How to detect is decorator has been applied to method or function?

Question

The goal is that I wan't to have one decorator that will work with both, function and instance methods, and I would like to retrieve within wrapping function the self object when decorator has been applied on method or the function object itself when applied to function.

Here's what I have found almost working, this is only the func I'm using to detect on what decorator has been applied:

def _is_method(func):
    for stack_frame in inspect.stack():
        # if the code_context of the stack frame starts with 'class' this
        # function is defined within a class and so a method.
        if inspect.getframeinfo(stack_frame[0]).code_context[0].strip().startswith('class'):
            return True
    return False

This does work for me, with one small exception, It throws exceptions when I run tests in parallel in multiple processes.

Answer 1

You can solve this problem using descriptor protocol. By returning non-data descriptor from decorator you get to implement __get__ where you can save the method's instance/class.

Another (simpler) way would be to detect instance/class late, in decorator-made wrapper which may have self or cls as first of *args. This improves "inspectability" of decorated function, as it's still a plain function and not a custom non-data-desctiptor/function-object.

Problem we have to solve is that we cannot hook into or before the method binding:

Note that the transformation from function object to (unbound or bound) method object happens each time the attribute is retrieved from the class or instance.

In other words: when our wrapper runs, its descriptor protocol, namely __get__ method-wrapper of function, has already bound function with class/instance and resulting method is already being executed. We're left with args/kwargs and no straightforwardly accessible class-related info in current stack frame.

Let's start with solving class/staticmethod special cases and implementing wrapper as simple printer:

def decorated(fun):
    desc = next((desc for desc in (staticmethod, classmethod)
                 if isinstance(fun, desc)), None)
    if desc:
        fun = fun.__func__

    @wraps(fun)
    def wrap(*args, **kwargs):
        cls, nonselfargs = _declassify(fun, args)
        clsname = cls.__name__ if cls else None
        print('class: %-10s func: %-15s args: %-10s kwargs: %-10s' %
              (clsname, fun.__name__, nonselfargs, kwargs))

    wrap.original = fun

    if desc:
        wrap = desc(wrap)
    return wrap

Here comes the tricky part - if this was a method/classmethod call, first of args must be instance/class respectively. If so, we can get the very method we execute from this arg. If so, wrapper which we implemented above will be inside as __func__. If so, original member will be in our wrapper. If it is identical to fun from closure, we're home and can slice instance/class safely from remaining args.

def _declassify(fun, args):
    if len(args):
        met = getattr(args[0], fun.__name__, None)
        if met:
            wrap = getattr(met, '__func__', None)
            if getattr(wrap, 'original', None) is fun:
                maybe_cls = args[0]
                cls = maybe_cls if isclass(maybe_cls) else maybe_cls.__class__
                return cls, args[1:]
    return None, args

Let's see if this works with different variants of functions/methods:

@decorated
def simplefun():
    pass

class Class(object):
    @decorated
    def __init__(self):
        pass

    @decorated
    def method(self, a, b):
        pass

    @decorated
    @staticmethod
    def staticmethod(a1, a2=None):
        pass

    @decorated
    @classmethod
    def classmethod(cls):
        pass

Let's see if this actually runs:

simplefun()
instance = Class()
instance.method(1, 2)
instance.staticmethod(a1=3)
instance.classmethod()
Class.staticmethod(a1=3)
Class.classmethod()

output:

$ python Example5.py 
class: None       func: simplefun       args: ()         kwargs: {}        
class: Class      func: __init__        args: ()         kwargs: {}        
class: Class      func: method          args: (1, 2)     kwargs: {}        
class: None       func: staticmethod    args: ()         kwargs: {'a1': 3} 
class: Class      func: classmethod     args: ()         kwargs: {}        
class: None       func: staticmethod    args: ()         kwargs: {'a1': 3} 
class: Class      func: classmethod     args: ()         kwargs: {}        

Answer 2

You can use inspect.getargspec:

import inspect

def _is_method(func):
    spec = inspect.getargspec(func)
    return spec.args and spec.args[0] == 'self'

Example usage:

>>> def dummy_deco(f):
...     print('{} is method? {}'.format(f.__name__, _is_method(f)))
...     return f
... 
>>> @dummy_deco
... def add(a, b):
...     return a + b
... 
add is method? False
>>> class A:
...     @dummy_deco
...     def meth(self, a, b):
...         return a + b
... 
meth is method? True

NOTE This code depend on the name of the first argument. If the name is not self it will treat it as non-instance-method even though it is.

Answer 3

Thanks to this SO answer: Using the same decorator (with arguments) with functions and methods

I came to this solution witch works for me flawlessly:

def proofOfConcept():
    def wrapper(func):

        class MethodDecoratorAdapter(object):
            def __init__(self, func):
                self.func = func
                self.is_method = False

            def __get__(self, instance, owner):
                if not self.is_method:
                    self.is_method = True
                self.instance = instance

                return self

            def __call__(self, *args, **kwargs):
                # Decorator real logic goes here
                if self.is_method:
                    return self.func(self.instance, *args, **kwargs)
                else:
                    return self.func(*args, **kwargs)

        return wraps(func)(MethodDecoratorAdapter(func))

    return wrapper

NOTE This is not thread safe, to have a thread safe method one must return a callable object from __get__ that will have scope tied to instance

Answer 4

Solution for python3:

import inspect

def _is_method(func):
    spec = inspect.signature(func)
    if len(spec.parameters) > 0:
        if list(spec.parameters.keys())[0] == 'self':
            return True
    return False

Answer 5

import functools
import inspect


def mydec():
    def decorator(func):
        @functools.wraps(func)
        def pickled_func(*args, **kwargs):
            is_method = False
            if len(args) > 0:
                method = getattr(args[0], func.__name__, False)
                if method:
                    wrapped = getattr(method, "__wrapped__", False)
                    if wrapped and wrapped == func:
                        print("Used mydec to a method")
                        is_method = True

            if is_method is False:
                print("Used mydec not to a method.")
            result = func(*args, **kwargs)
            return result
    return decorator

Check whether the __wrapped__ variable is the same function as decorated.

Answer 6

My solution after taking a lot of inspiration from the answer of @canni. And from @aurzenligl to deal with static and classmethods.

Edit: I also added support for decorating properties

I have this class that can be reused for creating multiple decorators:

class MethodCheck:
    def __init__(self, func) -> None:
        self.func = func

    def __get__(self, instance, owner):
        if isinstance(self.func, staticmethod):
            return lambda *args, **kwargs: self.on_staticmethod(self.func.__func__, *args, **kwargs)
        elif isinstance(self.func, classmethod):
            return lambda *args, **kwargs: self.on_classmethod(self.func.__func__, owner, *args, **kwargs)
        elif isinstance(self.func, property):
            return self.on_property(self.func, instance)
        return lambda *args, **kwargs: self.on_method(self.func, instance, *args, **kwargs)

    def __call__(self, *args, **kwargs):
        return self.on_function(self.func, *args, **kwargs)

    def on_method(self, func, instance, *args, **kwargs):
        return func(instance, *args, **kwargs)

    def on_property(self, func:property, instance, *args, **kwargs):
        return func.fget(instance, *args, **kwargs)

    def on_classmethod(self, func, cls, *args, **kwargs):
        return func(cls, *args, **kwargs)

    def on_staticmethod(self, func, *args, **kwargs):
        return func(*args, **kwargs)

    def on_function(self, func, *args, **kwargs):
        return func(*args, **kwargs)

The implementation of a decorator using the above class:

from functools import wraps

class MethodCheckTester(MethodCheck):
    def on_method(self, *args, **kwargs):
        print(f"Executing Method: ", end='')
        return super().on_method(*args, **kwargs)

    def on_property(self, *args, **kwargs):
        print(f"Accessing Property: ", end='')
        return super().on_property(*args, **kwargs)

    def on_classmethod(self, *args, **kwargs):
        print(f"Executing Class Method: ", end='')
        return super().on_classmethod(*args, **kwargs)

    def on_staticmethod(self, *args, **kwargs):
        print(f"Executing Static Method: ", end='')
        return super().on_staticmethod(*args, **kwargs)

    def on_function(self, *args, **kwargs):
        print(f"Executing Function: ", end='')
        return super().on_function(*args, **kwargs)


def decorator(func):
    return wraps(func)(MethodCheckTester(func))

Decorated test functions/methods:

class MyClass:
    @decorator
    def f1(self, arg1):
        print(arg1)

    @decorator
    @classmethod
    def f2(cls, arg1):
        print(arg1)

    @decorator
    @staticmethod
    def f3(arg1):
        print(arg1)

    @decorator
    @property
    def f5(self):
        return "f5"

@decorator
def f4(arg1):
    print(arg1)

obj = MyClass()
obj.f1("f1")
obj.f2("f2")
obj.f3("f3")
f4("f4")
print(obj.f5)

This will print:

Executing Method: f1
Executing Class Method: f2
Executing Static Method: f3
Executing Function: f4
Accessing Property: f5