Yeah it appears to be a hybrid Trie implementation. Which I personally find extremely amusing. The primary drawback with using Trie's is that they are typically terribly wasteful of memory. (Except if the data set is extremely dense.) Which is why it turns out that most times Tries are implemented they are implemented in a hybrid form, or are designed to work on a limited data set or both. OTOH one advantage of a pure Trie lookup is that a C implementation of the algorithm performs similarly or better than a hash lookup but does prefix matching as well.