Report error when modifying final parameter

[Henrik Tidefelt]

Model B below modifies a final parameter in A, but validates fine:

package FinalParam
  model A
    final parameter Integer n = 2;
    Real x[2](each start = 1.0);
  equation
    der(x[2]) = 0;
    der(x[1]) = -x[n];
  end A;

  model B
    A a(n = 3);
  end B;
end FinalParam;

The illegal modification should be detected and reported as an error. Right now, this slips all the way through code generation and results an error during compilation of the C code.

[Martin Sjölund]

This is both a front-end and back-end error on OMC's behalf (front-end does not check the modification, uses the old one for array dimensions; back-end does not check index out of bounds).

[Francesco Casella]

An even simpler example:

package P
  model A
    parameter Real p = 10;
    final parameter Real q = 3*p;
  end A;

  model B
    A a(q = 0);
  end B;
end P;

Model B is compiled without any error or waring and produces a simulation code with q = 0, which is clearly against the language spec., section 7.2.6.

Failure to produce an error (or at least a warning) can cause very subtle bugs to go undetected in complex models.

Please make sure that in no cases can final elements be modified in a model.

[Per Östlund]

Partially fixed in 415e95d. I say partially, because the check has several exceptions due to the way we handle modifiers. Most importantly I don't check extends modifiers, because that leads to all kinds of issues. The check will catch the examples given in this ticket though. I don't consider this issue properly solved though, and will leave the ticket open until we can fix it properly.

[Per Östlund]

The new frontend gives an error:

[test.mo:11:9-11:14:writable] Notification: From here:
[test.mo:3:5-3:34:writable] Error: Trying to override final element n with modifier '= 3'.

It also does not have exceptions like mentioned in the above comment, so I think we can consider this fixed.