There are two aspects to this question that are inter-related.
1) Change I've made to the trie module in python (PyPI, github) to make it serialisable (and deserialisable).
The change I made is visible on github. I added two functions to the Node class used in serialisation.
def __getstate__(self):
return (self.parent, self.key, self.nodes, self.value)
def __setstate__(self, state):
(self.parent, self.key, self.nodes, self.value) = state
if type(self.value) == object:
self.value = Node.no_value
The Node class uses an instance of object to signify that there is no value for that Node. If you serialise and deserialise a Trie of Nodes then the values are set to different instances of object, this is why I added a check for the type being object and replacing it with Node.no_value.
Is this a reasonable way of doing things? Is anyone likely to want to store an actual instance of object?
2) Is there a better way of signifying that a Node has no value, that is more robust to serialisation?
