We throttle the refresh driver; we should throttle DOM timers too. The issue is web compat...
On the other hand, I hear tell that other browsers are considering doing this, so at least we're not alone.
Oh, for refresh driver we throttle to 1Hz, but once 2.0 ships I'll land a patch to throttle to an doubling-each-time interval.
I don't think we want the doubling behavior for DOM timers, though.
nsGlobalWindow has its own concept of "active" which seems to do with focus. I don't think we want to use that here, fwiw.
Created attachment 511636 [details] [diff] [review]
Pretty easy to do if we want to!
Chrome is going to set the minimum interval to 0.5Hz I believe.
Yeah, that's what the attached patch uses as the default pref value.
Or rather it does 0.5s, which is what I think you meant. We should clearly make it 1Hz to avoid that problem....
Chromium issue is http://code.google.com/p/chromium/issues/detail?id=66078 and they seem to use 1s.
dev-doc-needed: https://developer.mozilla.org/en/DOM/window.setTimeout#Minimum_delay_and_timeout_nesting will need to mention this change if it lands.
Comment on attachment 511636 [details] [diff] [review]
Pretty easy to do if we want to!
Let's give this a shot! Patch looks good (and I'm fine with either 1 or 2Hz here, maybe go with 1 to match chrome?).
*** Bug 532904 has been marked as a duplicate of this bug. ***
Pushed http://hg.mozilla.org/mozilla-central/rev/e328ab1bcb03 except I forgot to fix the commit message... I used a 1000ms clamp, but the commit message still says 500ms. The correct number is 1000.
is there anything here that mobile/e10s needs to do to get this benefit?
I don't think so. You're already flagging docshells as inactive so images get discarded, right?
I can write up a testcase for you to test whether this patch is working if you want (pretty easy to do a manual test; hard to do an automated one).
yes, i think we do. alon would know specifically. happy to run a test case.
I tested to make sure, and this works perfectly on mobile. Great stuff!
On a related note, this might make bug 588975 (stop animating images in background tabs) easier to do, which is also about reducing activity in inactive tabs.
Created attachment 522433 [details]
Testcase that can be used to test this
This works on background tabs, but not on active tabs on minimized windows. Is this the expected behavior?
For comparison, in Chrome, the active tab of a minimized window is considered inactive.
> Is this the expected behavior?
For now, yes. Which tabs are active is under control of the user interface; it might be worth a bug on Firefox to mark the active tab in a minimized window as inactive.
I filed bug 648045 and bug 648046 on applying the throttling in more cases.
Also updated Firefox 5 for developers.
Is there a specific DOM event that gets fired when the clamping to 1000 milliseconds starts (We would need to register a listener for it first of course though for it to be exposed...)? This currently trashes the JS GameBoy Color emulator audio when it's running inside of an inactive tab, as everything is driven in it at 16 ms intervals by a sole setInterval.
So the question remains, is there a unique DOM event that would fire solely for this throttling, to warn the web app of the slowdown? Doing normal timestamp comparisons is not the best, due to slow browsers (mobile especially) and timestamp checking for automatic pausing could get hacky.
> Is there a specific DOM event that gets fired
At this point, yes: http://www.w3.org/TR/2011/WD-page-visibility-20110602/#sec-visibilitychange-event