在Python中创建一个参数化类型，但使所有实例的子项为“超类型”

Question

假设我有一个带有t属性的Python类型。我想创建一个“参数化元类型”，以便以下工作：

class MySuperClass(type):
    pass

class MySubClass(MySuperClass):
    # Here is the problem -- How do I define types that contain stuff,
    # independent of an object?
    def __init__(self, t): # Or __getitem__
        self.t = t

    def __instancecheck__(self, instance):
        return isinstance(instance, MySubClass) and instance.t == self.t

    def __subclasscheck__(self, subclass):
        return MySubClass in subclass.__mro__ and subclass.t == self.t

class MyObject(metaclass=MySubClass):
    def __init__(self, t):
        self.t = t

# Test code:
## Both of these, square brackets work too
assert isinstance(MyObject(0), MySubClass(0))
assert not isinstance(MyObject(0), MySubClass(1))

## Ideally
assert isinstance(MyObject(0), MySuperClass) or isinstance(MyObject(0), MySubClass)

目前我收到以下错误：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-4-99ad08881526> in <module>
     14         return MySubClass in subclass.__mro__ and subclass.t == self.t
     15 
---> 16 class MyObject(metaclass=MySubClass):
     17     def __init__(self, t):
     18         self.t = t

TypeError: __init__() takes 2 positional arguments but 4 were given

Answer 1

可以满足第一部分或要求。但它需要辅助检查员课程。 MySubClass是类型的后代，MySubClass(0)应该是一个类。它足以在InstanceChecker中创建一个内部类MySubClass类，并将__instancecheck__覆盖它们。

代码可以是：

class MySubClass(MySuperClass):
    def __new__(cls, name, bases=None, namespace=None, *args, **kwargs):
        if bases is not None:
            return super().__new__(cls, name, bases, namespace, **kwargs)
        return cls.InstanceChecker(name)

    class InstanceChecker:
        def __init__(self, t):
            self.t = t
        def __instancecheck__(self, instance):
            return isinstance(instance.__class__, MySubClass) and instance.t == self.t            

class MyObject(metaclass=MySubClass):
    def __init__(self, t):
        self.t = t

# Test code:
## Both of these, square brackets work too
assert isinstance(MyObject(0), MySubClass(0))
assert not isinstance(MyObject(0), MySubClass(1))

顺便说一下，我已经删除了__subclasscheck__覆盖，因为t只在MyObject的一个实例属性中

---

或者，元类可以在bases参数中自动添加超类。在下面的代码中，MySuperClass不再是MySubClass的超类，而是MyObject的超类：

class MySuperClass():
    pass


class MySubClass(type):
    def __new__(cls, name, bases=None, namespace=None, *args, **kwargs):
        if bases is not None:
            return super().__new__(cls, name, bases + (MySuperClass,), namespace, **kwargs)
        return cls.InstanceChecker(name)
    class InstanceChecker:
        def __init__(self, t):
            self.t = t
        def __instancecheck__(self, instance):
            return isinstance(instance.__class__, MySubClass) and instance.t == self.t

class MyObject(metaclass=MySubClass):
    def __init__(self, t):
        self.t = t

# Test code:
## Both of these, square brackets work too
assert isinstance(MyObject(0), MySubClass(0))
assert not isinstance(MyObject(0), MySubClass(1))

## Ideally
assert isinstance(MyObject(0), MySuperClass)

Answer 2

我似乎找到了一个解决方案，它比我原先想到的更清洁。

class MyMetaSuper(type):
    pass

class MyObject:
    def __init__(self, t):
        self.t = t

    def __class_getitem__(cls, key):        
        class MyMeta(MyMetaSuper):
            t = key
            def __instancecheck__(self, instance):
                return isinstance(instance, cls) and self.t == instance.t

            def __subclasscheck__(self, subclass):
                return isinstance(subclass, MyMetaSuper) and self.t == subclass.t

        class MyObjectSpecific(MyObject, metaclass=MyMeta):
            pass

        return MyObjectSpecific

# Check for specific condition
assert isinstance(MyObject(0), MyObject[0])
# Make sure isinstance fails when condition fails
assert not isinstance(MyObject(0), MyObject[1])

# Test the generic object
assert isinstance(MyObject(0), MyObject)