Weird MRO result when inheriting directly from typing.NamedTuple

Question

I am confused why FooBar.__mro__ doesn't show <class '__main__.Parent'> like the above two.

I still don't know why after some digging into the CPython source code.

from typing import NamedTuple
from collections import namedtuple

A = namedtuple('A', ['test'])

class B(NamedTuple):
  test: str

class Parent:
  pass

class Foo(Parent, A):
  pass

class Bar(Parent, B):
  pass

class FooBar(Parent, NamedTuple):
  pass

print(Foo.__mro__)
# prints (<class '__main__.Foo'>, <class '__main__.Parent'>, <class '__main__.A'>, <class 'tuple'>, <class 'object'>)

print(Bar.__mro__)
# prints (<class '__main__.Bar'>, <class '__main__.Parent'>, <class '__main__.B'>, <class 'tuple'>, <class 'object'>)

print(FooBar.__mro__)
# prints (<class '__main__.FooBar'>, <class 'tuple'>, <class 'object'>)
# expecting: (<class '__main__.FooBar'>, <class '__main__.Parent'>, <class 'tuple'>, <class 'object'>) 

Answer 1

This is because typing.NamedTuple is not really a proper type. It is a class. But its singular purpose is to take advantage of meta-class magic to give you a convenient nice way to define named-tuple types. And named-tuples derive from tuple directly.

Note, unlike most other classes,

from typing import NamedTuple
class Foo(NamedTuple):
    pass

print(isinstance(Foo(), NamedTuple))

prints False.

This is because in NamedTupleMeta essentially introspects __annotations__ in your class to eventually use it to return a class created by a call to collections.namedtuple:

def _make_nmtuple(name, types):
    msg = "NamedTuple('Name', [(f0, t0), (f1, t1), ...]); each t must be a type"
    types = [(n, _type_check(t, msg)) for n, t in types]
    nm_tpl = collections.namedtuple(name, [n for n, t in types])
    # Prior to PEP 526, only _field_types attribute was assigned.
    # Now __annotations__ are used and _field_types is deprecated (remove in 3.9)
    nm_tpl.__annotations__ = nm_tpl._field_types = dict(types)
    try:
        nm_tpl.__module__ = sys._getframe(2).f_globals.get('__name__', '__main__')
    except (AttributeError, ValueError):
        pass
    return nm_tpl

class NamedTupleMeta(type):

    def __new__(cls, typename, bases, ns):
        if ns.get('_root', False):
            return super().__new__(cls, typename, bases, ns)
        types = ns.get('__annotations__', {})
        nm_tpl = _make_nmtuple(typename, types.items())
        ...
        return nm_tpl

And of course, namedtuple essentially just creates a class which derives from tuple. Effectively, any other classes your named-tuple class derives from in the class definition statement are ignored, because this subverts the usual class machinery. It might feel wrong, in a lot of ways it is ugly, but practicality beats purity. And it is nice and practical to be able to write things like:

class Foo(NamedTuple):
    bar: int
    baz: str