How does a dictionary work exactly?

Skip Montanaro skip.montanaro at gmail.com
Thu Jul 16 19:56:34 CEST 2015


> I was trying to see how some have implemented a hashtable.  I took a gather at dictobject.h/.c.  It seems that underneath it all it's a linked list and that is used in order to store the actual information (I'm looking at PyDictEntry.)
>
> Am I correct in my assumption or is there more to this?  I'm still looking into how new entries are handled.

The Python dictionary implementation has been pretty well optimized
over the years, so it might not be the most easy-to-read code. You
might actually try and latch onto a copy of dictobject.[ch] from a
very old version of Python (1.5-ish). That will have much less in the
way of speed tricks obfuscating the code.

Very generally (I'm writing with a lot of water under the bridge since
I last thought about this), a dictionary contains an array whose
length is typically a power of two (2**n). When considering a key for
insertion or lookup, a hash value is computed, and the last n bits of
the resulting value are used as an index into that array. Each element
of the array consists of a linked list of all the key/value pairs
whose keys hash to that value. Once you've identified an element in
the hash array, the linked list is traversed looking for the key.
There are three operations: get, set, delete. Each operation has one
of two actions to perform depending whether the key is found or not:

* get - if found, return the corresponding value, if not, raise KeyError
* set - if found, replace the old value with the new, if not, add a
new key/value pair to the list
* del if found, delete the key/value pair, if not, raise KeyError

The challenge is to come up with a reasonable size hash array and a
suitable hash function which balances the desire to not chew up all of
memory with the desire to have very short lists. In Python's case, I
believe that dictionaries with strings as keys (and the hash function
used for strings) have been optimized for how Python's runtime works.

Skip


More information about the Python-list mailing list