Loss of modification equations during transformation

[Bernhard Thiele]

Following state machine model has two states for which the input xfoox is provided by modification equations.

loadString("
model test
  StateX state1(xfoox=1.0);
  StateX state2(xfoox=0.1);
equation
  initialState(state1);
  transition(state1, state2, true, immediate=false);
  transition(state2, state1, true, immediate=false);
end test;

model StateX
  input Real xfoox;
  Real y;
equation
  y = xfoox * sample(1, Clock(1,10));
end StateX;
"); getErrorString();
simulate(test); getErrorString();

The problem is that after simulation both input variables state1.xfoox and state2.xfoox remain on their "initial value" 0. If one provides an explicit start value, e.g., xfoox(start=0.1), that value is used instead. First analysis by Lennart and me suggests that the modification equations are "lost" either at the back-end transformation or the code generation.

[Lennart Ochel]

Fixed with ​OMCompiler#1485.
Test case added with ​OpenModelica-testsuite#562.

Well, these changes are pretty radical. I think we should also handle clocked partitions in the future. Therefore, we probably just need to add some restrictions on what is allowed to remove in case of clocked partitions.