Implementation of Value Object [1] for Python.
from precious import Value, assign_attributes
class Color(Value):
@assign_attributes
def __init__(self, red, green, blue, alpha=0): pass
@property
def grayscale(self):
return (self.r + self.g + self.b) / 3
>>> red = Color(255, 0, 0)
>>> red
Color(255, 0, 0, 0)
>>> red == Color(255, 0, 0)
True
>>> Color.__slots__
('red', 'green', 'blue', 'alpha')
>>> hash(red)
8736776571231852889
pip install precious
Value Object definitions should subclass Value
. Every Value
subclass has to define attributes
, which is an iterable containing names of all attributes.
This may be achieved by explicitly setting the attribute on the class:
class Point(Value):
attributes = ('x', 'y')
def __init__(self, x, y):
self.x = x
self.y = y
Attribute names can also be extracted directly from __init__
definition using one of provided helper decorators:
from precious import Value, extract_attributes
class Point(Value):
@extract_attributes
def __init__(self, x, y):
self.x = x
self.y = y
Finally, the assign_attributes
decorator allows to reduce a common boilerplate of extracting and assigning all the attributes in __init__
:
from precious import Value, assign_attributes
class Point(Value):
@assign_attributes
def __init__(self, x, y): pass
Note that in the example above attributes are not being assigned in parent's class __init__
, thus no super()
call is required.
__eq__
__repr__
__hash__
__slots__
Value
class footprint is as small as possible. It provides only methods listed above and a single attributes
attribute.
Subclassing Value
automaticaly assignes names of all attributes to __slots__
[2].
Just run tox
in package directory:
$ tox
Namedtuple definition is equally fine for simple use cases.
Point = namedtuple('Point', ('x', 'y'))
Having to repeat the classname is a minor inconvinience, but the definition is pretty readable and concise. Also, class gets iterable interface and indexing support, which sometimes is what you want. However, things with namedtuple get very ugly when a default value or a method or a property is required. Subclassing is the only way to go. Consider the following example:
class Color(namedtuple('Color_', ('r', 'g', 'b', 'alpha'))):
__slots__ = ()
def __new__(cls, r, g, b, alpha=0):
return super().__new__(cls, r, g, b, alpha)
@property
def grayscale(self):
return (self.r + self.g + self.b) / 3
# Equivalent to
class Color(Value):
@assign_attributes
def __init__(self, red, green, blue, alpha=0): pass
@property
def grayscale(self):
return (self.r + self.g + self.b) / 3
To sum up, problems with extending namedtuple include:
- Having to define empty
__slots__
[2]. - Overriding
__new__
when a default values is required. - Repeating attributes names in several places.
- Unintuitive inheritance by generating parent class on the fly.
[1] | http://martinfowler.com/bliki/ValueObject.html |
[2] | (1, 2) https://docs.python.org/3/reference/datamodel.html#slots |