Problem with code generation involving records in Buildings

[Francesco Casella]

Please check ​Buildings.Fluid.Geothermal.Borefields.BaseClasses.HeatTransfer.Validation.Analytic_20Years. The following error is issued during C compilation

Buildings_7.0.0_Buildings.Fluid.Geothermal.Borefields.BaseClasses.HeatTransfer.Validation.Analytic_20Years.c:2442:137: error: passing 'integer_array' (aka 'struct base_array_s') to parameter of incompatible type 'int'
    data->localData[0]->realVars[243] /* groTemRes.delTBor0 DISCRETE */ = mul_real_scalar_product((*real_array_element_addr1(&tmp10, 1, tmp12)), (*real_array_element_addr1(&tmp13, 1, tmp15)));
                                                                                                                                        ^~~~~

[Mahder Alemseged Gebremedhin]

My guess is that these are most probably NF problems.

I will try to localize the issues and we can pass them to Per. He usually fixes them quite easily.

[Francesco Casella]

The only term I can recognize in the broken C code is groTemRes.delTBor0, I guess this could help pinpointing what goes wrong with the NF. Unfortunately the OF breaks ​very early with this model, so we don't have an easy comparison term.

@mahge930, if you manage to do this in your spare time, it would be much appreciated :)

[Per Östlund]

Here's an example that gives the same error:

function f
  input Integer sz;
  input Real x[sz];
  input Real y[sz];
  input Integer i;
  output Real z;
algorithm
  z := x[1:i] * y[1:i];
  annotation(Inline = true); 
end f;

model M
  Real x[3] = {time, time, time};
  Real y[3] = {time, time, time};
  Real z;
  Integer i = integer(time);
algorithm
  when time > 0 then
    z := f(3, x, y, i);
  end when;
end M;

The error occurs with both the new and the old frontend, so I don't think it's a frontend issue. The issue seems to occur because the function is inlined, leading to an assignment that the backend can't scalarize and which the code generation doesn't seem to handle correctly.

[Francesco Casella]

The problem was fixed by mahge930's fixes reported ​here. These broke most PowerSystems models, but this was also fixed later, see ​report.

The model cited in the original description now fails at runtime because of a ​singular matrix, but this is most likely due to some other issue.

Thanks Mahder!