TL;DR
How can I get superkeys to be autovivified in a Python dict when assigning values to subkeys, without also getting them autovivified when checking for subkeys?

Background: Normally in Python, setting values in a nested dictionary requires manually ensuring that higher-level keys exist before assigning to their sub-keys. That is,

my_dict[1][2] = 3

will not reliably work as intended without first doing something like

if 1 not in my_dict:my_dict[1] = {}

Now, it is possible to set up a kind of autovivification by making my_dict an instance of a class that overrides __missing__, as shown e.g. in https://stackoverflow.com/a/19829714/6670909.

Question: However, that solution silently autovivifies higher-level keys if you check for the existence of a sub-key in such a nested dict. That leads to the following unfortunateness:

>>> vd = Vividict()
>>> 1 in vd
False
>>> 2 in vd[1]
False
>>> 1 in vd
True

How can I avoid that misleading result? In Perl, by the way, I can get the desired behavior by doing

no autovivification qw/exists/;

And basically I'd like to replicate that behavior in Python if possible.

This is not an easy problem to solve, because in your example:

my_dict[1][2] = 3

my_dict[1] results in a __getitem__ call on the dictionary. There is no way at that point to know that an assignment is being made. Only the last [] in the sequence is a __setitem__ call, and it can't succeed unless mydict[1] exists, because otherwise, what object are you assigning into?

So don't use autovivication. You can use setdefault() instead, with a regular dict.

my_dict.setdefault(1, {})[2] = 3

Now that's not exactly pretty, especially when you are nesting more deeply, so you might write a helper method:

class MyDict(dict):def nest(self, keys, value):for key in keys[:-1]:self = self.setdefault(key, {})self[keys[-1]] = valuemy_dict = MyDict()my_dict.nest((1, 2), 3)       # my_dict[1][2] = 3

But even better is to wrap this into a new __setitem__ that takes all the indexes at once, instead of requiring the intermediate __getitem__ calls that induce the autovivication. This way, we know from the beginning that we're doing an assignment and can proceed without relying on autovivication.

class MyDict(dict):def __setitem__(self, keys, value):if not isinstance(keys, tuple):return dict.__setitem__(self, keys, value)for key in keys[:-1]:self = self.setdefault(key, {})dict.__setitem__(self, keys[-1], value)my_dict = MyDict()
my_dict[1, 2] = 3

For consistency, you could also provide __getitem__ that accepts keys in a tuple as follows:

def __getitem__(self, keys):if not isinstance(keys, tuple):return dict.__getitem__(self, keys)for key in keys:self = dict.__getitem__(self, key)return self

The only downside I can think of is that we can't use tuples as dictionary keys as easily: we have to write that as, e.g. my_dict[(1, 2),].