描述符是一种在多个属性上重复利用同一个存取逻辑的方式,他能"劫持"那些本对于self.__dict__的操作。描述符通常是一种包含__get__、__set__、__delete__三种方法中至少一种的类,给人的感觉是「把一个类的操作托付与另外一个类」。静态方法、类方法、property都是构建描述符的类。

我们先看一个简单的描述符的例子(基于我之前的分享的Python高级编程改编,这个PPT建议大家去看看):

class MyDescriptor(object):
     _value = ''
     def __get__(self, instance, klass):
         return self._value

     def __set__(self, instance, value):
         self._value = value.swapcase()


class Swap(object):
     swap = MyDescriptor()

注意MyDescriptor要用新式类。调用一下:

In [1]: from descriptor_example import Swap
In [2]: instance = Swap()
In [3]: instance.swap  # 没有报AttributeError错误,因为对swap的属性访问被描述符类重载了
Out[3]: ''
In [4]: instance.swap = 'make it swap'  # 使用__set__重新设置_value
In [5]: instance.swap
Out[5]: 'MAKE IT SWAP'
In [6]: instance.__dict__  # 没有用到__dict__:被劫持了
Out[6]: {}

这就是描述符的威力。我们熟知的staticmethod、classmethod如果你不理解,那么看一下用Python实现的效果可能会更清楚了:

>>> class myStaticMethod(object):
...     def __init__(self, method):
...         self.staticmethod = method
...     def __get__(self, object, type=None):
...         return self.staticmethod
...
>>> class myClassMethod(object):
...     def __init__(self, method):
...         self.classmethod = method
...     def __get__(self, object, klass=None):
...         if klass is None:
...             klass = type(object)
...         def newfunc(*args):
...             return self.classmethod(klass, *args)
...         return newfunc

在实际的生产项目中,描述符有什么用处呢?首先看MongoEngine中的Field的用法:

from mongoengine import *                      

class Metadata(EmbeddedDocument):                   
    tags = ListField(StringField())
    revisions = ListField(IntField())

class WikiPage(Document):                           
    title = StringField(required=True)              
    text = StringField()                            
    metadata = EmbeddedDocumentField(Metadata)

有非常多的Field类型,其实它们的基类就是一个描述符,我简化下,大家看看实现的原理:

class BaseField(object):
    name = None
    def __init__(self, **kwargs):
        self.__dict__.update(kwargs)
        ...

    def __get__(self, instance, owner):
        return instance._data.get(self.name)

    def __set__(self, instance, value):
        ...
        instance._data[self.name] = value

很多项目的源代码看起来很复杂,在抽丝剥茧之后,其实原理非常简单,复杂的是业务逻辑。

接着我们再看Flask的依赖Werkzeug中的cached_property:

class _Missing(object):
    def __repr__(self):
        return 'no value'

    def __reduce__(self):
        return '_missing'


_missing = _Missing() 


class cached_property(property):
    def __init__(self, func, name=None, doc=None):
        self.__name__ = name or func.__name__
        self.__module__ = func.__module__
        self.__doc__ = doc or func.__doc__
        self.func = func

    def __set__(self, obj, value):
        obj.__dict__[self.__name__] = value

    def __get__(self, obj, type=None):
        if obj is None:
            return self
        value = obj.__dict__.get(self.__name__, _missing)
        if value is _missing:
            value = self.func(obj)
            obj.__dict__[self.__name__] = value
        return value

其实看类的名字就知道这是缓存属性的,看不懂没关系,用一下:

class Foo(object):
    @cached_property
    def foo(self):
        print 'Call me!'
        return 42

调用下:

In [1]: from cached_property import Foo
   ...: foo = Foo()
   ...:

In [2]: foo.bar
Call me!
Out[2]: 42

In [3]: foo.bar
Out[3]: 42

可以看到在从第二次调用bar方法开始,其实用的是缓存的结果,并没有真的去执行。

说了这么多描述符的用法。我们写一个做字段验证的描述符:

class Quantity(object):
    def __init__(self, name):
        self.name = name

    def __set__(self, instance, value):
        if value > 0:
            instance.__dict__[self.name] = value
        else:
            raise ValueError('value must be > 0')


class Rectangle(object):
    height = Quantity('height')
    width = Quantity('width')

    def __init__(self, height, width):
        self.height = height
        self.width = width

    @property
    def area(self):
        return self.height * self.width

我们试一试:

In [1]: from rectangle import Rectangle
In [2]: r = Rectangle(10, 20)
In [3]: r.area
Out[3]: 200

In [4]: r = Rectangle(-1, 20)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-5-5a7fc56e8a> in <module>()
----> 1 r = Rectangle(-1, 20)

/Users/dongweiming/mp/2017-03-23/rectangle.py in __init__(self, height, width)
     15
     16     def __init__(self, height, width):
---> 17         self.height = height
     18         self.width = width
     19

/Users/dongweiming/mp/2017-03-23/rectangle.py in __set__(self, instance, value)
      7             instance.__dict__[self.name] = value
      8         else:
----> 9             raise ValueError('value must be > 0')
     10
     11

ValueError: value must be > 0

看到了吧,我们在描述符的类里面对传值进行了验证。ORM就是这么玩的!

但是上面的这个实现有个缺点,就是不太自动化,你看height = Quantity('height'),这得让属性和Quantity的name都叫做height,那么可不可以不用指定name呢?当然可以,不过实现的要复杂很多:

class Quantity(object):
    __counter = 0
    def __init__(self):
        cls = self.__class__
        prefix = cls.__name__
        index = cls.__counter
        self.name = '_{}#{}'.format(prefix, index)
        cls.__counter += 1

    def __get__(self, instance, owner):
        if instance is None:
            return self
        return getattr(instance, self.name)
    ...


class Rectangle(object):
    height = Quantity()
    width = Quantity() 
    ...

Quantity的name相当于类名+计时器,这个计时器每调用一次就叠加1,用此区分。有一点值得提一提,在__get__中的:

if instance is None:
    return self

在很多地方可见,比如之前提到的MongoEngine中的BaseField。这是由于直接调用Rectangle.height这样的属性时候会报AttributeError, 因为描述符是实例上的属性。

PS:这个灵感来自《Fluent Python》,书中还有一个我认为设计非常好的例子。就是当要验证的内容种类很多的时候,如何更好地扩展的问题。现在假设我们除了验证传入的值要大于0,还得验证不能为空和必须是数字(当然三种验证在一个方法中验证也是可以接受的,我这里就是个演示),我们先写一个abc的基类:

class Validated(abc.ABC):
    __counter = 0

    def __init__(self):
        cls = self.__class__
        prefix = cls.__name__
        index = cls.__counter
        self.name = '_{}#{}'.format(prefix, index)
        cls.__counter += 1

    def __get__(self, instance, owner):
        if instance is None:
            return self
        else:
            return getattr(instance, self.name)
    def __set__(self, instance, value):
        value = self.validate(instance, value)
        setattr(instance, self.name, value) 

    @abc.abstractmethod
    def validate(self, instance, value):
        """return validated value or raise ValueError"""

现在新加一个检查类型,新增一个继承了Validated的、包含检查的validate方法的类就可以了:

class Quantity(Validated):
    def validate(self, instance, value):
        if value <= 0:
            raise ValueError('value must be > 0')
        return value


class NonBlank(Validated):
    def validate(self, instance, value):
        value = value.strip()
        if len(value) == 0:
            raise ValueError('value cannot be empty or blank')
        return value

前面展示的描述符都是一个类,那么可不可以用函数来实现呢?也是可以的:

def quantity():
    try:
        quantity.counter += 1
    except AttributeError:
        quantity.counter = 0

    storage_name = '_{}:{}'.format('quantity', quantity.counter)

    def qty_getter(instance):
        return getattr(instance, storage_name)

    def qty_setter(instance, value):
        if value > 0:
            setattr(instance, storage_name, value)
        else:
            raise ValueError('value must be > 0')
    return property(qty_getter, qty_setter)