Comments on Sergio's devlog: Performance tips for Clojure

You create a list of one single element, namely th...

2009-03-13T02:55:00.000-07:00

You create a list of one single element, namely the said range. Try to remove the list or use list* in your benchmark and you will get similar results to the vector. (Of course mapping over one element is faster than mapping over one million)

Creating a seq on something is O(1), not O(n). The collection is not turned somehow into a seq, but the seq is an abstract view on the collection.

(def ls (list (range 1000000)))(def vc (vec (range...

2009-03-12T23:56:00.000-07:00

(def ls (list (range 1000000)))
(def vc (vec (range 1000000)))

(time (dorun (map (fn [x]) ls)))
"Elapsed time: 3.184435 msecs"

(time (dorun (map (fn [x]) vc)))
"Elapsed time: 498.535303 msecs"

I was based on something like that. Probably the extra time taken is from converting the vector to a seq. That would still make it O(n) right? It would just be for another reason.

Maybe you are getting similar performance because you didn't use dorun and map returns a lazy seq.

Thanks, I'll edit the post.

rest is an operation on seqs. A list is a seq, but...

2009-03-12T23:45:00.000-07:00

rest is an operation on seqs. A list is a seq, but a vector is not. When map is given a vector, it first creates a seq on it. The rest operation on that seq is O(1). In some quick micro-benchmarking, the performance of iterating identity many thousands of times over a list or vector of a few hundred items gave very similar performance. I'm interested in what you see in the code or in a test that suggests that map performs much better over a list than over a vector.