The NF does not expand the arguments of the product() function to scalars

[Francesco Casella]

Please check ​Modelica.Electrical.PowerConverters.Examples.ACDC.RectifierBridge2Pulse.HalfControlledBridge2Pulse.err.

When flattening with the old FE, this equation is obtained:

rootMeanSquareVoltage.product.y = 
  rootMeanSquareVoltage.product.u[1] * rootMeanSquareVoltage.product.u[2];

while the NF gives

  rootMeanSquareVoltage.product.y = product(rootMeanSquareVoltage.product.u);

which eventually the code generation chain cannot handle.

[Martin Sjölund]

I believe the old FE also hides other problems by pretending it can always expand product. For example the following just expands the product to 1.0:

model M
function f
  input Real r[:];
  output Real p;
algorithm
  p := product(r);
end f;
  Real r[2] = {time,time*2};
  Real p = f(r);
end M;

That error should be fixed along with the code generation. product(v for v in r) works fine in the old FE, but the new FE still converts this to product(array(...)) which is not supported in code generation.

[Martin Sjölund]

I thought to fix the old FE by translating into a reduction expression by calling vector() to flatten the array (​https://github.com/OpenModelica/OMCompiler/pull/2403). But it only works if at most 1 dimension is >1. sum and product work in almost the same way, but only sum is handled (mostly) correctly (in Static, Ceval, Simplify). If we had an operator to reshape a multi-dim array into a vector that was handled properly everywhere (including efficient code generation for iterators), that would be great.

[Per Östlund]

Fixed in 30b0d61 by adding code generation for product. The runtime functionality was actually already implemented, I guess someone had just forgot to add a case for product in the code generator.