928971 - Run-once closure analysis bug can cause unnecessary script cloning

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jan de Mooij [:jandem]]

Noticed this on the "[coll (take 100000 (iterate inc 0))] (reduce + 0 coll)" Clojurescript test. We keep cloning some scripts (and hence never JIT them). See the function literals on line 2 and 10:

    var Fi = function Di(b, c) {
        return new kc(null, function () {
            if (0 < b) {
                var e = B(c);
                return e ? I(C(e), Di(b - 1, D(e))) : null
            }
            return null
        }, null, null)
    }(1E5, function Ei(b, c) {
        return I(c, new kc(null, function () {
            return Ei(b, b.a ? b.a(c) : b.call(null, c))
        }, null, null))
    }(function (a) {
        return a + 1
    }, 0));

Comment 1

[Luke Wagner [:luke]]

Not fixed by bug 880085?

Comment 2

[Luke Wagner [:luke]]

Oops, n/m

Comment 3

[Brian Hackett [Laid off!]]

Attachment: patch

We don't end up emitting a run once lambda here because of the outer name, but due to an inconsistency between how run once lambdas are detected the inner functions of the lambda are still given singleton type and end up being deep cloned each time the outer script runs.

Comment 4

[Jan de Mooij [:jandem]]

Excellent, on this subtest the patch improves our time from > 700 ms to 130-140 ms.

Comment 5

[Brian Hackett [Laid off!]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/d2b4e498b5aa

Comment 6

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/d2b4e498b5aa