1375676 - Can JSObject::callHook() tail recurse and therefore be surprisingly expensive?

Status: NEW

(Core :: JavaScript Engine, enhancement, P3)

Description

[(no longer active)]

See this profile: http://bit.ly/2t0DWM8

Why is JSObject::callHook() taking 284ms self time?

The only possible explanation I could come up with was that we end up in the following sequence:

 * https://searchfox.org/mozilla-central/rev/3291398f10dcbe192fb52e74974b172616c018aa/js/src/jsobj.cpp#2232
 * https://searchfox.org/mozilla-central/rev/3291398f10dcbe192fb52e74974b172616c018aa/js/src/proxy/Wrapper.cpp#286
 * https://searchfox.org/mozilla-central/rev/3291398f10dcbe192fb52e74974b172616c018aa/js/src/jsobj.cpp#2205
 * Back to JSObject::callHook()

Comment 1

[Jan de Mooij [:jandem]]

From the inverted stack it seems we're using instanceof on a CCW. So unwrapping there probably wouldn't be too surprising... Hopefully Jason's wrapper work will eliminate this one.

We also really need to inline JSObject::isCallable() and JSObject::callHook() - I noticed this also affects Object.prototype.toString.

Comment 2

[Jan de Mooij [:jandem]]

We inline JSObject::isCallable/callHook now so this is probably no longer as bad. Also the same-compartment-realms work will likely fix this by getting rid of the wrappers - I think we should just wait for that..

Comment 3

[Jan de Mooij [:jandem]]

Removing [qf] per comment 2.