Inheriting collections.namedtuple and calling super(C, cls).__new__(cls, ...) results in empty analysis set

[lostmsu]

Sample code:

import collections

class A(collections.namedtuple('A', [])):
  def __new__(cls):
    return super(A, cls).__new__(cls)

a = A()

Possibly related to #15, #174

[lostmsu]

A test can be cherry-picked from 010a67e

[lostmsu]

I am also unsure why in the following:

class B(object):
  def __new__(cls):
    return super(B, cls).__new__(cls)

class A(B):
  def __new__(cls):
    return super(A, cls).__new__(cls)

a = A()

a ends up possibly being either of A and B.

Initially I thought, that is because of my previous change to propagate return types, but if I comment it out, the result does not change.

[lostmsu]

I suspect a similar issue would happen to any non-specialized "built-in" class. I made a really dirty workaround for internal purposes here: 645da64

[jakebailey]

I think these specific cases aren't broken now, though I'm not sure if we do anything interesting with __new__.

[lostmsu]

@jakebailey I suggest to cherry-pick the regression test still. :) 010a67e