Skip to main content

General purpose proxy and wrapper types

Project description

The peak.util.proxies module provides some useful base classes for creating proxies and wrappers for ordinary Python objects. Proxy objects automatically delegate all attribute access and operations to the proxied object. Wrappers are similar, but can be subclassed to allow additional attributes and operations to be added to the wrapped object.

Note that these proxy types are not intended to be tamper-proof; the unproxied form of an object can be readily accessed using a proxy’s __subject__ attribute, and some proxy types even allow this attribute to be set. (This can be handy for algorithms that lazily create circular structures and thus need to be able to hand out “forward reference” proxies.)

Proxy Basics

Here’s a quick demo of the ObjectProxy type:

>>> from peak.util.proxies import ObjectProxy
>>> p = ObjectProxy(42)

>>> p

>>> isinstance(p, int)

>>> p.__class__
<... 'int'>

>>> p*2

>>> 'X' * p

>>> hex(p)

>>> chr(p)

>>> p ^ 1

>>> p ** 2

As you can see, a proxy is virtually indistinguishable from the object it proxies, except via its __subject__ attribute, and its type():

>>> p.__subject__

>>> type(p)
<class 'peak.util.proxies.ObjectProxy'>

You can change the __subject__ of an ObjectProxy, and it will then refer to something else:

>>> p.__subject__ = 99
>>> p
>>> p-33

>>> p.__subject__ = "foo"
>>> p

All operations are delegated to the subject, including setattr and delattr:

>>> class Dummy: pass
>>> d = Dummy()
>>> p = ObjectProxy(d)

>>> = "bar"

>>> del
>>> hasattr(d,'foo')

Callback Proxies

Sometimes, you may want a proxy’s subject to be determined dynamically whenever the proxy is used. For this purpose, you can use the CallbackProxy type, which accepts a callback function and creates a proxy that will invoke the callback in order to get the target. Here’s a quick example of a counter that gets incremented each time it’s used, from zero to three:

>>> from peak.util.proxies import CallbackProxy

>>> ct = -1
>>> def callback():
...     global ct
...     if ct == 3: raise StopIteration
...     ct += 1
...     return ct

>>> counter = CallbackProxy(callback)

>>> counter
>>> counter
>>> str(counter)
>>> hex(counter)

>>> counter
Traceback (most recent call last):

As you can see, the callback is automatically invoked on any attempt to use the proxy. This is a somewhat silly example; a better one would be something like a thread_id proxy that is always equal to the ID # of the thread it’s running in.

A callback proxy’s callback can be obtained or changed via the get_callback and set_callback functions:

>>> from peak.util.proxies import get_callback, set_callback
>>> set_callback(counter, lambda: 42)

>>> counter

>>> get_callback(counter)
<function <lambda> at ...>

Lazy Proxies

A LazyProxy is similar to a CallbackProxy, but its callback is called at most once, and then cached:

>>> from peak.util.proxies import LazyProxy

>>> def callback():
...     print("called!")
...     return 42

>>> lazy = LazyProxy(callback)
>>> lazy
>>> lazy

You can use the get_callback and set_callback functions on lazy proxies, but it has no effect if the callback was already called:

>>> set_callback(lazy, lambda: 99)
>>> lazy

But you can use the get_cache and set_cache functions to tamper with the cached value:

>>> from peak.util.proxies import get_cache, set_cache
>>> get_cache(lazy)
>>> set_cache(lazy, 99)
>>> lazy


The ObjectWrapper, CallbackWrapper and LazyWrapper classes are similar to their proxy counterparts, except that they are intended to be subclassed in order to add custom extra attributes or methods. Any attribute that exists in a subclass of these classes will be read or written from the wrapper instance, instead of the wrapped object. For example:

>>> from peak.util.proxies import ObjectWrapper
>>> class NameWrapper(ObjectWrapper):
...     name = None
...     def __init__(self, ob, name):
...         ObjectWrapper.__init__(self, ob)
... = name
...     def __str__(self):
...         return

>>> w = NameWrapper(42, "The Ultimate Answer")
>>> w

>>> print(w)
The Ultimate Answer

>>> w * 2

'The Ultimate Answer'

Notice that any attributes you add must be defined in the class. You can’t add arbitary attributes at runtime, because they’ll be set on the wrapped object instead of the wrapper:

>>> = 'bar'
Traceback (most recent call last):
AttributeError: 'int' object has no attribute 'foo'

Note that this means that all instance attributes must be implemented as either slots, properties, or have a default value defined in the class body (like the name = None shown in the example above).

The CallbackWrapper and LazyWrapper base classes are basically the same as ObjectWrapper, except that they use a callback or cached lazy callback instead of expecting an object as their subject.

Creating Custom Subclasses and Mixins

In addition to all the concrete classes described above, there are also two abstract base classes: AbstractProxy and AbstractWrapper. If you want to create a mixin type that can be used with any of the concrete types, you should subclass the abstract version and set __slots__ to an empty list:

>>> from peak.util.proxies import AbstractWrapper

>>> class NamedMixin(AbstractWrapper):
...     __slots__ = []
...     name = None
...     def __init__(self, ob, name):
...         super(NamedMixin, self).__init__(ob)
... = name
...     def __str__(self):
...         return

Then, when you mix it in with the respective base class, you can add back in any necessary slots, or leave off __slots__ to give the subclass instances a dictionary of their own:

>>> from peak.util.proxies import CallbackWrapper, LazyWrapper

>>> class NamedObject(NamedMixin, ObjectWrapper): pass
>>> class NamedCallback(NamedMixin, CallbackWrapper): pass
>>> class NamedLazy(NamedMixin, LazyWrapper): pass

>>> print(NamedObject(42, "The Answer"))
The Answer

>>> n = NamedCallback(callback, "Test")
>>> n
>>> n

>>> n = NamedLazy(callback, "Once")
>>> n
>>> n

Both the AbstractProxy and AbstractWrapper base classes work by assuming that self.__subject__ will be the wrapped or proxied object. If you don’t want to use any of the standard three ways of defining __subject__ (i.e., as an object, callback, or lazy callback), you will need to subclass AbstractProxy or AbstractWrapper and provide your own way of defining __subject__.

Mailing List

Please direct questions regarding this package to the PEAK mailing list; see for details.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ProxyTypes-0.10.0.tar.gz (9.2 kB view hashes)

Uploaded Source

Built Distributions

ProxyTypes-0.10.0-py3-none-any.whl (9.7 kB view hashes)

Uploaded Python 3

ProxyTypes-0.10.0-py2-none-any.whl (9.7 kB view hashes)

Uploaded Python 2

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page