestComplexParameter
  import CM = Modelica.ComplexMath;
  parameter Real v_abs_start = 2, v_phase_start = 0.5;
  parameter Modelica.SIunits.ComplexVoltage v_start = CM.fromPolar(v_abs_start, v_phase_start);
  Modelica.SIunits.ComplexVoltage v(re(start=v_start.re), im(start=v_start.im));
equation
  CM.'abs'(v) = 1;
  CM.arg(v) = 0.5;
end TestComplexParameter;

When building the model with the NF, the generated TestComplex_06inz.c C-code contains:

/*
 equation index: 1
 type: ALGORITHM
 
   v_start := Complex(v_abs_start * cos(v_phase_start), v_abs_start * sin(v_phase_start));
 */
void TestComplex_eqFunction_1(DATA *data, threadData_t *threadData)
{
  TRACE_PUSH
  const int equationIndexes[2] = {1,1};
  Complex tmp0;
  Modelica_SIunits_ComplexVoltage tmp1;
  tmp0._re = (data->simulationInfo->realParameter[0]) * (cos(data->simulationInfo->realParameter[1]));
  tmp0._im = (data->simulationInfo->realParameter[0]) * (sin(data->simulationInfo->realParameter[1]));
  tmp1 = tmp0;
  data->simulationInfo->realParameter[3] = tmp1._re;
  data->simulationInfo->realParameter[2] = tmp1._im;
  TRACE_POP
}

which breaks the compiler because Modelica_SIunits_ComplexVoltage is undefined.

I understand binding equations for Complex parameters are now moved to the initial equations. For some reason, the call to the fromPolar function is not inlined correctly, as it is instead the case for regular binding equations, such as in

model TestComplexVariable
  import CM = Modelica.ComplexMath;
  Modelica.SIunits.ComplexVoltage v = CM.fromPolar(time, sin(time));
end TestComplexVariable;

Perhaps the Inline annotation of the fromPolar function is lost on these added initial equations?

[Francesco Casella]

Reporting @perost's e-mail comment on 12 July 2018

The model does indeed not work using the NF, and I'm afraid it's not trivial to fix (at least not in a way that won't cause a lot of other issues). Your analysis in #5009 that the function isn't inlined is not correct, the comment above the generated function shows that it has been inlined.

The NF has some pretty bad issues with functions and record constructors due to sharing too much between class instances, and it will take some time to unravel that particular knot.

[Francesco Casella]

It seems that the recent updates to the NF were beneficial, since this model now works correctly with v1.13.0-dev-1096-g6770e6300.