What is the difference between class syntax and type()?

Question

I am well aware of fact that classes can be declared dynamically in python using type and have used it here and there. But I am still unclear what are the differences between these two functions.

def class_factory():
    def init(self, attr):
        self.attr = attr
    class_ = type('DynamicClass', (), {'__init__' : init})
    return class_

and

def class_factory():
    class DynamicClass:
        def __init__(self, attr):
            self.attr = attr
    return DynamicClass

I have seen first pattern used in a lot of code-bases (like django) but never seen second one so far. Still I find second one more clear, syntactically.

I am struggling to find proper explanation about why people use first pattern for declaring dynamic classes and not second one. I experimented with both functions and could not find any notable differences between classes obtained from both functions. I am hoping for a clear explanation about differences between above two patterns/syntaxes from any aspect (performance-wise or any other) they differ.

Thanks,

I *have* used the second pattern before, but rarely. As some of the other answers have stated, using `type()` directly offers the most flexibility. For example if you want to be able to compute the *name* of the class one must used `type()`; similarly if the class body is computed in some way (or even its base classes). — Iguananaut, Nov 07 '19 at 17:26

chepner · Accepted Answer · 2019-11-07T17:30:06.843

A class statement is a declarative syntax for a call to a metaclass.

class Foo(Bar, metaclass=SomeMeta):  # If not specified, the metaclass is type
    ...

is equivalent to

Foo = SomeMeta('Foo', (Bar,) {...})

where {...} is some mapping constructed from the body of the class statement. Just as a simple example, without going into full detail:

class Foo(Bar, metaclass=SomeMeta):
    some_attr = "foo string"

    def __init__(self, x=9):
        self.y = x = 3

would define a function name __init__, then pass {'__init__': __init__, 'some_attr': 'foo string'} to SomeMeta. The name __init__ only exists in a temporary namespace created by the class statement, without affecting any name __init__ that might exist outside the statement.

To answer your question, sometimes it is more flexible or simpler to construct the third argument the metaclass yourself than to let the class statement do it for you. The same goes for the second argument. The first argument is fixed by the syntax of the class statement, but could be generated at run-time if you call the metaclass yourself.

Note that this means you can really abuse the class statement. EnumMeta from the enum module, in my opinion, stretches what you should do with one. Here's a really abusive example. The metaclass doesn't actually have to be a type; it just has to be callable, and it needs to accept three positional arguments. It can accept additional keyword arguments as well. It can then do anything it wants with those arguments, and return anything it wants as well.

def FooMeta(name, bases, dct, msg):
    print(f'{msg} from {name}')
    return 3

class Foo(metaclass=FooMeta, msg="hi"):
    pass

Foo isn't even a class; it's bound to 3.

So basically first syntax gives us more control by letting us construct first and third argument of metaclass. Are there any other differences (performance wise)? — Nafees Anwar, Nov 07 '19 at 17:25
Not really. If the bottleneck in your code becomes the difference between using a `class` statement can calling `type`, congratulations on writing some screamingly fast code. — chepner, Nov 07 '19 at 17:34

score 0 · Answer 2 · answered Nov 07 '19 at 17:10

You chose a the right exemple, how do you create dynamicly a class with your second exemple ? You can't properly do it, that why the first case is prefered. Have a look to this toy exemple based on yours.

def class_factory(many_args=False):
    def init1(self, attr):
        self.attr = attr

    def init2(self, *attr):
        self.attr = attr

    init = init1
    if many_args:
        init = init2

    class_ = type('DynamicClass', (), {'__init__': init})
    return class_

many = class_factory(True)(1, 2, 3, 4)
one = class_factory()(1)

Results :

>>> print(many.attr, one.attr)
(1, 2, 3, 4) 1

In your second exemple is difficult to compose a class with differents attributes and methods.

You can put an `if` statement inside the `class` statement to select which definition of `__init__` to use based on the value of `many_args`. — chepner, Nov 07 '19 at 17:22
Well, similarly, one could still assign `DynamicClass.__init__ = init2 if many_args else init1`. — r.ook, Nov 07 '19 at 17:22

What is the difference between class syntax and type()?

2 Answers2