pickle and then object changes

[Peter Otten]

James Smith wrote:

> Say an object like this exists:
> class test:
>     a = ""
>     b = ""
> 
> You pickle it.
> 
> You change the object definition to have a new field:
> class test
>     a = ""
>     b = ""
>     c = ""
> 
> You read the pickled object.
> Will it load but ignore the new field?
> That is what I want.

Classes are pickled by name. You are free to replace them with whatever you 
want:

>>> class Test:
...     a = 1
...     b = 2
... 
>>> import pickle
>>> p = pickle.dumps(Test)
>>> class Test:
...     a = 10
...     b = 20
...     c = 30
... 
>>> P = pickle.loads(p)
>>> P.c
30
>>> P.a
10
>>> P is Test
True

If you introduce a new *instance* attribute that will not magically appear 
on unpickling:

>>> class T:
...     def __init__(self, a, b):
...         self.a = a
...         self.b = b
...     def __str__(self): return "T(a={0.a}, b={0.b})".format(self)
... 
>>> t = T(1, 2)
>>> print(t)
T(a=1, b=2)
>>> p = pickle.dumps(t)
>>> class T:
...     def __init__(self, a, b, c):
...          self.a = a
...          self.b = b
...          self.c = c
...     def __str__(self): return "T(a={0.a}, b={0.b}, 
c={0.c})".format(self)
... 
>>> print(T(10, 20, 30))
T(a=10, b=20, c=30)
>>> v = pickle.loads(p)

No problem so far as the initializer is not called so that the missing c is 
not a problem. But then:

>>> print(v)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<stdin>", line 6, in __str__
AttributeError: 'T' object has no attribute 'c'

One easy fix is to use a class attribute as the fallback:

>>> T.c = "class attribute"
>>> print(v)
T(a=1, b=2, c=class attribute)

But remember not to try this with mutable attributes. When `c` is a list

v.c.append(42)

will affect all instances with a missing `c` instance attribute.