decorators are a powerful concept that allow you to “wrap” a function with another function. Python The idea of a decorator is to abstract away something that you want a function or class to do, besides its normal responsibility. This can be helpful for many reasons such as code reuse, and sticking to . curlys law By learning how to write your own decorators, you can significantly improve readability of your own code. They can change how the function behaves, without needing to actually change the code (such as adding logging lines). They are a fairly common in python, familiar to those who use frameworks such as flask or click, but many only know how to use them, not how to write their own. tool This is a guest-post brought to you by your friends @ Timber. If you’re interested in writing for us, feel free to reach out on Twitter . First off, let’s show an example of a decorator in python. The following is a very basic example of what a decorator would like like if you were using it. 1 2 3 @my_decorator def hello(): print('hello') When you define a function in python, that function becomes an object. The function above is a function object. The is actually a function that has the ability to use the object, and return a different object to the interpreter. The object that the decorator returns, is what becomes known as . Essentially, it's the same thing as if you were going to write your own normal function, such as: . Decorate is passed the function -- which it can use however it wants -- then returns another object. The decorator has the ability to swallow the function, or return something that is not a function, if it wanted. hello @my_decorator hello hello hello = decorate(hello) As mentioned, a decorator is simply a function that is passed a function, and returns an object. So, to start writing a decorator, we just need to define a function. 1 2 def my_decorator(f): return 5 Any function can be used as a decorator. In this example the decorator is passed a function, and returns a different object. It simply swallows the function it is given entirely, and always returns 5. 1 2 3 @my_decorator def hello(): print('hello')

1 2 3 4 5 >>> hello() Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: 'int' object is not callable 'int' object is not callable Because our decorator is returning an int, and not a , it can not be called as a function. Remember, the decorator's return value . callable replaces hello 1 2 >>> hello 5 In most cases, we want the object our decorator returns, to actually mimic the function we decorated. . For example, let’s say we wanted to simply print every time the function was called, we could write a function that prints that information, then calls the function. But that function would need to be returned by the decorator. This usually leads to function nesting such as: This means that the object the decorator returns, needs to be a function itself 1 2 3 4 5 def mydecorator(f): # f is the function passed to us from python def log_f_as_called(): print(f'{f} was called.') f() return log_f_as_called As you can see, we have a nested function definition, and the decorator function is returning that function. This way, the function can still be called like a standard function, the caller does not need to know that it is decorated. We can now define as the following: hello hello 1 2 3 @mydecorator def hello(): print('hello') We get the following output: 1 2 3 >>> hello() <function hello at 0x7f27738d7510> was called. hello (Note: the number inside the _<function hello at 0x7f27738d7510>_ reference will be different for everyone, it represents the memory address) A function can be decorated many times, if desired. In this case, the decorators cause a chain effect. Essentially, the top decorator is passed the object from the former, and so on and so forth. For example, if we have the following code: 1 2 3 4 5 @a @b @c def hello(): print('hello') The interpreter is essentially doing and all of the decorators would wrap each other. You can test this out yourself by using our existing decorator, and using it twice. hello = a(b(c(hello))) 1 2 3 4 5 6 7 8 9 @mydecorator @mydecorator def hello(): print('hello') >>> hello() <function mydec.<locals>.a at 0x7f277383d378> was called. <function hello at 0x7f2772f78ae8> was called. hello You will notice that the first decorator, wrapped the second, and printed it separately. One interesting thing you might have noticed here, is that first line printed said instead of what the second line printed, which is what you would expect: . The is because the object returned by the decorator is a new function, not called . This is fine for our trivial example, but can often break tests and things that might be trying to introspect the function attributes. If the idea is for a decorator to act like the function it decorates, it needs to also mimic that function. Luckily, there is a python standard library decorator called for that in functools module. <function mydec.<locals>.a at 0x7f277383d378> <function hello at 0x7f2772f78ae8> hello wraps 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 import functools def mydecorator(f):  @functools.wraps(f) # we tell wraps that the function we are wrapping is f def log_f_as_called(): print(f'{f} was called.') f() return log_f_as_called @mydecorator @mydecorator def hello(): print('hello') >>> hello() <function hello at 0x7f27737c7950> was called. <function hello at 0x7f27737c7f28> was called. hello Now, our new function appears just like the one its wrapping/decorating. However, we are still reliant on the fact that it returns nothing, and accepts no input. If we wanted to make that more general, we would need to pass in the arguments and return the same value. We can modify our function to look like this: 1 2 3 4 5 6 7 8 9 import functools def mydecorator(f):  @functools.wraps(f) # wraps is a decorator that tells our function to act like f def log_f_as_called(*args, **kwargs): print(f'{f} was called with arguments={args} and kwargs={kwargs}') value = f(*args, **kwargs) print(f'{f} return value {value}') return value return log_f_as_called Now we are printing everytime the function is called, including all the inputs the function receives, and what it returns. Now you can simply decorate any existing function and get debug logging on all its inputs and outputs without needed to manually write the logging code. If you were using the decorator for any code that you wanted to ship, and not just for yourself locally, you might want to replace all the statements with logging statements. In which case, you would need to define a log level. It might be safe to assume you always want the debug log level, but what that also might depend on the function. We can provide variables to the decorator itself, to define how it should behave. For example: print 1 2 3 @debug(level='info') def hello(): print('hello') The above code would allow us to specify that this specific function should log at the info level instead of the debug level. This is achieved in python by writing a function, that returns a decorator. Yes, a decorator is also a function. So this is essentially saying . That double parenthesis might look funky, but basically, debug is function, that returns a function. Adding this to our existing decorator, we would need to nest one more time, now making our code look like the following: hello = debug('info')(hello) 1 2 3 4 5 6 7 8 9 10 11 import functools def debug(level): def mydecorator(f) @functools.wraps(f) def log_f_as_called(*args, **kwargs): logger.log(level, f'{f} was called with arguments={args} and kwargs={kwargs}') value = f(*args, **kwargs) logger.log(level, f'{f} return value {value}') return value return log_f_as_called return mydecorator The changes above turn into a function, which returns a decorator which uses the correct logging level. This is starting to get a bit ugly, and overly nested. A cool little trick to solve that, which I like to do, is simply add a default kwarg to and return a . A partial is a "non-complete function call" that includes a function and some arguments, so that they are passed around as one object without actually calling the function yet. debug level debug partial 1 2 3 4 5 6 7 8 9 10 11 import functools def debug(f=None, *, level='debug'): if f is None: return functools.partial(debug, level=level)  @functools.wraps(f) # we tell wraps that the function we are wrapping is f def log_f_as_called(*args, **kwargs): logger.log(level, f'{f} was called with arguments={args} and kwargs={kwargs}') value = f(*args, **kwargs) logger.log(level, f'{f} return value {value}') return value return log_f_as_called Now the decorator can work as normal: 1 2 3 @debug def hello(): print('hello') And use debug logging. Or, you could override the log level: 1 2 3 @debug('warning') def hello(): print('hello') Writing your logs to a hosted log aggregation service makes your life easier, so you can focus on building better software, not debugging. Disclaimer: I’m a current employee @ Timber. This section is entirely optional, but I sincerely believe that logging to the cloud will make your life easier (and you can try it for completely free). 1 pip install timber

1 2 3 4 5 6 7 import logging import timber logger = logging.getLogger(__name__) timber_handler = timber.TimberHandler(api_key='...') logger.addHandler(timber_handler) That’s it. All you need to do is get your API key from timber.io, and you’ll be able to see your logs. We automatically capture them from the logging module (and seamlessly add context), so you can continue to log normally. Originally published at timber.io .

Twitter

Too Long; Didn't Read

In your car, at home, or at work — Bosch technology shapes many areas of life.

Decorators in Python: What you need to know

Too Long; Didn't Read

Company Mentioned

Timber.io

About Author

TOPICS

THIS ARTICLE WAS FEATURED IN...

Decorators in Python: What you need to know

Too Long; Didn't Read

Company Mentioned

Timber.io

About Author

TOPICS

THIS ARTICLE WAS FEATURED IN...

RELATED STORIES