sentinel 0.1.2

Create sentinel and singleton objects

Creates simple sentinel objects which are the only instance of their own anonymous class. As a singleton, there is a guarantee that there will only ever be one instance: they can be safely used with pickle and cPickle alike, as well as being able to be used properly with copy.deepcopy(). In addition, a self-documenting __repr__ is provided for free!

Usage

Sentinels are singleton objects that typically represent some end or terminating condition. Some singletons already exist in Python, like None NotImplemented, and Ellipsis.

All that’s needed to create a sentinel is its name:

>>> import sentinel
>>> Nothing = sentinel.create('Nothing')
>>> Nothing
Nothing

This by itself is useful when other objects such as None, False, 0, -1, etc. are entirely valid values. For example, setting default values when all other values are valid with: dict.setdefault():

>>> MissingEntry = sentinel.create('MissingEntry')
>>> d = {'stdout': None, 'stdin': 0, 'EOF': -1}
>>> [d.setdefault(key, MissingEntry) for key in ('stdin', 'stdout', 'stderr')]
[0, None, MissingEntry]

Alternatively, using dict.get() when fetching values:

>>> d = {'stdout': None, 'stdin': 0, 'EOF': -1}
>>> d.get('stdout', MissingEntry)
None
>>> d.get('stdin', MissingEntry)
0
>>> d.get('stderr', MissingEntry)
MissingEntry

It’s known immediately which value was missing from the dictionary in a self-documenting manner.

Adding extra methods and class attributes

Sentinels may also inherit from base classes, or implement extra methods.

Consider a binary search tree with two kinds of nodes: interior nodes (Node) which contain some payload and leaves (Leaf), which simply terminate traversal.

To create singleton leaf which implements a search method and an is_leaf property, you may provide any extra class attributes in the extra_methods keyword argument. The following is a full example of both the singleton Leaf and its Node counterpart:

def _search_leaf(self, key):
    raise KeyError(key)

Leaf = sentinel.create('Leaf', extra_methods={
    'search': _search_leaf,
    'is_leaf': property(lambda self: True)
})

class Node(object):
    def __init__(self, key, payload, left=Leaf, right=Leaf):
        self.left = left
        self.right = right
        self.key = key
        self.payload = payload

    def search(self, key):
        if key < self.__key:
            return self.left.search(key)
        elif key > self.key:
            return self.right.search(key)
        else:
            return self.payload

    is_leaf = property(lambda: false)

Example usage:

>>> tree = Node(2, 'bar', Node(1, 'foo'), Node(3, 'baz'))
>>> tree.search(1)
'foo'
>>> tree.search(4)
Traceback (most recent call last):
    ...
KeyError: 2

Inheriting from a base class

Another usage is inheriting from a tuple, in order to do tuple comparison. For example, consider a scenario where a certain order must be maintained, but ordering matters. If the key being used to sort is an integer, a plain object instance will always sort greater (in Python 2—see below for Python 3 fix):

>>> (1, [], []) < (object(), None, None)
True

Now say we want to encode this in a neat, self-documenting package. This is can be done by create a sentinel that inherits from tuple and is instantiated with the given tuple:

arg = (object(), None, None)
AlwaysGreater = sentinel.create('AlwaysGreater', (tuple,), {}, args)

This will call tuple((object(), None, None)). This means the singleton will now behave exactly as expected:

>>> (1, [], []) < AlwaysGreater
True

Python 3 fix

An int and any old object are no longer comparable in Python 3:

>>> (1, ..., ...) < (object(), None, None)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unorderable types: int() < object()

This makes the above example more difficult. Luckily, sentinels can easily fix this. Creating a sentinel that is always less than any number:

IntInfinity = sentinel.create('IntInfinity', (int,), extra_methods={
    '__lt__': lambda self, other: False,
    '__gt__': lambda self, other: True,
    '__ge__': lambda self, other: True,
    '__le__': lambda self, other: True if self is other else False
})

Since we inherit from int, it is, for all intents and purposes, an int:

>>> isinstance(MinInf, int)
True
>>> IntInfinity > 10 ** 1000
True
>>> 10 ** 1000 > IntInfinity
False

Note that if not provided any explicit instantiation, it is equal to 0:

>>> IntInfinity == 0
True
>>> bool(IntInfinity)
False
>>> IntInfinity + 8
8

Nonetheless, it serves its purpose in our example:

arg = (IntInfinity, None, None)
AlwaysGreater = sentinel.create('AlwaysGreater', (tuple,), {}, arg)

Usage:

>>> (1, ..., ...) < AlwaysGreater
True
>>> AlwaysGreater < (1, ..., ...)
False