Is a Python Decorator the same as Java annotation, or Java with Aspects?

Python decorators are just syntactic sugar for passing a function to another function and replacing the first function with the result:

@decoratordef function():    pass

is syntactic sugar for

def function():    passfunction = decorator(function)

Java annotations by themselves just store metadata, you must have something that inspects them to add behaviour.

Java AOP systems are huge things built on top of Java, decorators are just language syntax with little to no semantics attached, you can't really compare them.

This is a very valid question that anyone dabbling in both these languages simultaneously, can get. I have spent some time on python myself, and have recently been getting myself up to speed with Java and here's my take on this comparison.

Java annotations are - just that: annotations. They are markers; containers of additional metadata about the underlying object they are marking/annotating. Their mere presence doesn't change execution flow of the underlying, or doesn't add encapsulation/wrapper of some sort on top of the underlying. So how do they help? They are read and processed by - Annotation Processors. The metadata they contain can be used by custom-written annotation processors to add some auxiliary functionality that makes lives easier; BUT, and again, they NEITHER alter execution flow of an underlying, NOR wrap around them.

The stress on "not altering execution flow" will be clear to someone who has used python decorators. Python decorators, while being similar to Java annotations in look and feel, are quite different under the hood. They take the underlying and wrap themselves around it in any which way, as desired by the user, possibly even completely avoiding running the underlying itself as well, if one chooses to do so. They take the underlying, wrap themselves around it, and replace the underlying with the wrapped ones. They are effectively 'proxying' the underlying!

Now that is quite similar to how Aspects work in Java! Aspects per se are quite evolved in terms of their mechanism and flexibility. But in essence what they do is - take the 'advised' method (I am talking in spring AOP nomenclature, and not sure if it applies to AspectJ as well), wrap functionality around them, along with the predicates and the likes, and 'proxy' the 'advised' method with the wrapped one.

Please note these musings are at a very abstract and conceptual level, to help get the big picture. As you start delving deeper, all these concepts - decorators, annotations, aspects - have quite an involving scope. But at an abstract level, they are very much comparable.

TLDR

In terms of look and feel, python decorators can be considered similar to Java annotations, but under the hood, they work very very similar to the way Aspects work in Java.

I use both of them in a similar way: to turn on/off debugging or testing options.

For example (Python decorators):

def measure_time(func):    def _measure_time(*args, **kwargs):        t0 = time.time()        ret = func(*args, **kwargs)        print "time=%lf" % (time.time()-t0)        ...        return ret    return _measure_time@measure_timedef train_model(self):    ...

For Java annotations, use getAnnotation, etc. can do the similar jobs or more complicated ones.