A Deep Dive Into Python's functools.wraps Decorator

Decorators in Python are great! So far, my favorite use case for them is using decorators to store first-class functions, which is what Flask's app.route decorator does. However, due to the language's underlying mechanics, wrapping one object over another can result in the loss of valuable metadata from the encapsulated object. This is why it's crucial to use the wraps decorator from the Python Standard Library's functools (function tools) module when developing your own Python decorators.

What Does functools.wraps Do?

functools.wraps, an easy-to-use interface for functools.update_wrapper, is a decorator that automatically transfers the key metadata from a callable (generally a function or class) to its wrapper. Typically, this wrapper is another function, but it can be any callable object such as a class.

Besides the wrapped parameter, which accepts the callable that gets enclosed by the wrapper, there are two more arguments that we can play around with:

@functools.wraps(wrapped, assigned=WRAPPER_ASSIGNMENTS, updated=WRAPPER_UPDATES)

The assigned parameter tells functools.wraps which attributes should be taken from the object being wrapped and transferred over to the wrapper. The parameter has a default tuple WRAPPER_ASSIGNMENTS, which has the following dunder attributes:

  1. __module__: The name of the module in which the object is declared.
  2. __name__: The object name.
  3. __qualname__: A more detailed version of the __name__.
  4. __doc__: The doc string at the top of an object.

The updated parameter also has a default tuple value of (‘__dict__’,). The updated parameter tells the wraps decorator which attributes of the wrapper callable need to be updated with the values from the original object. By default, the wrapping object's __dict__ attribute gets updated with the key-value pairs of the __dict__ from the wrapped object.

The wraps decorator also adds a new attribute to the wrapping object called __wrapped__, which holds a pointer to the enclosed function or class. This is very useful as it allows you to peek into the actual object being wrapped to see more metadata about the object that functools.wraps doesn't automatically include, such as __defaults__.

Decorators Without functools.wraps

First, let's see what a function that has a decorator looks like when there is no metadata transfer. Below is the decorator that I'll be using in the example; this decorator isn't actually doing anything, but note that it has its own docstring (“““wrapper function”“”) and the __name__ of the wrapping function is called “wrapper.“

def example_decorator(func):
    def wrapper(*args, **kwargs):
        """Wrapper function"""
        return func(*args, **kwargs)

    return wrapper

Now, let's use the decorator on the following function and look at some of it’s metadata:

@example_decorator
def hello_world(planet: str = 'earth'):
    """Say hello to a world"""
    print(f"Hello, {planet}!")

# Checking metadata of the decorated function
print(f'{hello_world.__name__        =  }')
print(f'{hello_world.__doc__         =  }')
print(f'{hello_world.__annotations__ =  }')
print(f'{hello_world.__dict__        =  }')

Output:

hello_world.__name__        =  'wrapper'
hello_world.__doc__         =  'Wrapper function'
hello_world.__annotations__ =  {}
hello_world.__dict__        =  {}

From the output, we can see that none of the metadata was transferred to the wrapping function.

Furthermore, if we print the function object:

print(hello_world)

We get the following output, which gives us no information about the hello_world function.

<function example_decorator.<locals>.wrapper at 0x7a122c8a9090>

Decorators With functools.wraps

Now let's make the same decorator, but add the functools.wraps decorator. To use functools.wraps, you just need to add it to the wrapping object like such:

from functools import wraps


def example_decorator(func):
    @wraps(func)
    def wrapper(*args, **kwargs):
        """Wrapper function"""
        return func(*args, **kwargs)

    return wrapper

And if we now look at the same metadata as before:

@example_decorator
def hello_world(planet: str='earth'):
   """Say hello to a world"""
   print(f"Hello, {planet}!")

# Checking metadata of the decorated function
print(f'{hello_world.__name__         =  }')
print(f'{hello_world.__doc__          =  }')
print(f'{hello_world.__annotations__  =  }')
print(f'{hello_world.__dict__         =  }')

We get the following output:

hello_world.__name__         =  'hello_world'
hello_world.__doc__          =  'Say hello to a world'
hello_world.__annotations__  =  {'planet': <class 'str'>}
hello_world.__dict__         =  {'__wrapped__': <function hello_world at 0x7b49d5fccc10>}

We can see that the __name__ attribute has been updated to match the name of the function that the decorator is wrapping. The docstring of the wrapped function also gets updated, as well as the static typing annotations. The __dict__ attribute also contains all of the attributes from hello_world (which is nothing since it's a function) and also has the __wrapped__ attribute, which holds a direct link to the original function.

As mentioned above, the __wrapped__ attribute holds a pointer to the original object. This can be very useful if another program or developer wants to introspect/get more information on the encapsulated object. There is another Python Standard Library module for this called “inspect.“ In that module, there's a function called inspect.signature which actually looks for a __wrapped__ attribute in __dict__ and if one is found, it follows the __wrapped__ value to be able to inspect the actual object!

However, you can also manually use the __wrapped__ attribute to get extra information on the encapsulated object. For example, functools.wraps doesn't automatically transfer over default values to the wrapping object (because classes don’t have a __defaults__ attribute and a class can be a wrapper), but with __wrapped__, we can see the default values which are stored in the __defaults__ attribute:

print(f'{hello_world.__dict__["__wrapped__"].__defaults__ =  }')

And we can see that the output is as expected; the planet argument from the function above has a default value of “earth.”

hello_world.__dict__["__wrapped__"].__defaults__ =  ('earth',)

Lastly, we can also print the function object; it shows the correct function:

print(hello_world)

Output:

<function hello_world at 0x7a122d7295a0>

Transferring Extra Metadata

Using the predefined values from the wraps decorator is usually enough, but if you want to transfer more (or less) metadata over to the wrapper function, you can pass in your own arguments. Let's say that we want to save the default values of your function's arguments; we can add more dunder attributes to the assigned argument:

MORE_WRAPPER_ASSIGNMENTS = (
   '__module__', '__name__',
   '__qualname__', '__annotations__',
   '__doc__', '__defaults__',
   '__kwdefaults__'
)

def example_decorator(func):
    @wraps(func, assigned=MORE_WRAPPER_ASSIGNMENTS)
    def wrapper(*args, **kwargs):
        """Wrapper function"""
        return func(*args, **kwargs)

    return wrapper

@example_decorator
def hello_world(planet: str='earth'):
    """Say hello to a world"""
    print(f"Hello, {planet}!")

Now if we try to print out the __defaults__ dunder attribute, we will see (‘earth’,) because the planet parameter in the hello_world function has a default value of “earth”:

print(f'{hello_world.__defaults__ =  }')

And the output:

hello_world.__defaults__ =  ('earth',)

Final Thoughts

Saving the metadata of objects that use decorators is extremely important as it leads to code that's easier to debug and ensures that your objects still work with other parts of the language, such as introspection. Adding functools.wraps to your decorators allows you to easily carry over the most important attributes to the wrapper object. That being said, just be mindful that functools.wraps doesn't automatically move every attribute over, and you shouldn't have any problems!

Thanks for reading - I hope you learned something new!

About Me

Thanks for reading my article! I'm Jacob Padilla - a student at NYU Stern studying Business and Data/Computer Science. Besides programming, my main interests are rock climbing, sailing, ceramics, and photography.

Feel free to check out my open-source projects on GitHub, and follow me on Twitter or LinkedIn to stay up-to-date on my latest articles and other interesting activities.