From bidict to ndict

[wmjd]

Have you considered a 1-to-1 mapping data structure between not just two, but any arbitrary number of domains? In the implementation I have roughly ideated, there are n backing dictionaries
D_1 -> D_2 -> … -> D_n -> D_1. Since function composition is a transitive relation, we can map into any domain from any domain in at most n-1 steps (but I think we still keep constant time lookups since the input size is maybe len(D_i) not n)

Perhaps this is a naive, wacky, or impractical idea. However, if you have any thoughts, I would be very happy to know them! I think bidict is super cool and I am, essentially, at your service :)

[jab]

@wmjd Thanks for posting! Perhaps the interesting question here is for what values of n is it worth having a special-purpose library like bidict vs. using something like https://docs.python.org/3/library/sqlite3.html. As Raymond Hettinger once put it,

look at the generic description of the problem: "I have a list of n-ary tuples with named fields and would like to be able to retrieve a tuple using any of m-fields as a lookup key (where 2 <= m <= n). Also, I would like to enforce the constraint that those m-fields are all unique keys."

That pretty much sounds like a typical database problem featuring a single table with multiple indexes. One can wish-away the complexity of a database but you will end-up re-inventing an in-memory, one-table version of database.

I think for n=2, bidict is usually a better fit than something like sqlite3, but I buy Raymond's view that as n increases, at a certain point, a database-shaped thing is the right solution at least for many use cases.

P.S. I noticed your https://github.com/wmjd/TransitiveDictionary experiment -- fun :)