I’m upgrading an app from Graphene 2 to 3 and have run into an issue with some of the more complex ObjectType inheritance patterns that we have.

We have some code that looks similar to this:

class Person(ObjectType):
    first_name = graphene.String()
    last_name = graphene.String()
    
    def __init__(self, first_name, last_name):
        self.first_name = first_name
        self.last_name = last_name

class Employee(Person):
    email = graphene.String()

    def __init__(self, first_name, last_name):
        super().__init__(first_name, last_name)
        self.email = f"{self.first_name}{self.last_name}@company.com"

This used to work in graphene 2, but in graphene 3 it raises an Exception with the message:
'Employee' object has no attribute 'first_name'.

I believe this may be related to Graphene 3 optimizations to ObjectType initialization: https://github.com/graphql-python/graphene/pull/1157.

Looking at the MRO of Employee it looks something like:

ipdb> Employee.mro()
[<Employee meta=<ObjectTypeOptions name='Employee'>>, <class 'graphene.types.objecttype.ObjectTypeMeta.__new__.<locals>.InterObjectType'>, <Person meta=None>, <class 'graphene.types.objecttype.ObjectTypeMeta.__new__.<locals>.InterObjectType'>, <ObjectType meta=None>, <class 'graphene.types.objecttype.ObjectTypeMeta.__new__.<locals>.InterObjectType'>, <BaseType meta=None>, <SubclassWithMeta meta=None>, <class 'object'>]

So I think super().__init__() is resolving to the __init__ method on 'graphene.types.objecttype.ObjectTypeMeta.__new__.<locals>.InterObjectType'. Thus, __init__ is never called on Person.

I can overcome this issue by calling __init__ directly on Person rather than as a method on super. But, generally this is assumed to be a bad practice.

Is there a recommended way to implement an ObjectType class hierarchy with custom __init__ methods in Graphene 3?