Indeed. Unification-based type inference doesn't work great when the type constraints are inequalities.