Question

I have been told that declaring dynamic attributes within a classes scope is not the ‘Python Way’ but I do not understand why.
Could someone explain this to me or point me at some documentation as to why this is a bad thing? Honestly, I thought this was good practice, if anything for self documenting code.

Example:

class ClassA(object):
    user_data = {}

    def set_user(self):
        self.user_data['username'] = 'fred'

The only reason I can see for not using this is that attributes are static (and so could be misleading)..

Answer 1

With the code as shown, user_data isn’t a dynamic attribute (It’s not created on the class instance “dynamically”). It’s a class attribute which is a lot more like “static” attributes in some other languages I believe. This means that it is an attribute declared on the class at the time the class is read and initialized. This has the side-effect/benefit of all of the instances being able to access the same object via self.whatever.

In other words:

class Foo(object):
    whatever = {}
    def __init__(self):
        print self.whatever is Foo.whatever

will always print True. Of course, you can change this behavior by adding a different whatever attribute to an instance:

class Bar(object):
     whatever = {}
     def __init__(self):
         self.whatever = {}
         print self.whatever is Bar.whatever

With Foo, if I add items: foo_instance.whatever['foo'] = 'bar', then foo_instance2 will see that change as well, whereas that won’t be the case with Bar.