How to inject variable into scope with a decorator

问题:

[Disclaimer: there may be more pythonic ways of doing what I want to do, but I want to know how python's scoping works here]

I'm trying to find a way to make a decorator that does something like injecting a name into the scope of another function (such that the name does not leak outside the decorator's scope). For example, if I have a function that says to print a variable named var that has not been defined, I would like to define it within a decorator where it is called. Here is an example that breaks:

c = 'Message'

def decorator_factory(value):
    def msg_decorator(f):
        def inner_dec(*args, **kwargs):
            var = value
            res = f(*args, **kwargs)
            return res
        return inner_dec
    return msg_decorator

@decorator_factory(c)
def msg_printer():
    print var

msg_printer()

I would like it to print "Message", but it gives:

NameError: global name 'var' is not defined

The traceback even points to wher var is defined:

<ipython-input-25-34b84bee70dc> in inner_dec(*args, **kwargs)
      8         def inner_dec(*args, **kwargs):
      9             var = value
---> 10             res = f(*args, **kwargs)
     11             return res
     12         return inner_dec

So I don't understand why it can't find var.

Is there any way to do something like this?

回答1:

You can't. Scoped names (closures) are determined at compile time, you cannot add more at runtime.

The best you can hope to achieve is to add global names, using the function's own global namespace:

def decorator_factory(value):
    def msg_decorator(f):
        def inner_dec(*args, **kwargs):
            g = f.__globals__  # use f.func_globals for py < 2.6
            sentinel = object()

            oldvalue = g.get('var', sentinel)
            g['var'] = value

            try:
                res = f(*args, **kwargs)
            finally:
                if oldvalue is sentinel:
                    del g['var']
                else:
                    g['var'] = oldvalue

            return res
        return inner_dec
    return msg_decorator

f.__globals__ is the global namespace for the wrapped function, so this works even if the decorator lives in a different module. If var was defined as a global already, it is replaced with the new value, and after calling the function, the globals are restored.

This works because any name in a function that is not assigned to, and is not found in a surrounding scope, is marked as a global instead.

Demo:

>>> c = 'Message'
>>> @decorator_factory(c)
... def msg_printer():
...     print var
... 
>>> msg_printer()
Message
>>> 'var' in globals()
False

But instead of decorating, I could just as well have defined var in the global scope directly.

Note that altering the globals is not thread safe, and any transient calls to other functions in the same module will also still see this same global.

回答2:

You can't. Python has lexical scoping. That means the meaning of an identifier is determined solely based on the scopes that physically surround it when you look at the source code.

回答3:

Python is lexically scoped, so I'm afraid there is no clean way to do what you want without some potentially nasty side effects. I recommend just passing var into the function via the decorator.

c = 'Message'

def decorator_factory(value):
    def msg_decorator(f):
        def inner_dec(*args, **kwargs):
            res = f(value, *args, **kwargs)
            return res
        inner_dec.__name__ = f.__name__
        inner_dec.__doc__ = f.__doc__
        return inner_dec
    return msg_decorator

@decorator_factory(c)
def msg_printer(var):
    print var

msg_printer()  # prints 'Message'

回答4:

There is a clean way to do what you want without using global variable. If you want to be stateless and threads safe, you don't really have the choice.

Use the "kwargs" variable:

c = 'Message'

def decorator_factory(value):
    def msg_decorator(f):
    def inner_dec(*args, **kwargs):
        kwargs["var"] = value
        res = f(*args, **kwargs)
        return res
    return inner_dec
return msg_decorator

@decorator_factory(c)
def msg_printer(*args, **kwargs):
    print kwargs["var"]

msg_printer()

回答5:

Here is a simple demonstration of using a decorator to add a variable into the scope of a function.

>>> def add_name(name):
...     def inner(func):
...         # Same as defining name within wrapped
...         # function.
...         func.func_globals['name'] = name
...
...         # Simply returns wrapped function reference.
...         return func
... 
...     return inner
...
>>> @add_name("Bobby")
... def say_hello():
...     print "Hello %s!" % name
...
>>> print say_hello()
Hello Bobby!
>>>

回答6:

Here's a way of injecting multiple variables into a function's scope in a manner somewhat to similar to what @Martijn Pieters does in his answer. I'm posting it primarily because it's a more general solution than applying the decorator multiple times as would be required doing the same thing with the code in Martijn's (and many of the other) answers.

from functools import wraps

def inject_variables(context):
    """ Decorator factory. """
    def variable_injector(func):
        @wraps(func)
        def decorator(*args, **kwargs):
            try:
                func_globals = func.__globals__  # Python 2.6+
            except AttributeError:
                func_globals = func.func_globals  # Earlier versions.

            saved_values = func_globals.copy()  # Shallow copy of dict.
            func_globals.update(context)

            try:
                result = func(*args, **kwargs)
            finally:
                func_globals = saved_values  # Undo changes.

            return result

        return decorator

    return variable_injector

if __name__ == '__main__':
    namespace = {'a': 5, 'b': 3}

    @inject_variables(namespace)
    def test():
        print('a:', a)
        print('b:', b)

    test()

回答7:

Assuming that in python functions are objects, you can do...

#!/usr/bin/python3


class DecorClass(object):
    def __init__(self, arg1, arg2):
        self.a1 = arg1
        self.a2 = arg2

    def __call__(self, function):
        def wrapped(*args):
            print('inside class decorator >>')
            print('class members: {0}, {1}'.format(self.a1, self.a2))
            print('wrapped function: {}'.format(args))
            function(*args, self.a1, self.a2)
        return wrapped


    @DecorClass(1, 2)
    def my_function(f1, f2, *args):
        print('inside decorated function >>')
        print('decorated function arguments: {0}, {1}'.format(f1, f2))
        print('decorator class args: {}'.format(args))


    if __name__ == '__main__':
        my_function(3, 4)

and the result is:

inside class decorator >>
class members: 1, 2
wrapped function: (3, 4)
inside decorated function >>
decorated function arguments: 3, 4
decorator class args: (1, 2)

more explanation here http://python-3-patterns-idioms-test.readthedocs.io/en/latest/PythonDecorators.html