Bug in simplification of rem() or div()

[hkiel]

These two models behave wrong (and even differently):

model test1
  Real x(start=0,fixed=true);
  Real y1,y2;
equation
  der(x)=10;
  y1 = rem(x,2);
  y2 = rem(x+1,2);
end test1;

model test2
  Real x(start=0,fixed=true);
  Real y2;
equation
  der(x)=10;
  y2 = rem(x+1,2);
end test2;

Model test1 shows sawtooth, but not in the correct value range of 0..2, but in 1..3 for y2.

Model test2 does not show any sawtooth for y2, it is just plain ramp x+1.

Also the flattened code is not valid Modelica x + -2.0
instantiateModel(test1)

"class test1
  Real x(start = 0.0, fixed = true);
  Real y1;
  Real y2;
equation
  der(x) = 10.0;
  y1 = x + -2.0 * div(x, 2.0);
  y2 = 1.0 + x + -2.0 * div(1.0 + x, 2.0);
end test1;
"

I think that simplification does wrong things. The (syntactically fixed) model from instantiateModel(test1) is also simulated wrong, so maybe the problem lies within div simplifications.

[hkiel]

This seems to be a problem with zero crossing detection in div().
This model behaves wrong:

model d
  Real x(start=0,fixed=true);
  Real y;
equation
  der(x)=10;
  y=div(1.0+x,2.0);
end d;

this model behaves correct using noEvent() around the div():

model d
  Real x(start=0,fixed=true);
  Real y;
equation
  der(x)=10;
  y=noEvent(div(1.0+x,2.0));
end d;

However, adding noEvent() around the rem() in the initial problem does not fix it. It even makes things look worse...

[lochel]

Thanks for this very good report. I will fix the issue.

[lochel]

Fixed with ​OMCompiler#1655.

[hkiel]

@lochel Can you add a test case for this?

[lochel]

Replying to hkiel:

@lochel Can you add a test case for this?

Test added with ​OpenModelica-testsuite#640.