The objproxies
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.)
Table of Contents
This is Python 3 port of ProxyTypes wrote by Phillip J. Eby as part of PEAK for Python 2.
The namespace was changed from peak.util.proxies
to objproxies
. Other than that it should be a compatible replacement.
So far the following was accomplished:
- Streamlined files and setup
- Ported unittests and doctests
- Cleaned up syntax
- Turn the module in a package, separate functionalities in different modules
- Simplify code wherever possible
- Get positive feedback from a couple of users
Contributions and bug reports are welcome.
When nose is available all tests can be run using:
nosetests3 --with-doctest --doctest-extension=rst .
Otherwise standard python will suffice:
python -m unittest objproxies_tests.py
python -m doctest README.rst
Here's a quick demo of the ObjectProxy
type:
>>> from objproxies import ObjectProxy
>>> p = ObjectProxy(42)
>>> p
42
>>> isinstance(p, int)
True
>>> p.__class__
<class 'int'>
>>> p*2
84
>>> 'X' * p
'XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX'
>>> hex(p)
'0x2a'
>>> chr(p)
'*'
>>> p ^ 1
43
>>> p ** 2
1764
As you can see, a proxy is virtually indistinguishable from the object it proxies, except via its __subject__
attribute, and its type()
:
>>> p.__subject__
42
>>> type(p)
<class 'objproxies.ObjectProxy'>
You can change the __subject__
of an ObjectProxy
, and it will then refer to something else:
>>> p.__subject__ = 99
>>> p
99
>>> p-33
66
>>> p.__subject__ = "foo"
>>> p
'foo'
All operations are delegated to the subject, including setattr
and delattr
:
>>> class Dummy: pass
>>> d = Dummy()
>>> p = ObjectProxy(d)
>>> p.foo = "bar"
>>> d.foo
'bar'
>>> del p.foo
>>> hasattr(d,'foo')
False
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 objproxies import CallbackProxy
>>> callback = iter(range(4)).__next__
>>> counter = CallbackProxy(callback)
>>> counter
0
>>> counter
1
>>> str(counter)
'2'
>>> hex(counter)
'0x3'
>>> counter
Traceback (most recent call last):
...
StopIteration
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 objproxies import get_callback, set_callback
>>> set_callback(counter, lambda: 42)
>>> counter
42
>>> type(get_callback(counter))
<class 'function'>
A LazyProxy
is similar to a CallbackProxy
, but its callback is called at most once, and then cached:
>>> from objproxies import LazyProxy
>>> def callback():
... print("called")
... return 42
>>> lazy = LazyProxy(callback)
>>> lazy
called
42
>>> lazy
42
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
42
But you can use the get_cache
and set_cache
functions to tamper with the cached value:
>>> from objproxies import get_cache, set_cache
>>> get_cache(lazy)
42
>>> set_cache(lazy, 99)
>>> lazy
99
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 objproxies import ObjectWrapper
>>> class NameWrapper(ObjectWrapper):
... name = None
... def __init__(self, ob, name):
... ObjectWrapper.__init__(self, ob)
... self.name = name
... def __str__(self):
... return self.name
>>> w = NameWrapper(42, "The Ultimate Answer")
>>> w
42
>>> print(w)
The Ultimate Answer
>>> w * 2
84
>>> w.name
'The Ultimate Answer'
Notice that any attributes you add must be defined in the class. You can't add arbitrary attributes at runtime, because they'll be set on the wrapped object instead of the wrapper:
>>> w.foo = '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.
LazyWrapper
objects are particularly useful when working with expensive resources, like connections or web browsers, to avoid their creation unless absolutely needed.
However resources usually must be released after use by calling a "close
" method of some sort. In this case the lazy creation could be triggered just when the object is not needed anymore, by the call to close
itself. For this reason when extending LazyWrapper
these methods can be overridden with a @lazymethod
replacement:
>>> from objproxies import LazyWrapper, lazymethod
>>> class LazyCloseable(LazyWrapper):
... @lazymethod
... def tell(self):
... return 0
... @lazymethod
... def close(self):
... print("bye")
... @lazymethod
... def __bool__(self):
... return False
>>> import tempfile
>>> def openf():
... print("called")
... return tempfile.TemporaryFile('w')
>>> lazyfile = LazyCloseable(openf)
>>> lazyfile.tell()
0
>>> lazyfile.close()
bye
>>> bool(lazyfile)
False
>>> lazyfile = LazyCloseable(openf)
>>> lazyfile.write('wake up')
called
7
>>> lazyfile.tell()
7
>>> lazyfile.close() # close for real
>>> bool(lazyfile)
True
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 objproxies import AbstractWrapper
>>> class NamedMixin(AbstractWrapper):
... __slots__ = []
... name = None
... def __init__(self, ob, name):
... super(NamedMixin, self).__init__(ob)
... self.name = name
... def __str__(self):
... return self.name
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 objproxies 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
called
42
>>> n
called
42
>>> n = NamedLazy(callback, "Once")
>>> n
called
42
>>> n
42
Both the AbstractProxy
and AbstractWrapper
base classes work by assuming that self.__subject__
will be the wrapped or proxed 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__
.