Fails to encode attribute of nested collections.defaultdict

[kylebgorman]

I use nested defaultdicts to represent sparse matrices. With default settings (i.e., no directives to the backend) on Python 2.7.5 and Python 3.4.1, the outer default_factory is simply not encoded. The following example shows this:

import jsonpickle
import collections
mydict = collections.defaultdict(lambda: collections.defaultdict(int))
mydict[3][2] = 4
encoded = jsonpickle.encode(mydict)
print(encoded)

This prints:

{"py/object": "collections.defaultdict", "3": {"default_factory": {"py/type": "__builtin__.int"}, "2": 4, "py/object": "collections.defaultdict"}}

This is wrong: there's no outer default_factory at all. (Note this has nothing to do with the use of a lambda function, as far as I can tell: non-anonymous default_factories behave the same way.)

This also doesn't go round-trip:

mydict_new = jsonpickle.decode(encoded)
print(mydict.default_factory) # prints pointer to the lambda function
print(mydict_new.default_factory) # prints None

[davvid]

You're using integer keys so make sure to pass keys=True when calling encode and decode as demonstrated in the unit test.