final or not final?

[d-hedberg]

In MyPackage.MyModel below, is the modification of a(x=4) legal (considering that the model A has a final prefix to it)?

package MyPackage
  model B
    parameter Real x;
  end B;

  final model A = B(x = 3); // Model A is made final here.

  model MyModel
    A a(x=4); // Legal?
  end MyModel;
end MyPackage;

[HansOlsson]

I would prefer it is legal, and think that it is easier to understand for replaceable.
The reason is that I would assume some people use 'final redeclare' to say that the class is choice, but still works as a normal class.

I would actually see three possibilities:

1. It's only the choice of class A that is final; this prevents some modifications that are a bit odd.
2. The entire class A is final, so even for 'final model A=B;' then MyModel would be illegal.
3. Only the modified elements "inherit" final; the example suggests this.

I prefer 1, but https://specification.modelica.org/master/inheritance-modification-and-redeclaration.html#final-element-modification-prevention indicates 2.

[henrikt-ma]

Alternative 2 corresponds to my intuition about final, which I suppose I got from the cited 7.2.6 Final Element Modification Prevention one way or another.

[d-hedberg]

I'd have to agree that alternative 2 is the only alternative that makes sense to me. I'd have a really hard time trying to convince Modelica users that if they declare a class final , parts of it can still be modified. That's confusing, and I'm pretty sure that the follow-up question would be: "How do I really make the class final then?".

[HansOlsson]

I'd have to agree that alternative 2 is the only alternative that makes sense to me. I'd have a really hard time trying to convince Modelica users that if they declare a class final , parts of it can still be modified. That's confusing, and I'm pretty sure that the follow-up question would be: "How do I really make the class final then?".

Agreed on second thought, and thus it seems this is just a question that has been answered and can be closed.
(If people just want to prohibit redeclare the solution is to use "redeclare" without final.)

[henrikt-ma]

Agreed on second thought, and thus it seems this is just a question that has been answered and can be closed.
(If people just want to prohibit redeclare the solution is to use "redeclare" without final.)

Perhaps we should still clarify that a final element doesn't have to be a component, and give an example of what it means for a type to be final?

[HansOlsson]

However, there are a number of modifiers for final classes (mostly/entirely final redeclare, of course) in MSL.

One example is in Modelica.Fluid.Examples.HeatExchanger.BaseClasses.BasicHX:

model BasicHX "Simple heat exchanger model"
 ...
Pipes.DynamicPipe pipe_1(
    redeclare final package Medium = Medium_1,
    ...
    redeclare final model HeatTransfer = HeatTransfer_1,
    ...
    redeclare final model FlowModel = FlowModel_1,

with Modelica.Fluid.Pipes.DynamicPipe:

model DynamicPipe "Dynamic pipe model with storage of mass and energy"
   ...
   HeatTransfer heatTransfer(
    redeclare final package Medium = Medium,
    final n=n,

(and similarly for the inherited FlowModel).

[d-hedberg]

Are you saying that the final prefix for a class if occurring in a modification should be interpreted differently? That it should be allowed to change a parameter inside a redeclared final class? See example below.

package MyPackage
  model B
  end B;

  model C
    parameter Real x;
  end C;

  replaceable model A = B constrainedby B;
end MyPackage;

model MyModel
  package P = MyPackage(redeclare final model A = MyPackage.C(x=3)); // Model A is made final here.
  P.A a(x=4); // Legal?
end MyModel;

[HansOlsson]

Are you saying that the final prefix for a class if occurring in a modification should be interpreted differently?

No, it was just an observation that some had that idea - and we need to correct MSL to be consistent with this.
I'll make a PR.

[HansOlsson]

Agreed on second thought, and thus it seems this is just a question that has been answered and can be closed.
(If people just want to prohibit redeclare the solution is to use "redeclare" without final.)

Perhaps we should still clarify that a final element doesn't have to be a component, and give an example of what it means for a type to be final?

Clarifying that and adding an example makes sense.