Wrong flat Modelica output of records with start attributes

[casella]

Steps to Reproduce

Consider the following MWE:

model M
  record R
    Real x;
    Real y;
  end R;
  
  R r1(x(start = 1), y(start = 2));
  R r2(x(start = 3), y(start = 4));
equation
  r1 = R(0,0);
  r2 = R(0,0);
end M;

When I instantiate it with the default options, I get

function M.R "Automatically generated record constructor for M.R"
  input Real x;
  input Real y;
  output R res;
end M.R;

class M
  Real r1.x(start = 1.0);
  Real r1.y(start = 2.0);
  Real r2.x(start = 3.0);
  Real r2.y(start = 4.0);
equation
  r1 = M.R(0.0, 0.0);
  r2 = M.R(0.0, 0.0);
end M;

However, if I turn on the -f translation flag, I get

model 'M'
record 'R'
  Real 'x'(start = 1.0);
  Real 'y'(start = 2.0);
end 'R';

  'R' 'r1';
  'R' 'r2';
equation
  'r1' = 'M.R'(0.0, 0.0);
  'r2' = 'M.R'(0.0, 0.0);
end 'M';

This is wrong for two reasons:

1. The start attribute on the two record instances 'r1' and 'r2' were lost in the declarationss
2. Some of those start attributes are incorrectly assigned to the record type itself.

@perost I hope this is a quick fix

Keeping @mscuttari in the loop

[perost]

#11812 fixes the second issue, that the record declaration itself is wrong. The issue was that the frontend does not have unmodified instances of records, only record instances with modifiers applied (e.g. r1 and r2) and record constructors. With the fix the frontend now reinstantiates each record type we want to dump without modifiers, similar to how we create record constructors.

Regarding the first issue the problem is that the variables in the flat model do not have modifiers, only binding equations and type attributes. But it should be possible to reconstruct the modifiers from the instance tree when dumping record components.

[casella]

#11812 fixes the second issue, that the record declaration itself is wrong. The issue was that the frontend does not have unmodified instances of records, only record instances with modifiers applied (e.g. r1 and r2) and record constructors. With the fix the frontend now reinstantiates each record type we want to dump without modifiers, similar to how we create record constructors.

Good.

Regarding the first issue the problem is that the variables in the flat model do not have modifiers, only binding equations and type attributes.

I see. So, if I flatten

model M3
  Real x(start = 3);
equation
  der(x) = 1;
end M3;

I get

class M3
  Real x(start = 3.0);
equation
  der(x) = 1.0;
end M3;

but then start = 3.0 is a type attribute of Real?

But it should be possible to reconstruct the modifiers from the instance tree when dumping record components.

I guess that is necessary, if we want to support records in BaseModelica output

[perost]

I see. So, if I flatten

model M3
  Real x(start = 3);
equation
  der(x) = 1;
end M3;

I get

class M3
  Real x(start = 3.0);
equation
  der(x) = 1.0;
end M3;

but then start = 3.0 is a type attribute of Real?

Yes, each variable has a flat list of type attributes (just a list of name/binding tuples). There's never been any need to be able to represent nested binding equations before since those are normally resolved during flattening.

[casella]

OK, now after #11812 I get

package 'M'
  record 'R'
    Real 'x';
    Real 'y';
  end 'R';

  model 'M'
    'R' 'r1';
    'R' 'r2';
  equation
    'r1' = 'M.R'(0.0, 0.0);
    'r2' = 'M.R'(0.0, 0.0);
  end 'M';
end 'M';

@perost, getting the record member modifiers is essential if we want to get power grids models to run in MARCO. Do you think you can fix this before the summer break? Thanks!

[perost]

#12453 improves the handling of records in Base Modelica, the output is now:

package 'M'
  record 'R'
    Real 'x';
    Real 'y';
  end 'R';

  model 'M'
    'R' 'r1'('x'(start = 1.0), 'y'(start = 2.0));
    'R' 'r2'('x'(start = 3.0), 'y'(start = 4.0));
  equation
    'r1' = 'M.R'(0.0, 0.0);
    'r2' = 'M.R'(0.0, 0.0);
  end 'M';
end 'M';

There's still an issue here, M.R in the equation is wrong since the record is just called R, but that's probably not hard to fix.

However, I'm not sure if we actually want this fix, because it changes the flattening to do the same as for the new backend and not split record instances instead of having to reconstruct them when dumping Base Modelica. I realized after implementing the fix that the complicated and error-prone way we previously did it was to try to avoid having the --baseModelica flag influence the flat model that we pass to the backend.

With this fix that's no longer the case, since using --baseModelica will now keep record instances in the flat model which for example the old backend can't handle, so dumping Base Modelica and also compiling the model in the same pass doesn't work. That could be a big issue if we have any plans on making Base Modelica the default output.

[perost]

#12453 has now been replaced by #12454, which achieves the same result but without changing the flattening behaviour. It turned out that the current code only needed some very slight modification to work with the new code I wrote to dump record modifiers. The reconstruction of record instances still has issues, like probably not working with nested records, but that's a different issue.

[perost]

#12455 fixes the issue I mentioned earlier that record expressions use the wrong name for the record type.

[casella]

The reconstruction of record instances still has issues, like probably not working with nested records, but that's a different issue.

That is not really necessary for the time being. We just need to be able to handle Complex numbers.

[casella]

I checked with the latest nightly, All fine, thanks @perost!