The NF does not evaluate expressions with Integer indeces

[Francesco Casella]

Please check the ​ScalableTestSuite.Power.ConceptualPowerSystem.Verification.TwoGeneratorsStepLoad model. It currently fails during C code compilation.

Flattening the model with the NF leads to equations such as

generator[2].T_s[2] = 
  (generator[2].T_s[2 - 1] + generator[2].NTU / 4.0 * generator[2].T_source) /
  (1.0 + generator[2].NTU / 4.0);

and

  if 1 == 1 then
    P_diss[1,1] = 0.0;
    P_ex[1,1] = 0.0;
  else
    P_diss[1,1] = P_d * (generator[1].omega - generator[1].omega);
    P_ex[1,1] = P_f / /*Real*/(abs(1 - 1)) * sin(generator[1].theta - generator[1].theta);
  end if;

In the first case, the indeces of the array elements should be evaluated to literal constants (as with the old FE).

In the second case, the condition should be evaluated and only the first branch should be included in the flat code.

[Francesco Casella]

Something similar happens in ​Modelica.Blocks.Examples.PID_Controller. The flattened code contains the follwing lines

protected parameter Real kinematicPTP.p_deltaq[1] = 
  (if 1 == 1 then {1.0} * kinematicPTP.deltaq[1] else kinematicPTP.deltaq)[1];

kinematicPTP.sd_max = 1.0 / max(array(abs(kinematicPTP.aux1[$i1]) for $i1 in 1:1));

instead of

protected parameter Real kinematicPTP.p_deltaq[1] = kinematicPTP.deltaq[1];

kinematicPTP.sd_max = 1.0 / abs(kinematicPTP.aux1[1]);

[Martin Sjölund]

This does not seem to be the current error. Instead it is that scalar*array and array*scalar is translated to scalar*scalar. Note also that the old FE does not have scalar*array (it swaps the arguments to only implement one operator).

[Martin Sjölund]

​https://github.com/OpenModelica/OMCompiler/pull/2350

[Francesco Casella]

After ​PR #2350, the model now ​simulates