The appeal, at least as I understood it, is that it allows you to quickly and efficiently determine the number of own properties of an object, without some sort of ungainly loop structure. That said, it seems it isn't actually "efficient", at least in the sense of not being O(n) in the number of properties on the object, for normal JS objects, because it uses the mechanism of initializing an enumerator to get the property count -- but initializing an enumerator iterates through all properties! So either way you still have this linear-time loop; the question is whether it's hidden behind an engine-specific per-object magical property, hidden through some less magical mechanism, or not provided at all (left to the developer to write himself -- slower than the near-max-speed native-code loop in obj_enumerate, but not asymptotically so). I favor removing __count__ entirely. If experience from such a change demonstrates use such that we need to support it, we can make it a getter directly on Object.prototype, which would still give people the tool without giving it to them in a way that requires special treatment in the engine.
Just remove it. No turd-polishing now that we have ES5 and its (costly but who cares) ways to count properties. /be
Created attachment 431717 [details] [diff] [review] KILL THEM
Comment on attachment 431717 [details] [diff] [review] KILL THEM Aw yeah.
Great. __parent__ next, right? /be
Sure, will file another bug for that.
You should probably fix http://mxr.mozilla.org/comm-central/source/suite/browser/test/browser/browser_pluginnotification.js too, to avoid comm-central bustage when this lands on mozilla-central.
https://mxr.mozilla.org/addons/search?string=__count__ shows several hits too :(
I'm fixing the c-c bit once a pul -u completes. I have a blog post queued up on this topic, just waiting for a TM->m-c merge and a nightly cycle so there's something concrete to point developers toward...