[BUG] RealVar operations with LB<0 and UB>0 leads to SolverException

[fhpg]

Hello,

at first, thank you for this great software!

Maybe I have found an issue having the following code:

Model model = new Model();
RealVar a = model.realVar("A", -10, 10, 1e-8);
RealVar b = model.realVar("B", -10, 10, 1e-8);
a.div(b).realVar(1e-8);

Executing this leads to a SolverException:

Exception in thread "main" org.chocosolver.solver.exception.SolverException: RV_1: wrong domain definition, lower bound > upper bound at org.chocosolver.solver.variables.IVariableFactory.checkRealDomainRange(IVariableFactory.java:854) at org.chocosolver.solver.variables.IVariableFactory.realVar(IVariableFactory.java:642) at org.chocosolver.solver.variables.IVariableFactory.realVar(IVariableFactory.java:630) at org.chocosolver.solver.expression.continuous.arithmetic.BiCArExpression.realVar(BiCArExpression.java:105) at solver.SolverTester.main(SolverTester.java:18)

Actually my real world problem code uses ibex and a far more complex formula but I reduced the problem to these simple lines of code. Is this a BUG or am I missing something?

• Choco-solver version: 4.10.3
• Java version: OpenJDK Runtime Environment (build 11.0.7+10-post-Debian-3deb10u1)

Best regards,
Florian

[cprudhom]

Hmmm. you're right, there is something wrong in RealUtils.odiv(x,y).
I'll fix that asap.

Thank you for the bug reporting.
Meanwhile, you can replace the call to .ibex() by a call to .equation() that may fix the issue.

[fhpg]

Thank you for your reply. I had an eye on RealUtils.odiv() but I had no clue how to fix that correctly.
BTW: I am using pow() which is obviously only supported with Ibex. So I will wait for you to fix this and keep watching the upstream.

[cprudhom]

RealUtils.odiv() returns an interval that represents the result of a division of 'x' by 'y'.
I think that since the 2 intervals include the value 0, the return domain can't be anything else but (-infinity,+infinity).