Develop Reference

Should I use `__setattr__`, a property or...?

I have an object with two attributes, file_path and save_path. Unless save_path is explicitly set, I want it to have the same value as file_path.
I think the way to do this is with __setattr__, with something like the following:
class Class():
...
def __setattr__(self, name, value):
if name == 'file_path':
self.file_path = value
self.save_path = value if self.save_path == None else self.save_path
elif name == 'save_path':
self.save_path = value
But this looks like it's going to give me infinite loops since __setattr__ is called whenever an attribute is set. So, what's the proper way to write the above and avoid that?

First, the easiest way to do this would be with a property:
class Class(object):
def __init__(self, ...):
self._save_path = None
...
#property
def save_path(self):
if self._save_path is None:
return self.file_path
else:
return self._save_path
#save_path.setter
def save_path(self, val):
self._save_path = val
Second, if you ever find yourself needing to write a __setattr__, you should use super(Class, self).__setattr__ inside your __setattr__ to bypass your __setattr__ and set attributes the normal way, avoiding infinite recursion.

this looks kind of unpythonic. You can just use attributes. Three lines of code:
>>> class Class:
... def __init__(self, file_path, save_path=None):
... self.file_path=file_path
... self.save_path = save_path or file_path
...
>>> c = Class('file')
>>> c.file_path
'file'
>>> c.save_path
'file'
>>> c1 = Class('file', 'save')
>>> c1.file_path
'file'
>>> c1.save_path
'save'
>>>

Use super!
class Class:
def __init__(self):
self.save_path = None
self.file_path = None
def __setattr__(self, name, value):
super().__setattr__(name, value)
if name == 'file_path':
super().__setattr__('save_path', self.save_path or value)
c = Class()
c.file_path = 42
print(c.file_path)
print(c.save_path)
Note that there's a limitation to this particular implementation - self.save_path needs to be called first, or it's going to fail because it hasn't been set yet when the call to super happens and it looks for self.save_path or value.
I would probably use the property based approach, personally.

Ruby like DSL in Python

I'm currently writing my first bigger project in Python, and I'm now wondering how to define a class method so that you can execute it in the class body of a subclass of the class.
First to give some more context, a slacked down (I removed everything non essential for this question) example of how I'd do the thing I'm trying to do in Ruby:
If I define a class Item like this:
class Item
def initialize(data={})
#data = data
end
def self.define_field(name)
define_method("#{name}"){ instance_variable_get("#data")[name.to_s] }
define_method("#{name}=") do |value|
instance_variable_get("#data")[name.to_s] = value
end
end
end
I can use it like this:
class MyItem < Item
define_field("name")
end
item = MyItem.new
item.name = "World"
puts "Hello #{item.name}!"
Now so far I tried achieving something similar in Python, but I'm not happy with the result I've got so far:
class ItemField(object):
def __init__(self, name):
self.name = name
def __get__(self, item, owner=None):
return item.values[self.name]
def __set__(self, item, value):
item.values[self.name] = value
def __delete__(self, item):
del item.values[self.name]
class Item(object):
def __init__(self, data=None):
if data == None: data = {}
self.values = data
for field in type(self).fields:
self.values[field.name] = None
setattr(self, field.name, field)
#classmethod
def define_field(cls, name):
if not hasattr(cls, "fields"): cls.fields = []
cls.fields.append(ItemField(name, default))
Now I don't know how I can call define_field from withing a subclass's body. This is what I wished that it was possible:
class MyItem(Item):
define_field("name")
item = MyItem({"name": "World"})
puts "Hello {}!".format(item.name)
item.name = "reader"
puts "Hello {}!".format(item.name)
There's this similar question but none of the answers are really satisfying, somebody recommends caling the function with __func__() but I guess I can't do that, because I can't get a reference to the class from within its anonymous body (please correct me if I'm wrong about this.)
Somebody else pointed out that it's better to use a module level function for doing this which I also think would be the easiest way, however the main intention of me doing this is to make the implementation of subclasses clean and having to load that module function wouldn't be to nice either. (Also I'd have to do the function call outside the class body and I don't know but I think this is messy.)
So basically I think my approach is wrong, because Python wasn't designed to allow this kind of thing to be done. What would be the best way to achieve something as in the Ruby example with Python?
(If there's no better way I've already thought about just having a method in the subclass which returns an array of the parameters for the define_field method.)

Perhaps calling a class method isn't the right route here. I'm not quite up to speed on exactly how and when Python creates classes, but my guess is that the class object doesn't yet exist when you'd call the class method to create an attribute.
It looks like you want to create something like a record. First, note that Python allows you to add attributes to your user-created classes after creation:
class Foo(object):
pass
>>> foo = Foo()
>>> foo.x = 42
>>> foo.x
42
Maybe you want to constrain which attributes the user can set. Here's one way.
class Item(object):
def __init__(self):
if type(self) is Item:
raise NotImplementedError("Item must be subclassed.")
def __setattr__(self, name, value):
if name not in self.fields:
raise AttributeError("Invalid attribute name.")
else:
self.__dict__[name] = value
class MyItem(Item):
fields = ("foo", "bar", "baz")
So that:
>>> m = MyItem()
>>> m.foo = 42 # works
>>> m.bar = "hello" # works
>>> m.test = 12 # raises AttributeError
Lastly, the above allows you the user subclass Item without defining fields, like such:
class MyItem(Item):
pass
This will result in a cryptic attribute error saying that the attribute fields could not be found. You can require that the fields attribute be defined at the time of class creation by using metaclasses. Furthermore, you can abstract away the need for the user to specify the metaclass by inheriting from a superclass that you've written to use the metaclass:
class ItemMetaclass(type):
def __new__(cls, clsname, bases, dct):
if "fields" not in dct:
raise TypeError("Subclass must define 'fields'.")
return type.__new__(cls, clsname, bases, dct)
class Item(object):
__metaclass__ = ItemMetaclass
fields = None
def __init__(self):
if type(self) == Item:
raise NotImplementedError("Must subclass Type.")
def __setattr__(self, name, value):
if name in self.fields:
self.__dict__[name] = value
else:
raise AttributeError("The item has no such attribute.")
class MyItem(Item):
fields = ("one", "two", "three")

You're almost there! If I understand you correctly:
class Item(object):
def __init__(self, data=None):
fields = data or {}
for field, value in data.items():
if hasattr(self, field):
setattr(self, field, value)
#classmethod
def define_field(cls, name):
setattr(cls, name, None)
EDIT: As far as I know, it's not possible to access the class being defined while defining it. You can however call the method on the __init__ method:
class Something(Item):
def __init__(self):
type(self).define_field("name")
But then you're just reinventing the wheel.

When defining a class, you cannot reference the class itself inside its own definition block. So you have to call define_field(...) on MyItem after its definition. E.g.,
class MyItem(Item):
pass
MyItem.define_field("name")
item = MyItem({"name": "World"})
print("Hello {}!".format(item.name))
item.name = "reader"
print("Hello {}!".format(item.name))

python __getattribute__ override and #property decorator

I had to write a class of some sort that overrides __getattribute__.
basically my class is a container, which saves every user-added property to self._meta which is a dictionary.
class Container(object):
def __init__(self, **kwargs):
super(Container, self).__setattr__('_meta', OrderedDict())
#self._meta = OrderedDict()
super(Container, self).__setattr__('_hasattr', lambda key : key in self._meta)
for attr, value in kwargs.iteritems():
self._meta[attr] = value
def __getattribute__(self, key):
try:
return super(Container, self).__getattribute__(key)
except:
if key in self._meta : return self._meta[key]
else:
raise AttributeError, key
def __setattr__(self, key, value):
self._meta[key] = value
#usage:
>>> a = Container()
>>> a
<__main__.Container object at 0x0000000002B2DA58>
>>> a.abc = 1 #set an attribute
>>> a._meta
OrderedDict([('abc', 1)]) #attribute is in ._meta dictionary
I have some classes which inherit Container base class and some of their methods have #property decorator.
class Response(Container):
#property
def rawtext(self):
if self._hasattr("value") and self.value is not None:
_raw = self.__repr__()
_raw += "|%s" %(self.value.encode("utf-8"))
return _raw
problem is that .rawtext isn't accessible. (I get attributeerror.) every key in ._meta is accessible, every attributes added by __setattr__ of object base class is accessible, but method-to-properties by #property decorator isn't. I think it has to do with my way of overriding __getattribute__ in Container base class. What should I do to make properties from #property accessible?

I think you should probably think about looking at __getattr__ instead of __getattribute__ here. The difference is this: __getattribute__ is called inconditionally if it exists -- __getattr__ is only called if python can't find the attribute via other means.

I completely agree with mgilson. If you want a sample code which should be equivalent to your code but work well with properties you can try:
class Container(object):
def __init__(self, **kwargs):
self._meta = OrderedDict()
#self._hasattr = lambda key: key in self._meta #???
for attr, value in kwargs.iteritems():
self._meta[attr] = value
def __getattr__(self, key):
try:
return self._meta[key]
except KeyError:
raise AttributeError(key)
def __setattr__(self, key, value):
if key in ('_meta', '_hasattr'):
super(Container, self).__setattr__(key, value)
else:
self._meta[key] = value
I really do not understand your _hasattr attribute. You put it as an attribute but it's actually a function that has access to self... shouldn't it be a method?
Actually I think you should simple use the built-in function hasattr:
class Response(Container):
#property
def rawtext(self):
if hasattr(self, 'value') and self.value is not None:
_raw = self.__repr__()
_raw += "|%s" %(self.value.encode("utf-8"))
return _raw
Note that hasattr(container, attr) will return True also for _meta.
An other thing that puzzles me is why you use an OrderedDict. I mean, you iterate over kwargs, and the iteration has random order since it's a normal dict, and add the items in the OrderedDict. Now you have _meta which contains the values in random order.
If you aren't sure whether you need to have a specific order or not, simply use dict and eventually swap to OrderedDict later.
By the way: never ever use an try: ... except: without specifying the exception to catch. In your code you actually wanted to catch only AttributeErrors so you should have done:
try:
return super(Container, self).__getattribute__(key)
except AttributeError:
#stuff

Use class variables as instance vars?

What I would like to do there is declaring class variables, but actually use them as vars of the instance. I have a class Field and a class Thing, like this:
class Field(object):
def __set__(self, instance, value):
for key, v in vars(instance.__class__).items():
if v is self:
instance.__dict__.update({key: value})
def __get__(self, instance, owner):
for key, v in vars(instance.__class__).items():
if v is self:
try:
return instance.__dict__[key]
except:
return None
class Thing(object):
foo = Field()
So when I instantiate a thing and set attribute foo, it will be added to the instance, not the class, the class variable is never actually re-set.
new = Thing()
new.foo = 'bar'
# (foo : 'bar') is stored in new.__dict__
This works so far, but the above code for Field is rather awkward. It has too look for the Field object instance in the classes props, otherwise there seems no way of knowing the name of the property (foo) in __set__ and __get__. Is there another, more straight forward way to accomplish this?

Every instance of Field (effectively) has a name. Its name is the attribute name (or key) which references it in Thing. Instead of having to look up the key dynamically, you could instantiate Fields with the name at the time the class attribute is set in Thing:
class Field(object):
def __init__(self, name):
self.name = name
def __set__(self, instance, value):
instance.__dict__.update({self.name: value})
def __get__(self, instance, owner):
if instance is None:
return self
try:
return instance.__dict__[self.name]
except KeyError:
return None
def make_field(*args):
def wrapper(cls):
for arg in args:
setattr(cls, arg, Field(arg))
return cls
return wrapper
#make_field('foo')
class Thing(object):
pass
And it can be used like this:
new = Thing()
Before new.foo is set, new.foo returns None:
print(new.foo)
# None
After new.foo is set, 'foo' is an instance attribute of new:
new.foo = 'bar'
print(new.__dict__)
# {'foo': 'bar'}
You can access the descriptor (the Field instance itself) with Thing.foo:
print(Thing.foo)
# <__main__.Field object at 0xb76cedec>
PS. I'm assuming you have a good reason why
class Thing(object):
foo = None
does not suffice.

Reread your question and realized I had it wrong:
You don't need to override the default python behavior to do this. For example, you could do the following:
class Thing(object):
foo = 5
>>> r = Thing()
>>> r.foo = 10
>>> s = Thing()
>>> print Thing.foo
5
>>> print r.foo
10
>>> print s.foo
5
If you want the default to be 'None' for a particular variable, you could just set the class-wide value to be None. That said, you would have to declare it specifically for each variable.

The easiest way would be to call the attribute something else than the name of the descriptor variable - preferably starting with _ to signal its an implementation detail. That way, you end up with:
def __set__(self, instance, value):
instance._foo = value
def __get__(self, instance, owner):
return getattr(instance, '_foo', None)
The only drawback of this is that you can't determine the name of the key from the one used for the descriptor. If that increased coupling isn't a problem compared to the loop, you could just use a property:
class Thing:
#property
def foo(self):
return getattr(self, '_foo', None)
#foo.setter
def foo(self, value):
self._foo = value
otherwise, you could pass the name of the variable into the descriptor's __init__, so that you have:
class Thing:
foo = Field('_foo')
Of course, all this assumes that the simplest and most Pythonic way - use a real variable Thing().foo that you set to None in Thing.__init__ - isn't an option for some reason. If that way will work for you, you should prefer it.

Create per-instance property descriptor?

Usually Python descriptor are defined as class attributes. But in my case, I want every object instance to have different set descriptors that depends on the input. For example:
class MyClass(object):
def __init__(self, **kwargs):
for attr, val in kwargs.items():
self.__dict__[attr] = MyDescriptor(val)
Each object are have different set of attributes that are decided at instantiation time. Since these are one-off objects, it is not convenient to first subclass them.
tv = MyClass(type="tv", size="30")
smartphone = MyClass(type="phone", os="android")
tv.size # do something smart with the descriptor
Assign Descriptor to the object does not seem to work. If I try to access the attribute, I got something like
<property at 0x4067cf0>
Do you know why is this not working? Is there any work around?

This is not working because you have to assign the descriptor to the class of the object.
class Descriptor:
def __get__(...):
# this is called when the value is got
def __set__(...
def __del__(...
if you write
obj.attr
=> type(obj).__getattribute__(obj, 'attr') is called
=> obj.__dict__['attr'] is returned if there else:
=> type(obj).__dict__['attr'] is looked up
if this contains a descriptor object then this is used.
so it does not work because the type dictionairy is looked up for descriptors and not the object dictionairy.
there are possible work arounds:
put the descriptor into the class and make it use e.g. obj.xxxattr to store the value.
If there is only one descriptor behaviour this works.
overwrite setattr and getattr and delattr to respond to discriptors.
put a discriptor into the class that responds to descriptors stored in the object dictionairy.

You are using descriptors in the wrong way.
Descriptors don't make sense on an instance level. After all the __get__/__set__
methods give you access to the instance of the class.
Without knowing what exactly you want to do, I'd suggest you put the per-instance
logic inside the __set__ method, by checking who is the "caller/instance" and act accordingly.
Otherwise tell us what you are trying to achieve, so that we can propose alternative solutions.

I dynamically create instances by execing a made-up class. This may suit your use case.
def make_myclass(**kwargs):
class MyDescriptor(object):
def __init__(self, val):
self.val = val
def __get__(self, obj, cls):
return self.val
def __set__(self, obj, val):
self.val = val
cls = 'class MyClass(object):\n{}'.format('\n'.join(' {0} = MyDescriptor({0})'.format(k) for k in kwargs))
#check if names in kwargs collide with local names
for key in kwargs:
if key in locals():
raise Exception('name "{}" collides with local name'.format(key))
kwargs.update(locals())
exec(cls, kwargs, locals())
return MyClass()
Test;
In [577]: tv = make_myclass(type="tv", size="30")
In [578]: tv.type
Out[578]: 'tv'
In [579]: tv.size
Out[579]: '30'
In [580]: tv.__dict__
Out[580]: {}
But the instances are of different class.
In [581]: phone = make_myclass(type='phone')
In [582]: phone.type
Out[582]: 'phone'
In [583]: tv.type
Out[583]: 'tv'
In [584]: isinstance(tv,type(phone))
Out[584]: False
In [585]: isinstance(phone,type(tv))
Out[585]: False
In [586]: type(tv)
Out[586]: MyClass
In [587]: type(phone)
Out[587]: MyClass
In [588]: type(phone) is type(tv)
Out[588]: False

This looks like a use-case for named tuples

The reason it is not working is because Python only checks for descriptors when looking up attributes on the class, not on the instance; the methods in question are:
__getattribute__
__setattr__
__delattr__
It is possible to override those methods on your class in order to implement the descriptor protocol on instances as well as classes:
# do not use in production, example code only, needs more checks
class ClassAllowingInstanceDescriptors(object):
def __delattr__(self, name):
res = self.__dict__.get(name)
for method in ('__get__', '__set__', '__delete__'):
if hasattr(res, method):
# we have a descriptor, use it
res = res.__delete__(name)
break
else:
res = object.__delattr__(self, name)
return res
def __getattribute__(self, *args):
res = object.__getattribute__(self, *args)
for method in ('__get__', '__set__', '__delete__'):
if hasattr(res, method):
# we have a descriptor, call it
res = res.__get__(self, self.__class__)
return res
def __setattr__(self, name, val):
# check if object already exists
res = self.__dict__.get(name)
for method in ('__get__', '__set__', '__delete__'):
if hasattr(res, method):
# we have a descriptor, use it
res = res.__set__(self, val)
break
else:
res = object.__setattr__(self, name, val)
return res
#property
def world(self):
return 'hello!'
When the above class is used as below:
huh = ClassAllowingInstanceDescriptors()
print(huh.world)
huh.uni = 'BIG'
print(huh.uni)
huh.huh = property(lambda *a: 'really?')
print(huh.huh)
print('*' * 50)
try:
del huh.world
except Exception, e:
print(e)
print(huh.world)
print('*' * 50)
try:
del huh.huh
except Exception, e:
print(e)
print(huh.huh)
The results are:
hello!
BIG
really?
can't delete attribute
hello!
can't delete attribute
really?