Testcase: Redeclarations: Order of Evaluation

[modelica-trac-importer]

Reported by choeger on 13 Aug 2014 09:07 UTC
Consider the following small model:

model Test2
  model C
    constant Real k = 1;
    constant Real p = 0;
  end C;

  model D
    constant Real k = 2;
    constant Real p = 2;
  end D;    

  model E
    replaceable model C2 = C;
    replaceable model C3 = C2(k = C2.p + 10);
      
    Real x;
    equation
     x = C3.k;   
  end E;

  model F = E(redeclare model C3 = D, redeclare model C2 = D);

  F e;
end Test2;

I wonder whether this is actually legal? C2 is certainly not transitively non-replacable, so can I have a "redeclaration chain" like that? Dymola accepts this. OpenModelica complains about a field not found.

Assuming it is legal, how about the modifications?

Basically, there are 3 ways to deal with the modification of C2:

1. evaluate before redeclaration of C3 takes effect (i.e. C3=D, x=2)
   This may also be the case when the modification is discarded.
2. evaluate after redeclaration of C3 takes effect, but before C2 (i.e. C3=C(k=C.p+2), x=10)
3. evaluate after redeclaration of C3 and after C2 (i.e. C3=D(k=D.p+10), x=12)

AFAIK, modifications are "merged" according to the spec, i.e. the modification inside E is never lost. When we also assume that redeclarations are always applied before modifications, the only legal answer should be 3.

However we decide, I propose to add this as a test case to the compliance suite.

---

Migrated-From: https://trac.modelica.org/Modelica/ticket/1552

[HansOlsson]

I agree that only option 3 makes sense, and also that C2 isn't transitively non-replaceable as it should be.

Will need to check if there is any overlap, but otherwise I think it makes sense to add this as two two-cases, one for the "non-replaceable" issue and one for the value (modified in some way to avoid the previous issue).