Bug, rational produced where an integer is expected. about z3

[Drup]

I also saw a similar problem, and for me it was usually an issue of promotion between heterogeneous operands. for i and r integers and rational, r + i was a rational and i + r an integer, or something like that.

I just added manual promotions everywhere and rolled with it.

edit: rationals, not floats.

[wintersteiger]

I can't reproduce this problem with the latest unstable branch anymore, but I suspect the problem is just hidden now, because this file is significantly faster (and with less unknowns) than before. Does either of you still see this problem?

[NikolajBjorner]

Issue fix for #161 deals with the case where the arithmetic solver returns l_undef (due to for example non-linear arithmetic). In this case models could still be extracted even though no assurance that the model would satisfy the formula. To ensure that integers are returned during this model construction phase, the arithmetic module truncates non-integer solutions to integer variables. It is reasonable to assume this is the core of also this issue.