Handling of (tuple) = f(...) when solving non-linear

[Willi Braun]

Examples like following were handled wrong in SimCodeUtil:

model TupleNonLinear
function ftest
  input Real a;
  output Real b;
  output Integer c;
algorithm
  b := 0;
  for i in {1} loop //prevent inlining
    b := b + sin(a);
    c := integer(a);
  end for;
end ftest;

  Real a;
  Integer c;
  Real x,z;
equation
  der(x) = sin(a + time);
  (x,c) = ftest(a);
  z = if c>=1 then 10 else 0;
  annotation (experiment(StopTime=2));
end TupleNonLinear;

We try to solve the equation
(x,c) = ftest(a) non-linear for
{a,c} but actually only a should be solved
non-linear and c can be calculated then directly.

[Lennart Ochel]

Replying to wbraun:

We try to solve the equation
(x,c) = ftest(a) non-linear for
{a,c} but actually only a should be solved
non-linear and c can be calculated then directly.

Why? That is actually not an error. I think it is good enough to solve it non-linear.

Anyway, the model gets handled wrong. I changed the model a bit, so that OpenModelica produces wrong results:

model TupleNonLinear
  function ftest
    input Real a;
    output Real b;
    output Integer c;
  algorithm
    b := 0;
    for i in {1} loop //prevent inlining
      b := b + sin(a);
      c := i;
    end for;
  end ftest;

  Real a;
  Integer c;
  Real x;
equation
  der(x) = a + time;
  (x,c) = ftest(a);
end TupleNonLinear;

[Willi Braun]

Yes, the code we would produce is wrong, since for all
non-linear systems we assume that the iteration variables
are Real and not Integer.

[Martin Sjölund]

Are you actually allowed to output both a continuous-time and discrete-time expression for the same equation?

Spec 3.8.3 says:

For an equation expr1 = expr2 where neither expression is of base type Real, both expressions must be discrete-time expressions.

Exceptions are only made for records.

[Willi Braun]

Replying to sjoelund.se:

Are you actually allowed to output both a continuous-time and discrete-time expression for the same equation?

Spec 3.8.3 says:

For an equation expr1 = expr2 where neither expression is of base type Real, both expressions must be discrete-time expressions.

Exceptions are only made for records.

That's a really good remark.
I changed the model a bit, so that the mode
has a discrete change of c while the
continues integration.

So I think you are right we need to handle
that whole equation as discrete.