Closure analyzer output: let

[borkdude]

The following script demonstrates that returning a closure directly from the analyzer can have beneficial effect on performance:

;; run with:
;; clojure -J-Dclojure.compiler.direct-linking=true -M let_closure.clj
;; master: "Elapsed time: 6990.686673 msecs"
;; let-closure: "Elapsed time: 5351.171676 msecs"

(require '[sci.core :as sci])

(def ctx (sci/init {}))

(require '[sci.impl.analyzer :as ana])

(def analyzed (ana/analyze ctx '(let [x 1] x)))

(require '[sci.impl.evaluator :as eval])

(prn (eval/eval ctx analyzed))

(time (dotimes [_ 10000000]
        (eval/eval ctx analyzed)))

I didn't see a regression in the binary loop example:

$ time tmp/let-closure "(loop [val 0 cnt 1000000] (if (pos? cnt) (recur (inc val) (dec cnt)) val))"
1000000
tmp/let-closure    1.45s  user 0.09s system 99% cpu 1.544 total

(but unfortunately also no speedup)