map-invert should use transients and reduce-kv instead of reduce

Question

Two performance enhancements on clojure.set/map-invert:

1) Use reduce-kv to avoid realizing mapentry's from input map
2) Use transients to create the output map

*Perf:*


(use 'criterium.core)
(require '[clojure.set :as set])
(def m1 (zipmap (range 1) (range 1)))
(def m10 (zipmap (range 10) (range 10)))
(def m100 (zipmap (range 100) (range 100)))
(def m1000 (zipmap (range 1000) (range 1000)))
(quick-bench (set/map-invert m1000))
(quick-bench (set/map-invert m100))
(quick-bench (set/map-invert m10))
(quick-bench (set/map-invert m1))

;; means before:  138 ns  1.8 µs  20.6 µs  304 µs
;; means after:   151 ns  1.3 µs   9.0 µs  126 µs


*Patch:* clj-1808-map-invert-should-use-reduce-kv-and-transient.patch

*Screened by:* Alex Miller

Answer 1

Comment made by: alexmiller

Would be nice to see a quick perf test that compared before/after times.

Answer 2

Reference: https://clojure.atlassian.net/browse/CLJ-1808 (reported by tonsky)