We Keep Coding

Using Python Metaclasses to Limit the Number of Attributes

I trying to define a Python metaclass to limit the number of attributes that a class may contain, starting at creation time. I am constrained to use Python 3.7 due to system limitations
I managed to do it without a metaclass:
class MyClass:
def __init__(self):
self.param1 = 7
self.param2 = 8
def __setattr__(self, key, value):
if key not in self.__dict__ or len(self.__dict__.keys()) < 2:
self.__dict__[key] = value
raise ValueError("Excess of parameters.")
test = MyClass()
test.param1 = 3
test.param2 = 5
print(test.param1) # Prints 3
print(test.param2) # Prints 5
test.param3 = 9 # RAISES AN ERROR, AS IT SHOULD!
The problem that I am having is that I want to do this through a metaclass. Sure, I bet that there can be better options, but I find that, in this case, the best option is to do it through a metaclass (?)
I have read several posts and this is the closest Q&A I have found to solving the problem.
Question
Is there a way to limit the number of attributes of a class to a specific number, at creation time, using metaclasses?
Maybe use the dunder method __settattr__ to use the super class constructor?

Since your needs are actually "It is more a policy of my co-workers not being able to meddle with the number of attributes of memory sensitive classes, where adding more attributes at run time (or even develop time) may place unneeded stress on the system (kind of an API), so I want them to have the flexibility to, at the very least, choose the names of the variables but limit them to a certain number"
You actually need to use __slots__. You can use the metaclass to inspect some parameter of the class at creation time - possibly the parameters of the __init__ method, and create the names from there.
The metaclass can also inject a __setattr__ method with whatever limitations you want, of course - but without the use of __slots__, each instance will have a full __dict__ for the attributes, which will (1) use memory, and (2) could be used with a workaround to store extra attributes.
Since knowing the attribute names to set them as slots would be an extra task, and there should be some definition (for example, we could use the first two parameter names to __init__), I think it is easier to inject generic slot names, and use __setattr__ to map arbitrarily named attributes to the slots in the order they are assigned. The mapping itself can be recorded on the class object by the setattr, and once the first instance is fully initiated, no other attributes can be added in other instances.
Still - I am providing an example for this for the fun of it - from the point of view of "real world" usage, your concerns are not useful at all: if attributes are needed for a certain business logic, they are needed - trying to artificially limit them won't help. (they could just add a dict as one of the attributes. If you bar that, then use another container. Continue to try limting the code and it could degrade to the point of using raw lists, dictionaries and stand alone functions instead of being object oriented, and so on.
def limit_setattr(self, name, value):
cls = type(self)
# Acessing the attribute through "cls.__dict__"
# prevent it from being fetched from a superclass.
attrs = cls.__dict__["attr_names"]
# Allow the use of properties, without then counting to the maximum attributes:
if hasattr(cls, name) and hasattr(getattr(cls, name), "__set__"): # nice place to use the walrus - := - op, but that is supported from Python 3.8 on only
return object.__setattr__(self, name, value)
try:
index = attrs.index(name)
except ValueError:
index = None
if index is None and len(attrs) < cls.maxattrs:
index = len(attrs)
attrs += (name,)
cls.attr_names = attrs
elif index is None:
raise AttributeError(f"Class {cls.__name__} can't hold attribute {name}: max attributes exceeded!")
mapped_name = f"attr_{index}"
object.__setattr__(self, mapped_name, value)
def limit_getattr(self, name):
cls = type(self)
attrs = cls.__dict__["attr_names"]
try:
index = attrs.index(name)
except ValueError:
raise AttributeError(f"{name} not an attribute of {cls.__name__} instances" )
mapped_name = f"attr_{index}"
return object.__getattribute__(self, mapped_name)
class LimitAttrs(type):
def __new__(mcls, name, bases, namespace, maxattrs=2):
for base in bases:
if "__dict__" in dir(base):
raise TypeError(f"The base class {base} of {name} does not have slots and won't work for a slotted child class.")
namespace["__setattr__"] = limit_setattr
namespace["__getattribute__"] = limit_getattr
namespace["maxattrs"] = maxattrs
namespace["attr_names"] = tuple()
namespace["__slots__"] = tuple(f"attr_{i}" for i in range(maxattrs))
return super().__new__(mcls, name, bases, namespace)
And here is the code above being used in an interactive session:
In [81]: class A(metaclass=LimitAttrs):
...: pass
...:
In [82]: a = A()
In [83]: a.aa = 1
In [84]: a.bb = 2
In [85]: a.cc = 3
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
[...]
AttributeError: Class A can't hold attribute cc: max attributes exceeded!
In [86]: b = A()
In [87]: b.aa
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
Input In [87], in <cell line: 1>()
----> 1 b.aa
[...]
AttributeError: ...
In [88]: b.aa = 3
In [89]: b.cc = 4
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
Input In [89], in <cell line: 1>()
[...]
AttributeError: Class A can't hold attribute cc: max attributes exceeded!

If you really want to do it using metaclasses, you can, but I would recommend against it, because a class decorator suffices, and you can chain class decorators in future, whereas you cannot do so with metaclasses. Using a class decorator also frees up the possibility of using a metaclass in future.
With that in mind, here are the solutions I've cooked up. All code is tested and confirmed to function as intended on Python 3.7.3.
# A solution using a class decorator (recommended solution)
def limited_attributes(number_attributes):
def __setattr__(self,key,value):
if key not in self.__dict__ and len(self.__dict__.keys())>=number_attributes:
raise ValueError('Too many attributes')
self.__dict__[key]=value
def decorator(cls):
cls.__setattr__=__setattr__
return cls
return decorator
#limited_attributes(2)
class Foo:
def __init__(self):
self.param1=4
self.param2=5
test = Foo()
test.param1=3 # no error
test.param2='bar' # no error
test.param3='baz' # raises ValueError
# A solution using a function as a metaclass
def limited_attributes(number_attributes):
def __setattr__(self,key,value):
if key not in self.__dict__ and len(self.__dict__.keys())>=number_attributes:
raise ValueError('Too many attributes')
self.__dict__[key]=value
def metaclass(name,bases,attrs):
cls=type(name,bases,attrs)
cls.__setattr__=__setattr__
return cls
return metaclass
class Foo(metaclass=limited_attributes(2)):
def __init__(self):
self.param1=4
self.param2=5
test = Foo()
test.param1=3 # no error
test.param2='bar' # no error
test.param3='baz' # raises ValueError

Your existing code has some problems, indeed it should be written as:
class MyClass:
def __init__(self):
self.param1 = 7
self.param2 = 8
def __setattr__(self, key, value):
n_attr_max = 2
if key in self.__dict__ or len(self.__dict__.keys())+1 <= n_attr_max:
self.__dict__[key] = value
else:
raise ValueError("Excess of parameters.")
test = MyClass()
test.param1 = 3
test.param2 = 5
print(test.param1) # Prints 3
print(test.param2) # Prints 5
test.param3 = 9 # Raises exception here.
To override metaclass dunder methods you can pass new method through the dictionary. The following will do what you want:
class MyMetaClass(type):
def __new__(cls, name, bases, dic):
#
def settattr_modified(self, key, value):
n_attr_max = 2
if key in self.__dict__ or len(self.__dict__.keys())+1 <= n_attr_max:
self.__dict__[key] = value
else:
raise ValueError("Excess of parameters.")
#
dic["__setattr__"] = settattr_modified
return super().__new__(cls, name, bases, dic)
class MyClass(metaclass= MyMetaClass):
def __init__(self):
self.param1 = 7
self.param2 = 8
test = MyClass()
test.param1 = 3
test.param2 = 5
print(test.param1) # Prints 3
print(test.param2) # Prints 5
test.param3 = 9 # Raises exception here.

which one is preferable to use in oop: property definition or method definition [duplicate]

I'm doing it like:
def set_property(property,value):
def get_property(property):
or
object.property = value
value = object.property
What's the pythonic way to use getters and setters?

Try this: Python Property
The sample code is:
class C(object):
def __init__(self):
self._x = None
#property
def x(self):
"""I'm the 'x' property."""
print("getter of x called")
return self._x
#x.setter
def x(self, value):
print("setter of x called")
self._x = value
#x.deleter
def x(self):
print("deleter of x called")
del self._x
c = C()
c.x = 'foo' # setter called
foo = c.x # getter called
del c.x # deleter called

What's the pythonic way to use getters and setters?
The "Pythonic" way is not to use "getters" and "setters", but to use plain attributes, like the question demonstrates, and del for deleting (but the names are changed to protect the innocent... builtins):
value = 'something'
obj.attribute = value
value = obj.attribute
del obj.attribute
If later, you want to modify the setting and getting, you can do so without having to alter user code, by using the property decorator:
class Obj:
"""property demo"""
#
#property # first decorate the getter method
def attribute(self): # This getter method name is *the* name
return self._attribute
#
#attribute.setter # the property decorates with `.setter` now
def attribute(self, value): # name, e.g. "attribute", is the same
self._attribute = value # the "value" name isn't special
#
#attribute.deleter # decorate with `.deleter`
def attribute(self): # again, the method name is the same
del self._attribute
(Each decorator usage copies and updates the prior property object, so note that you should use the same name for each set, get, and delete function/method.)
After defining the above, the original setting, getting, and deleting code is the same:
obj = Obj()
obj.attribute = value
the_value = obj.attribute
del obj.attribute
You should avoid this:
def set_property(property,value):
def get_property(property):
Firstly, the above doesn't work, because you don't provide an argument for the instance that the property would be set to (usually self), which would be:
class Obj:
def set_property(self, property, value): # don't do this
...
def get_property(self, property): # don't do this either
...
Secondly, this duplicates the purpose of two special methods, __setattr__ and __getattr__.
Thirdly, we also have the setattr and getattr builtin functions.
setattr(object, 'property_name', value)
getattr(object, 'property_name', default_value) # default is optional
The #property decorator is for creating getters and setters.
For example, we could modify the setting behavior to place restrictions the value being set:
class Protective(object):
#property
def protected_value(self):
return self._protected_value
#protected_value.setter
def protected_value(self, value):
if acceptable(value): # e.g. type or range check
self._protected_value = value
In general, we want to avoid using property and just use direct attributes.
This is what is expected by users of Python. Following the rule of least-surprise, you should try to give your users what they expect unless you have a very compelling reason to the contrary.
Demonstration
For example, say we needed our object's protected attribute to be an integer between 0 and 100 inclusive, and prevent its deletion, with appropriate messages to inform the user of its proper usage:
class Protective(object):
"""protected property demo"""
#
def __init__(self, start_protected_value=0):
self.protected_value = start_protected_value
#
#property
def protected_value(self):
return self._protected_value
#
#protected_value.setter
def protected_value(self, value):
if value != int(value):
raise TypeError("protected_value must be an integer")
if 0 <= value <= 100:
self._protected_value = int(value)
else:
raise ValueError("protected_value must be " +
"between 0 and 100 inclusive")
#
#protected_value.deleter
def protected_value(self):
raise AttributeError("do not delete, protected_value can be set to 0")
(Note that __init__ refers to self.protected_value but the property methods refer to self._protected_value. This is so that __init__ uses the property through the public API, ensuring it is "protected".)
And usage:
>>> p1 = Protective(3)
>>> p1.protected_value
3
>>> p1 = Protective(5.0)
>>> p1.protected_value
5
>>> p2 = Protective(-5)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "<stdin>", line 3, in __init__
File "<stdin>", line 15, in protected_value
ValueError: protectected_value must be between 0 and 100 inclusive
>>> p1.protected_value = 7.3
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "<stdin>", line 17, in protected_value
TypeError: protected_value must be an integer
>>> p1.protected_value = 101
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "<stdin>", line 15, in protected_value
ValueError: protectected_value must be between 0 and 100 inclusive
>>> del p1.protected_value
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "<stdin>", line 18, in protected_value
AttributeError: do not delete, protected_value can be set to 0
Do the names matter?
Yes they do. .setter and .deleter make copies of the original property. This allows subclasses to properly modify behavior without altering the behavior in the parent.
class Obj:
"""property demo"""
#
#property
def get_only(self):
return self._attribute
#
#get_only.setter
def get_or_set(self, value):
self._attribute = value
#
#get_or_set.deleter
def get_set_or_delete(self):
del self._attribute
Now for this to work, you have to use the respective names:
obj = Obj()
# obj.get_only = 'value' # would error
obj.get_or_set = 'value'
obj.get_set_or_delete = 'new value'
the_value = obj.get_only
del obj.get_set_or_delete
# del obj.get_or_set # would error
I'm not sure where this would be useful, but the use-case is if you want a get, set, and/or delete-only property. Probably best to stick to semantically same property having the same name.
Conclusion
Start with simple attributes.
If you later need functionality around the setting, getting, and deleting, you can add it with the property decorator.
Avoid functions named set_... and get_... - that's what properties are for.

In [1]: class test(object):
def __init__(self):
self.pants = 'pants'
#property
def p(self):
return self.pants
#p.setter
def p(self, value):
self.pants = value * 2
....:
In [2]: t = test()
In [3]: t.p
Out[3]: 'pants'
In [4]: t.p = 10
In [5]: t.p
Out[5]: 20

Using #property and #attribute.setter helps you to not only use the "pythonic" way but also to check the validity of attributes both while creating the object and when altering it.
class Person(object):
def __init__(self, p_name=None):
self.name = p_name
#property
def name(self):
return self._name
#name.setter
def name(self, new_name):
if type(new_name) == str: #type checking for name property
self._name = new_name
else:
raise Exception("Invalid value for name")
By this, you actually 'hide' _name attribute from client developers and also perform checks on name property type. Note that by following this approach even during the initiation the setter gets called. So:
p = Person(12)
Will lead to:
Exception: Invalid value for name
But:
>>>p = person('Mike')
>>>print(p.name)
Mike
>>>p.name = 'George'
>>>print(p.name)
George
>>>p.name = 2.3 # Causes an exception

This is an old question but the topic is very important and always current. In case anyone wants to go beyond simple getters/setters i have wrote an article about superpowered properties in python with support for slots, observability and reduced boilerplate code.
from objects import properties, self_properties
class Car:
with properties(locals(), 'meta') as meta:
#meta.prop(read_only=True)
def brand(self) -> str:
"""Brand"""
#meta.prop(read_only=True)
def max_speed(self) -> float:
"""Maximum car speed"""
#meta.prop(listener='_on_acceleration')
def speed(self) -> float:
"""Speed of the car"""
return 0 # Default stopped
#meta.prop(listener='_on_off_listener')
def on(self) -> bool:
"""Engine state"""
return False
def __init__(self, brand: str, max_speed: float = 200):
self_properties(self, locals())
def _on_off_listener(self, prop, old, on):
if on:
print(f"{self.brand} Turned on, Runnnnnn")
else:
self._speed = 0
print(f"{self.brand} Turned off.")
def _on_acceleration(self, prop, old, speed):
if self.on:
if speed > self.max_speed:
print(f"{self.brand} {speed}km/h Bang! Engine exploded!")
self.on = False
else:
print(f"{self.brand} New speed: {speed}km/h")
else:
print(f"{self.brand} Car is off, no speed change")
This class can be used like this:
mycar = Car('Ford')
# Car is turned off
for speed in range(0, 300, 50):
mycar.speed = speed
# Car is turned on
mycar.on = True
for speed in range(0, 350, 50):
mycar.speed = speed
This code will produce the following output:
Ford Car is off, no speed change
Ford Car is off, no speed change
Ford Car is off, no speed change
Ford Car is off, no speed change
Ford Car is off, no speed change
Ford Car is off, no speed change
Ford Turned on, Runnnnnn
Ford New speed: 0km/h
Ford New speed: 50km/h
Ford New speed: 100km/h
Ford New speed: 150km/h
Ford New speed: 200km/h
Ford 250km/h Bang! Engine exploded!
Ford Turned off.
Ford Car is off, no speed change
More info about how and why here: https://mnesarco.github.io/blog/2020/07/23/python-metaprogramming-properties-on-steroids

Properties are pretty useful since you can use them with assignment but then can include validation as well. You can see this code where you use the decorator #property and also #<property_name>.setter to create the methods:
# Python program displaying the use of #property
class AgeSet:
def __init__(self):
self._age = 0
# using property decorator a getter function
#property
def age(self):
print("getter method called")
return self._age
# a setter function
#age.setter
def age(self, a):
if(a < 18):
raise ValueError("Sorry your age is below eligibility criteria")
print("setter method called")
self._age = a
pkj = AgeSet()
pkj.age = int(input("set the age using setter: "))
print(pkj.age)
There are more details in this post I wrote about this as well: https://pythonhowtoprogram.com/how-to-create-getter-setter-class-properties-in-python-3/

You can use accessors/mutators (i.e. #attr.setter and #property) or not, but the most important thing is to be consistent!
If you're using #property to simply access an attribute, e.g.
class myClass:
def __init__(a):
self._a = a
#property
def a(self):
return self._a
use it to access every* attribute! It would be a bad practice to access some attributes using #property and leave some other properties public (i.e. name without an underscore) without an accessor, e.g. do not do
class myClass:
def __init__(a, b):
self.a = a
self.b = b
#property
def a(self):
return self.a
Note that self.b does not have an explicit accessor here even though it's public.
Similarly with setters (or mutators), feel free to use #attribute.setter but be consistent! When you do e.g.
class myClass:
def __init__(a, b):
self.a = a
self.b = b
#a.setter
def a(self, value):
return self.a = value
It's hard for me to guess your intention. On one hand you're saying that both a and b are public (no leading underscore in their names) so I should theoretically be allowed to access/mutate (get/set) both. But then you specify an explicit mutator only for a, which tells me that maybe I should not be able to set b. Since you've provided an explicit mutator I am not sure if the lack of explicit accessor (#property) means I should not be able to access either of those variables or you were simply being frugal in using #property.
*The exception is when you explicitly want to make some variables accessible or mutable but not both or you want to perform some additional logic when accessing or mutating an attribute. This is when I am personally using #property and #attribute.setter (otherwise no explicit acessors/mutators for public attributes).
Lastly, PEP8 and Google Style Guide suggestions:
PEP8, Designing for Inheritance says:
For simple public data attributes, it is best to expose just the attribute name, without complicated accessor/mutator methods. Keep in mind that Python provides an easy path to future enhancement, should you find that a simple data attribute needs to grow functional behavior. In that case, use properties to hide functional implementation behind simple data attribute access syntax.
On the other hand, according to Google Style Guide Python Language Rules/Properties the recommendation is to:
Use properties in new code to access or set data where you would normally have used simple, lightweight accessor or setter methods. Properties should be created with the #property decorator.
The pros of this approach:
Readability is increased by eliminating explicit get and set method calls for simple attribute access. Allows calculations to be lazy. Considered the Pythonic way to maintain the interface of a class. In terms of performance, allowing properties bypasses needing trivial accessor methods when a direct variable access is reasonable. This also allows accessor methods to be added in the future without breaking the interface.
and cons:
Must inherit from object in Python 2. Can hide side-effects much like operator overloading. Can be confusing for subclasses.

You can use the magic methods __getattribute__ and __setattr__.
class MyClass:
def __init__(self, attrvalue):
self.myattr = attrvalue
def __getattribute__(self, attr):
if attr == "myattr":
#Getter for myattr
def __setattr__(self, attr):
if attr == "myattr":
#Setter for myattr
Be aware that __getattr__ and __getattribute__ are not the same. __getattr__ is only invoked when the attribute is not found.

Python: Class private attributes [duplicate]

I am generally confused about the difference between a "property" and an "attribute", and can't find a great resource to concisely detail the differences.

Properties are a special kind of attribute. Basically, when Python encounters the following code:
spam = SomeObject()
print(spam.eggs)
it looks up eggs in spam, and then examines eggs to see if it has a __get__, __set__, or __delete__ method — if it does, it's a property. If it is a property, instead of just returning the eggs object (as it would for any other attribute) it will call the __get__ method (since we were doing lookup) and return whatever that method returns.
More information about Python's data model and descriptors.

With a property you have complete control on its getter, setter and deleter methods, which you don't have (if not using caveats) with an attribute.
class A(object):
_x = 0
'''A._x is an attribute'''
#property
def x(self):
'''
A.x is a property
This is the getter method
'''
return self._x
#x.setter
def x(self, value):
"""
This is the setter method
where I can check it's not assigned a value < 0
"""
if value < 0:
raise ValueError("Must be >= 0")
self._x = value
>>> a = A()
>>> a._x = -1
>>> a.x = -1
Traceback (most recent call last):
File "ex.py", line 15, in <module>
a.x = -1
File "ex.py", line 9, in x
raise ValueError("Must be >= 0")
ValueError: Must be >= 0

In general speaking terms a property and an attribute are the same thing. However, there is a property decorator in Python which provides getter/setter access to an attribute (or other data).
class MyObject(object):
# This is a normal attribute
foo = 1
#property
def bar(self):
return self.foo
#bar.setter
def bar(self, value):
self.foo = value
obj = MyObject()
assert obj.foo == 1
assert obj.bar == obj.foo
obj.bar = 2
assert obj.foo == 2
assert obj.bar == obj.foo

The property allows you to get and set values like you would normal attributes, but underneath there is a method being called translating it into a getter and setter for you. It's really just a convenience to cut down on the boilerplate of calling getters and setters.
Lets say for example, you had a class that held some x and y coordinates for something you needed. To set them you might want to do something like:
myObj.x = 5
myObj.y = 10
That is much easier to look at and think about than writing:
myObj.setX(5)
myObj.setY(10)
The problem is, what if one day your class changes such that you need to offset your x and y by some value? Now you would need to go in and change your class definition and all of the code that calls it, which could be really time consuming and error prone. The property allows you to use the former syntax while giving you the flexibility of change of the latter.
In Python, you can define getters, setters, and delete methods with the property function. If you just want the read property, there is also a #property decorator you can add above your method.
http://docs.python.org/library/functions.html#property

I learnt 2 differences from site of Bernd Klein, in summary:
1. A property is a more convenient way to achieve data encapsulation
For example, let's say you have a public attribute length. Later on, your project requires you to encapsulate it, i.e. to change it to private and provide a getter and setter => you have to change the the code you wrote before:
# Old code
obj1.length = obj1.length + obj2.length
# New code (using private attributes and getter and setter)
obj1.set_length(obj1.get_length() + obj2.get_length()) # => this is ugly
If you use #property and #length.setter => you don't need to change that old code.
2. A property can encapsulate multiple attributes
class Person:
def __init__(self, name, physic_health, mental_health):
self.name = name
self.__physic_health = physic_health
self.__mental_health = mental_health
#property
def condition(self):
health = self.__physic_health + self.__mental_health
if(health < 5.0):
return "I feel bad!"
elif health < 8.0:
return "I am ok!"
else:
return "Great!"
In this example, __physic_health and __mental_health are private and cannot be accessed directly from outside.

There is also one not obvious difference that i use to cache or refresh data , often we have a function connected to class attribute. For instance i need to read file once and keep content assigned to the attribute so the value is cached:
class Misc():
def __init__(self):
self.test = self.test_func()
def test_func(self):
print 'func running'
return 'func value'
cl = Misc()
print cl.test
print cl.test
Output:
func running
func value
func value
We accessed the attribute twice but our function was fired only once. Changing the above example to use property will cause attribute's value refresh each time you access it:
class Misc():
#property
def test(self):
print 'func running'
return 'func value'
cl = Misc()
print cl.test
print cl.test
Output:
func running
func value
func running
func value

I like to think that, if you want to set a restriction for an attribute, use a property.
Although all attributes are public, generally programmers differentiate public and private attributes with an underscore(_). Consider the following class,
class A:
def __init__(self):
self.b = 3 # To show public
self._c = 4 # To show private
Here, b attribute is intended to be accessed from outside class A. But, readers of this class might wonder, can b attribute be set from outside class A?
If we intend to not set b from outside, we can show this intention with #property.
class A:
def __init__(self):
self._c = 4 # To show private
#property
def b(self):
return 3
Now, b can't be set.
a = A()
print(a.b) # prints 3
a.b = 7 # Raises AttributeError
Or, if you wish to set only certain values,
class A:
#property
def b(self):
return self._b
#b.setter
def b(self, val):
if val < 0:
raise ValueError("b can't be negative")
self._b = val
a = A()
a.b = 6 # OK
a.b = -5 # Raises ValueError

What's the difference between a Python "property" and "attribute"?

I am generally confused about the difference between a "property" and an "attribute", and can't find a great resource to concisely detail the differences.

Properties are a special kind of attribute. Basically, when Python encounters the following code:
spam = SomeObject()
print(spam.eggs)
it looks up eggs in spam, and then examines eggs to see if it has a __get__, __set__, or __delete__ method — if it does, it's a property. If it is a property, instead of just returning the eggs object (as it would for any other attribute) it will call the __get__ method (since we were doing lookup) and return whatever that method returns.
More information about Python's data model and descriptors.

With a property you have complete control on its getter, setter and deleter methods, which you don't have (if not using caveats) with an attribute.
class A(object):
_x = 0
'''A._x is an attribute'''
#property
def x(self):
'''
A.x is a property
This is the getter method
'''
return self._x
#x.setter
def x(self, value):
"""
This is the setter method
where I can check it's not assigned a value < 0
"""
if value < 0:
raise ValueError("Must be >= 0")
self._x = value
>>> a = A()
>>> a._x = -1
>>> a.x = -1
Traceback (most recent call last):
File "ex.py", line 15, in <module>
a.x = -1
File "ex.py", line 9, in x
raise ValueError("Must be >= 0")
ValueError: Must be >= 0

In general speaking terms a property and an attribute are the same thing. However, there is a property decorator in Python which provides getter/setter access to an attribute (or other data).
class MyObject(object):
# This is a normal attribute
foo = 1
#property
def bar(self):
return self.foo
#bar.setter
def bar(self, value):
self.foo = value
obj = MyObject()
assert obj.foo == 1
assert obj.bar == obj.foo
obj.bar = 2
assert obj.foo == 2
assert obj.bar == obj.foo

The property allows you to get and set values like you would normal attributes, but underneath there is a method being called translating it into a getter and setter for you. It's really just a convenience to cut down on the boilerplate of calling getters and setters.
Lets say for example, you had a class that held some x and y coordinates for something you needed. To set them you might want to do something like:
myObj.x = 5
myObj.y = 10
That is much easier to look at and think about than writing:
myObj.setX(5)
myObj.setY(10)
The problem is, what if one day your class changes such that you need to offset your x and y by some value? Now you would need to go in and change your class definition and all of the code that calls it, which could be really time consuming and error prone. The property allows you to use the former syntax while giving you the flexibility of change of the latter.
In Python, you can define getters, setters, and delete methods with the property function. If you just want the read property, there is also a #property decorator you can add above your method.
http://docs.python.org/library/functions.html#property

I learnt 2 differences from site of Bernd Klein, in summary:
1. A property is a more convenient way to achieve data encapsulation
For example, let's say you have a public attribute length. Later on, your project requires you to encapsulate it, i.e. to change it to private and provide a getter and setter => you have to change the the code you wrote before:
# Old code
obj1.length = obj1.length + obj2.length
# New code (using private attributes and getter and setter)
obj1.set_length(obj1.get_length() + obj2.get_length()) # => this is ugly
If you use #property and #length.setter => you don't need to change that old code.
2. A property can encapsulate multiple attributes
class Person:
def __init__(self, name, physic_health, mental_health):
self.name = name
self.__physic_health = physic_health
self.__mental_health = mental_health
#property
def condition(self):
health = self.__physic_health + self.__mental_health
if(health < 5.0):
return "I feel bad!"
elif health < 8.0:
return "I am ok!"
else:
return "Great!"
In this example, __physic_health and __mental_health are private and cannot be accessed directly from outside.

There is also one not obvious difference that i use to cache or refresh data , often we have a function connected to class attribute. For instance i need to read file once and keep content assigned to the attribute so the value is cached:
class Misc():
def __init__(self):
self.test = self.test_func()
def test_func(self):
print 'func running'
return 'func value'
cl = Misc()
print cl.test
print cl.test
Output:
func running
func value
func value
We accessed the attribute twice but our function was fired only once. Changing the above example to use property will cause attribute's value refresh each time you access it:
class Misc():
#property
def test(self):
print 'func running'
return 'func value'
cl = Misc()
print cl.test
print cl.test
Output:
func running
func value
func running
func value

I like to think that, if you want to set a restriction for an attribute, use a property.
Although all attributes are public, generally programmers differentiate public and private attributes with an underscore(_). Consider the following class,
class A:
def __init__(self):
self.b = 3 # To show public
self._c = 4 # To show private
Here, b attribute is intended to be accessed from outside class A. But, readers of this class might wonder, can b attribute be set from outside class A?
If we intend to not set b from outside, we can show this intention with #property.
class A:
def __init__(self):
self._c = 4 # To show private
#property
def b(self):
return 3
Now, b can't be set.
a = A()
print(a.b) # prints 3
a.b = 7 # Raises AttributeError
Or, if you wish to set only certain values,
class A:
#property
def b(self):
return self._b
#b.setter
def b(self, val):
if val < 0:
raise ValueError("b can't be negative")
self._b = val
a = A()
a.b = 6 # OK
a.b = -5 # Raises ValueError

Polluting a class's environment

I have an object that holds lots of ids that are accessed statically. I want to split that up into another object which holds only those ids without the need of making modifications to the already existen code base. Take for example:
class _CarType(object):
DIESEL_CAR_ENGINE = 0
GAS_CAR_ENGINE = 1 # lots of these ids
class Car(object):
types = _CarType
I want to be able to access _CarType.DIESEL_CAR_ENGINE either by calling Car.types.DIESEL_CAR_ENGINE, either by Car.DIESEL_CAR_ENGINE for backwards compatibility with the existent code. It's clear that I cannot use __getattr__ so I am trying to find a way of making this work (maybe metaclasses ? )

Although this is not exactly what subclassing is made for, it accomplishes what you describe:
class _CarType(object):
DIESEL_CAR_ENGINE = 0
GAS_CAR_ENGINE = 1 # lots of these ids
class Car(_CarType):
types = _CarType

Something like:
class Car(object):
for attr, value in _CarType.__dict__.items():
it not attr.startswith('_'):
locals()[attr] = value
del attr, value
Or you can do it out of the class declaration:
class Car(object):
# snip
for attr, value in _CarType.__dict__.items():
it not attr.startswith('_'):
setattr(Car, attr, value)
del attr, value

This is how you could do this with a metaclass:
class _CarType(type):
DIESEL_CAR_ENGINE = 0
GAS_CAR_ENGINE = 1 # lots of these ids
def __init__(self,name,bases,dct):
for key in dir(_CarType):
if key.isupper():
setattr(self,key,getattr(_CarType,key))
class Car(object):
__metaclass__=_CarType
print(Car.DIESEL_CAR_ENGINE)
print(Car.GAS_CAR_ENGINE)

Your options fall into two substantial categories: you either copy the attributes from _CarType into Car, or set Car's metaclass to a custom one with a __getattr__ method that delegates to _CarType (so it isn't exactly true that you can't use __getattr__: you can, you just need to put in in Car's metaclass rather than in Car itself;-).
The second choice has implications that you might find peculiar (unless they are specifically desired): the attributes don't show up on dir(Car), and they can't be accessed on an instance of Car, only on Car itself. I.e.:
>>> class MetaGetattr(type):
... def __getattr__(cls, nm):
... return getattr(cls.types, nm)
...
>>> class Car:
... __metaclass__ = MetaGetattr
... types = _CarType
...
>>> Car.GAS_CAR_ENGINE
1
>>> Car().GAS_CAR_ENGINE
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
AttributeError: 'Car' object has no attribute 'GAS_CAR_ENGINE'
You could fix the "not from an instance" issue by also adding a __getattr__ to Car:
>>> class Car:
... __metaclass__ = MetaGetattr
... types = _CarType
... def __getattr__(self, nm):
... return getattr(self.types, nm)
...
to make both kinds of lookup work, as is probably expected:
>>> Car.GAS_CAR_ENGINE
1
>>> Car().GAS_CAR_ENGINE
1
However, defining two, essentially-equal __getattr__s, doesn't seem elegant.
So I suspect that the simpler approach, "copy all attributes", is preferable. In Python 2.6 or better, this is an obvious candidate for a class decorator:
def typesfrom(typesclass):
def decorate(cls):
cls.types = typesclass
for n in dir(typesclass):
if n[0] == '_': continue
v = getattr(typesclass, n)
setattr(cls, n, v)
return cls
return decorate
#typesfrom(_CarType)
class Car(object):
pass
In general, it's worth defining a decorator if you're using it more than once; if you only need to perform this task for one class ever, then expanding the code inline instead (after the class statement) may be better.
If you're stuck with Python 2.5 (or even 2.4), you can still define typesfrom the same way, you just apply it in a slightly less elegant matter, i.e., the Car definition becomes:
class Car(object):
pass
Car = typesfrom(_CarType)(Car)
Do remember decorator syntax (introduced in 2.2 for functions, in 2.6 for classes) is just a handy way to wrap these important and frequently recurring semantics.

class _CarType(object):
DIESEL_CAR_ENGINE = 0
GAS_CAR_ENGINE = 1 # lots of these ids
class Car:
types = _CarType
def __getattr__(self, name):
return getattr(self.types, name)
If an attribute of an object is not found, and it defines that magic method __getattr__, that gets called to try to find it.
Only works on a Car instance, not on the class.