Creating Truly Empty Types in Python

Question

On a regular basis, I want to create completely empty objects in Python. Basically, I want to have a dict where the keys are attributes. In Python 2.x, I can create an old-style type like this:

class Empty: pass

This will create a type that only has two attributes (__doc__ and __module__). In Python 3.x, everything is a new-style class, so I get 18 attributes.

In my current situation, I have a class that allows you to specify types that need monkey patched within a unit test. When the patches are applied, I am creating a type with attributes with the names of each mocked-out type. This is pretty much what my current implementation is doing:

class EmptyType: pass
...
mocks = EmptyType()
for mock_name, mock in patches:
    setattr(mocks, mock_name, mock)

My concern is that should someone be mocking a private member, they might run into naming collisions with the names in the EmptyType object. That's why I'd like to keep as few attributes in the EmptyType as possible. And it is way easier to say mocks.find_users than it is to say mocks["find_users"], especially when I know the name has to be a valid identifier.

Right now, I have provided the ability to give mocks different names, other than what would otherwise be the default. Still, it would be nice to avoid confusing errors in the first place. It is very easy to create almost empty types in JavaScript and I was hoping there was something similar in Python, since I keep finding good uses for them.

I don't really know why, but this pattern seems a little alien for Python. Monkey-patching is often frowned upon in Python circles, duck-typing and EAFP are encouraged. — Paulo Scardine
– Paulo Scardine, Commented Aug 17, 2012 at 1:29

jdi · Accepted Answer · 2012-08-17 02:12:49Z

6

What about creating your own custom container?

class Empty(object):

    def __init__(self, **kwargs):
        object.__setattr__(self, '_obj', kwargs)

    def __getattribute__(self, name):
        obj = object.__getattribute__(self, '_obj')
        try:
            return obj[name]
        except KeyError:
            cls_name = object.__getattribute__(self, '__class__').__name__
            raise AttributeError(
                "'%(cls_name)s' object has no attribute '%(name)s'" % locals())

    def __setattr__(self, name, val):
        obj = object.__getattribute__(self, '_obj')
        obj[name] = val

    def __getitem__(self, key):
        return getattr(self, key)

    def __setitem__(self, key, val):
        return setattr(self, key, val)

Usage:

e = Empty(initial='optional-value')
e.initial
# 'optional-value'
e.val = 'foo'
e.val
# 'foo'
e.bad
# AttributeError: 'Empty' object has no attribute 'bad'
setattr(e, 'user', 'jdi')
e.user
# 'jdi'
e['user']
# 'jdi'

# note that you dont even see anything when you dir()
dir(e)
# []

# and trying to access _obj is protected
e._obj
#AttributeError: 'Empty' object has no attribute '_obj'

# But you could even set your own _obj attribute
e._obj = 1
e._obj
# 1

It will store everything under a _obj dictionary so you basically get a clean space that doesn't conflict with the actual instance attributes.

edited Aug 17, 2012 at 2:12

answered Aug 17, 2012 at 1:38

jdi

93k20 gold badges174 silver badges208 bronze badges

Sign up to request clarification or add additional context in comments.

8 Comments

mgilson Over a year ago

+1 for using __getattribute__ instead of the more common __getattr__. ( I had to look up the difference to convince myself that __getattribute__ was really what you wanted ... )

jdi Over a year ago

@mgilson: Thanks! Ya I still have to remind myself about the special builtins. For anyone reading that comment, I used __getattribute__ to make sure it always handles the accessing of attributes and redirects them to the private dictionary, regardless of whether they might already exist on that Empty class.

mgilson Over a year ago

One other comment -- I can't think of any reason to have the last line in __getattribute__ (return object.__getattribute__(self, name) ). Is that just to raise the AttributeError if the attribute doesn't exist? If so, why not raise it more directly (explicitly) using a try/except catching the KeyError and re-raising it as an AttributeError?

jdi Over a year ago

@mgilson: My only reason for doing that was to let the object take care of the standard AttributeError string formatting and whatnot. I would just be reproducing the similar error. I could have done that if I wanted a custom message though. It just didn't seem all that critical to handle it explicitly in this case.

mgilson Over a year ago

I think this is my last comment. Instead of using '_obj' as the name of the single attribute, you could use '100' or something else which isn't a valid identifier. Then you eliminate the namespace clash problem entirely (since a.100 isn't valid syntax).

|

Collectives™ on Stack Overflow

Creating Truly Empty Types in Python

1 Answer 1

8 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

8 Comments

Your Answer

Sign up or log in

Post as a guest

Related