zope.cachedescriptors 5.0

Method and property caching decorators

zope.cachedescriptors

Cached descriptors cache their output. They take into account instance attributes that they depend on, so when the instance attributes change, the descriptors will change the values they return.

Cached descriptors cache their data in _v_ attributes, so they are also useful for managing the computation of volatile attributes for persistent objects.

Persistent descriptors:

• property

  A simple computed property.

  See src/zope/cachedescriptors/property.rst.

• method

  Idempotent method. The return values are cached based on method arguments and on any instance attributes that the methods are defined to depend on.

  Note

  Only a cache based on arguments has been implemented so far.

  See src/zope/cachedescriptors/method.rst.

Cached Properties

Cached properties are computed properties that cache their computed values. They take into account instance attributes that they depend on, so when the instance attributes change, the properties will change the values they return.

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
...
...     @property.CachedProperty('x', 'y')
...     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 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']

For backwards compatibility, the same thing can alternately be written without using decorator syntax:

>>> 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'

Documentation and the __name__ are preserved if the attribute is accessed through the class. This allows Sphinx to extract the documentation.

>>> class Point:
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @property.CachedProperty('x', 'y')
...     def radius(self):
...         '''The length of the line between self.x and self.y'''
...         print('computing radius')
...         return math.sqrt(self.x**2 + self.y**2)

>>> print(Point.radius.__doc__)
The length of the line between self.x and self.y
>>> print(Point.radius.__name__)
radius

It is possible to specify a CachedProperty that has no dependencies. For backwards compatibility this can be written in a few different ways:

>>> class Point:
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @property.CachedProperty
...     def no_deps_no_parens(self):
...         print("No deps, no parens")
...         return 1
...
...     @property.CachedProperty()
...     def no_deps(self):
...         print("No deps")
...         return 2
...
...     def no_deps_old_style(self):
...         print("No deps, old style")
...         return 3
...     no_deps_old_style = property.CachedProperty(no_deps_old_style)


>>> point = Point(1.0, 2.0)
>>> point.no_deps_no_parens
No deps, no parens
1
>>> point.no_deps_no_parens
1
>>> point.no_deps
No deps
2
>>> point.no_deps
2
>>> point.no_deps_old_style
No deps, old style
3
>>> point.no_deps_old_style
3

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'

Documentation and the __name__ are preserved if the attribute is accessed through the class. This allows Sphinx to extract the documentation.

>>> class Point:
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @property.Lazy
...     def radius(self):
...         '''The length of the line between self.x and self.y'''
...         print('computing radius')
...         return math.sqrt(self.x**2 + self.y**2)

>>> print(Point.radius.__doc__)
The length of the line between self.x and self.y
>>> print(Point.radius.__name__)
radius

The documentation of the attribute when accessed through the instance will be the same as the return-value:

>>> p = Point(1.0, 2.0)
>>> p.radius.__doc__ == float.__doc__
computing radius
True

This is the same behaviour as the standard Python property decorator.

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

Documentation and the __name__ are preserved if the attribute is accessed through the class. This allows Sphinx to extract the documentation.

>>> class Point:
...
...     def __init__(self, x, y):
...         self.x, self.y = x, y
...
...     @property.readproperty
...     def radius(self):
...         '''The length of the line between self.x and self.y'''
...         print('computing radius')
...         return math.sqrt(self.x**2 + self.y**2)

>>> print(Point.radius.__doc__)
The length of the line between self.x and self.y
>>> print(Point.radius.__name__)
radius

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'

Documentation is preserved if the attribute is accessed through the class. This allows Sphinx to extract the documentation.

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

>>> print(Point.radius.__doc__)
The length of the line between self.x and self.y

Method Cache

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):
...         """Compute the distance"""
...         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)

The documentation of the function is preserved (whether through the instance or the class), allowing Sphinx to extract it:

>>> print(point.distance.__doc__)
Compute the distance
>>> print(point.distance.__name__)
distance

>>> print(Point.distance.__doc__)
Compute the distance
>>> print(Point.distance.__name__)
distance

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

Changes

5.0 (2023-03-27)

• Add support for Python 3.11.

• Drop support for Python 2.7, 3.5, 3.6.

4.4 (2022-09-07)

• Drop support for Python 3.4.

• Add support for Python 3.7, 3.8, 3.9, 3.10.

4.3.1 (2017-12-09)

• Fix test which will break in the upcoming Python 3.7 release.

4.3.0 (2017-07-27)

• Add support for Python 3.6.

• Drop support for Python 3.3.

4.2.0 (2016-09-05)

• Add support for Python 3.5.

• Drop support for Python 2.6 and 3.2.

• The properties from the property module all preserve the documentation string of the underlying function, and all except cachedIn preserve everything that functools.update_wrapper preserves.

• property.CachedProperty is usable as a decorator, with or without dependent attribute names.

• method.cachedIn preserves the documentation string of the underlying function, and everything else that functools.wraps preserves.

4.1.0 (2014-12-26)

• Add support for PyPy and PyPy3.

• Add support for Python 3.4.

• Add support for testing on Travis.

4.0.0 (2013-02-13)

• Drop support for Python 2.4 and 2.5.

• Add support for Python 3.2 and 3.3.

3.5.1 (2010-04-30)

• Remove undeclared testing dependency on zope.testing.

3.5.0 (2009-02-10)

• Remove dependency on ZODB by allowing to specify storage factory for zope.cachedescriptors.method.cachedIn which is now dict by default. If you need to use BTree instead, you must pass it as factory argument to the zope.cachedescriptors.method.cachedIn decorator.

• Remove zpkg-related file.

• Clean up package description and documentation a bit.

• Change package mailing list address to zope-dev at zope.org, as zope3-dev at zope.org is now retired.

3.4.0 (2007-08-30)

Initial release as an independent package