Misc fixes

[skirpichev]

No description provided.

[skirpichev]

@AaronStiff, I'm glad you are using part of this work, related to sympy/sympy#20902, but please remember that such sorting (either in the solve_poly_inequality() or Poly.real_roots()) is just a workaround.

I think that the correct solution - is fixing is_disjoint() method of RealInterval by using strict inequalities.

[skirpichev]

@AaronStiff, please don't try to ping me in the sympy repo. I will be not notified about, in general.

This time I've seen your question, so I'll try to addess this. The problem is that solve_poly_inequalities() assume that (distinct) real roots are sorted by value. And such sorting should be expected after RooOf._reals_sorted(). Unfortunately, it doesn't work if compared intervals intersect by boundaries, e.g. corresponding left and right ends are equal. IIRC, in sympy/sympy#20902 the problem polynomial is something like (x + 1)**3*(3*x + 1). In this case, for distinct roots, we have intervals like r1=[-1, -1] (this root is from the ground domain, i.e. ZZ) and something like r2=[-1, 0] for x=-1/3. Second root is not going to be improved by sorting helper, since r1.b==r2.a.

I think it's perfectly fine to keep improving roots in such cases, using strict inequalities in the RealInterval.is_disjoint(), as it was suggested. I would appreciate @jksuom feedback (see above commit), for better safety.

[AaronStiff]

Oh sorry about that. Thanks for explaining!

[skirpichev]

Ok, rebased fix for RealInterval goes to #1117, merged.