We can save on unnecessary polls by looking at the expected next here->idle transition time instead of blindly polling every 5 seconds. (FYI, I'm interested in having the nsIIdleService send an observerService notification, once-a-day-on-idle, which will send that notification when the user goes idle, but will send at most one of these in a 24 hour period. This will require a timer to be available all the time, so I figured it would be good to make the polling "interval" smarter.)

Keep the minimum polling time at 5 seconds, but if we don't need to poll frequently, just check once every 5 minutes. (Actually we would only really need to check when the next here->idle transition might occur, but this is in preparation for an always-around timer for the once-a-day-on-idle notification.) Also store the required idle time in milliseconds to not need to calculate * 1000 every time we check away state.

Comment on attachment 359887 [details] [diff] [review] v1 - { mTimer->Cancel(); - mTimer = nsnull; - } Please don't remove the {} around single statements. That's our coding style last I checked.

This is going to suck for IM and such applications, where the idle time is actually directly reflected by automatic UI changes or server interaction. I mean, I wouldn't want my instant messaging program to wait 5 minutes before realizing I was back at the machine. :-\ So I'm not sure this is the right thing to do. I realize I don't have a say in this, but I'm a bit confused what your arguments are as to why this is even possible (rather than why it'd be a perf improvement - obviously we could get an even bigger perf improvement if we only polled once a day, but obviously that would also defeat the point).

When expecting an idle->back transition, it'll poll at the current 5 seconds. Only if all listeners are waiting for back->idle transition will it poll at 5 minute intervals (or somewhere between 5sec and 5min, e.g., 3 minutes if the next idle listener is looking for idle in 3 more minutes.)

(In reply to comment #4) > When expecting an idle->back transition, it'll poll at the current 5 seconds. > Only if all listeners are waiting for back->idle transition will it poll at 5 > minute intervals (or somewhere between 5sec and 5min, e.g., 3 minutes if the > next idle listener is looking for idle in 3 more minutes.) Ah, sorry, that's me not reading the patch (I just woke up). Apologies, and yes, that sounds fine, thanks!

(In reply to comment #2) > Please don't remove the {} around single statements. Restored the {}s and added them to the new code.