Tuning tries (IntSet, hashmap, bytestring-trie, etc.)
ekmett at gmail.com
Sat Jun 26 16:33:21 EDT 2010
On Sat, Jun 26, 2010 at 4:04 PM, Jim Apple
<jbapple+haskell-lib at gmail.com<jbapple%2Bhaskell-lib at gmail.com>
> Given the recent faster trie implementations (hooray!), I thought it
> would be worth pointing out that a pretty thorough discussion of the
> tradeoffs of functional trie design is available at
> An experimental study of compression methods for functional tries
> Jukka-Pekka Iivonen, Stefan Nilsson, and Matti Tikkanen
> FWIW, apparently clojure uses a 32-ary hashtrie:
> Because it is only binary, Data.IntSet is sometimes quite deep. All of
> that pointer chasing makes Data.Set actually faster than Data.IntSet
> for looking up 0 in the set 0:[2^i | i <-[0..30]].
To be fair that particular set is quite literally the worst case scenario
for a patricia trie.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Libraries