CachedProperty

Cached properties cache their data in _v_ attributes, so they are also useful for managing the computation of volatile attributes for persistent objects. Let's look at an example:

>>> from zope.cachedescriptors import property
>>> import math

>>> class Point:
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     def radius(self):
...         print 'computing radius'
...         return math.sqrt(self.x**2 + self.y**2)
...     radius = property.CachedProperty(radius, 'x', 'y')

>>> point = Point(1.0, 2.0)

If we ask for the radius the first time:

>>> '%.2f' % point.radius
computing radius
'2.24'

We see that the radius function is called, but if we ask for it again:

>>> '%.2f' % point.radius
'2.24'

The function isn't called. If we change one of the attribute the radius depends on, it will be recomputed:

>>> point.x = 2.0
>>> '%.2f' % point.radius
computing radius
'2.83'

But changing other attributes doesn't cause recomputation:

>>> point.q = 1
>>> '%.2f' % point.radius
'2.83'

Note that we don't have any non-volitile attributes added:

>>> names = [name for name in point.__dict__ if not name.startswith('_v_')]
>>> names.sort()
>>> names
['q', 'x', 'y']

Lazy Computed Attributes

The property module provides another descriptor that supports a slightly different caching model: lazy attributes. Like cached proprties, they are computed the first time they are used. however, they aren't stored in volatile attributes and they aren't automatically updated when other attributes change. Furthermore, the store their data using their attribute name, thus overriding themselves. This provides much faster attribute access after the attribute has been computed. Let's look at the previous example using lazy attributes:

>>> class Point:
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @property.Lazy
...     def radius(self):
...         print 'computing radius'
...         return math.sqrt(self.x**2 + self.y**2)

>>> point = Point(1.0, 2.0)

If we ask for the radius the first time:

>>> '%.2f' % point.radius
computing radius
'2.24'

We see that the radius function is called, but if we ask for it again:

>>> '%.2f' % point.radius
'2.24'

The function isn't called. If we change one of the attribute the radius depends on, it still isn't called:

>>> point.x = 2.0
>>> '%.2f' % point.radius
'2.24'

If we want the radius to be recomputed, we have to manually delete it:

>>> del point.radius

>>> point.x = 2.0
>>> '%.2f' % point.radius
computing radius
'2.83'

Note that the radius is stored in the instance dictionary:

>>> '%.2f' % point.__dict__['radius']
'2.83'

The lazy attribute needs to know the attribute name. It normally deduces the attribute name from the name of the function passed. If we want to use a different name, we need to pass it:

>>> def d(point):
...     print 'computing diameter'
...     return 2*point.radius

>>> Point.diameter = property.Lazy(d, 'diameter')
>>> '%.2f' % point.diameter
computing diameter
'5.66'

readproperty

readproperties are like lazy computed attributes except that the attribute isn't set by the property:

>>> class Point:
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @property.readproperty
...     def radius(self):
...         print 'computing radius'
...         return math.sqrt(self.x**2 + self.y**2)

>>> point = Point(1.0, 2.0)

>>> '%.2f' % point.radius
computing radius
'2.24'

>>> '%.2f' % point.radius
computing radius
'2.24'

But you can replace the property by setting a value. This is the major difference to the builtin property:

>>> point.radius = 5
>>> point.radius
5

cachedIn

The cachedIn property allows to specify the attribute where to store the computed value:

>>> class Point:
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @property.cachedIn('_radius_attribute')
...     def radius(self):
...         print 'computing radius'
...         return math.sqrt(self.x**2 + self.y**2)

>>> point = Point(1.0, 2.0)

>>> '%.2f' % point.radius
computing radius
'2.24'

>>> '%.2f' % point.radius
'2.24'

The radius is cached in the attribute with the given name, _radius_attribute in this case:

>>> '%.2f' % point._radius_attribute
'2.24'

When the attribute is removed the radius is re-calculated once. This allows invalidation:

>>> del point._radius_attribute

>>> '%.2f' % point.radius
computing radius
'2.24'

>>> '%.2f' % point.radius
'2.24'

cachedIn

The cachedIn property allows to specify the attribute where to store the computed value:

>>> import math
>>> from zope.cachedescriptors import method

>>> class Point(object):
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @method.cachedIn('_cache')
...     def distance(self, x, y):
...         print 'computing distance'
...         return math.hypot(self.x - x, self.y - y)
...
>>> point = Point(1.0, 2.0)

The value is computed once:

>>> point.distance(2, 2)
computing distance
1.0
>>> point.distance(2, 2)
1.0

Using different arguments calculates a new distance:

>>> point.distance(5, 2)
computing distance
4.0
>>> point.distance(5, 2)
4.0

The data is stored at the given _cache attribute:

>>> isinstance(point._cache, dict)
True

>>> sorted(point._cache.items())
[(((2, 2), ()), 1.0), (((5, 2), ()), 4.0)]

It is possible to exlicitly invalidate the data:

>>> point.distance.invalidate(point, 5, 2)
>>> point.distance(5, 2)
computing distance
4.0

Invalidating keys which are not in the cache, does not result in an error:

>>> point.distance.invalidate(point, 47, 11)

It is possible to pass in a factory for the cache attribute. Create another Point class:

>>> class MyDict(dict):
...     pass
>>> class Point(object):
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @method.cachedIn('_cache', MyDict)
...     def distance(self, x, y):
...         print 'computing distance'
...         return math.sqrt((self.x - x)**2 + (self.y - y)**2)
...
>>> point = Point(1.0, 2.0)
>>> point.distance(2, 2)
computing distance
1.0

Now the cache is a MyDict instance:

>>> isinstance(point._cache, MyDict)
True