Is that OK to use the MRO in order to override a mixin?

Question

Problem description: I have a class C inheriting from mixins A and B. I want a new class, C_, having all the methods/attributes defined in the class C but with B swapped with B_ (same API) in the inheritance scheme (one possible use of this is easy mocking). All classes are new style classes.

I got what I wanted by messing with the inheritance order, therefore the MRO:

A   B   B_    B_  A   B
 \ /   /       \   \ /
  C   /         \   C
   \ /           \ /
    C1            C2
C1(C,B_)       C2(B_,C)

C1.__mro__ = (C1, C , A, B, B_, object)
C2.__mro__ = (C2, B_, C, A, B , object)

The C2 method (inheriting the modified mixin before the C class) works without much surprise and if I call a method defined in the B mixin, the B_'s definition is chosen.

For the moment it works, but I feel like: "fingers crossed, I hope a special case does not arise and break the system !"

The question is: is it a finally-not-so-wrong way to solve the problem or is there a better way to do it ?

PS: I think I could take my bazooka and create a metaclass to redefine the mro (as said in the official doc), but my instinct says it's not going to be necessarily prettier.

Answer 1

Your approach should work just fine. It is legitimate to use a subclass to control the MRO and to shadow one class with another.

This blog post gives several examples: https://rhettinger.wordpress.com/2011/05/26/super-considered-super/