Patricia trie symbol tables?

[Florian Weimer]

* John Moser:

> Reading what Drepper says about -Wl,-O1 and why to use it, I get the
> gist of this:  You'll do comparisons against
> "_ZN14some_namespace22some_longer_class_name" twice, what a waste.  The
> point of -Wl,-O1 is to make the buckets smaller so this happens less
> often, as a side effect of making the hash table bigger (in other words,
> no guarantees; but it'll probably happen).

The technique described at <http://www.cs.princeton.edu/~rs/strings/>
might also be a win, compared to hash tables.  (Quoting from my
bookmark file, as I'm offline at the moment.)