Review Request 126078: [OS X] modernising the KIdleTime plugin (WIP!)

[René J.V. Bertin]

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/126078/
-----------------------------------------------------------

(Updated Nov. 17, 2015, 10:12 p.m.)


Review request for KDE Software on Mac OS X and KDE Frameworks.


Changes
-------

This revision takes comments from the mailing list into account. Pure end-of-idle detection is done in a way similar to how `WidgetBasedPoller` does it, by detecting global UI events. That part explains the newly introduced ObjC++ file.
Like `WidgetBasedPoller`, the new implementation has to resort to a polling approach, and like that example class, the the polling frequency is adaptive. However, the adaptive algorithm:
- uses the difference between the current idle time and the next timeout from the timeouts list as the source of the "error" value used to calculate the timer interval
- uses <1 gain (difference/2) 
- relies on a sorted timeouts list, which ensures that the initial "next timeout" is equal to the minimal timeout

In addition, using a sorted timeout list means that determination of a timeout hit becomes much simpler, basically `idle >= timeout[i]`. That removes the possibility to signal early, or to reject signalling because a timeout was detected after the acceptance window.

I have left the `kidletime_example` run idling for over 3 minutes; as far as I can tell it behaves as expected, and yields a better-than-1ms accuracy (on average timeouts are signalled less than 1ms late). It only shows near the top in a running `top` listing on rare occasions, and then only at a few percents CPU.

I have left in the hooks to configure a fixed, higher polling frequency, for now, until I'm convinced that they're indeed not justified given the 1ms resolution imposed by the API.


Repository: kidletime


Description
-------

I noticed that the KIdleTime example doesn't work properly on OS X, and that the plugin for OS X still uses the deprecated Carbon-based algorithm that I already patched for KDE4.

This patch is a work-in-progress (hence the qDebugs) update to use IOKit, IORegistry and CoreServices to do idle-time calculation as it should be done, and allow simulated user activity through a "less deprecated" function.


Diffs (updated)
-----

  src/plugins/osx/CMakeLists.txt e1b50b8 
  src/plugins/osx/macpoller.h ef51ea5 
  src/plugins/osx/macpoller.cpp ad9c10f 
  src/plugins/osx/macpoller_helper.mm PRE-CREATION 

Diff: https://git.reviewboard.kde.org/r/126078/diff/


Testing
-------

On OS X 10.9 with Qt 5.5.1 and frameworks 5.16.0 .

The example now works: when I set a QTimer with interval==0, the expected wait for user input (`resumingFromIdle` signal) works. However, I am getting a `stopCatchingIdleEvents` signal which means the application waits forever, without ever getting to compare idle time to the list of timeouts.
I haven't been able to figure out where that signal comes from, nor why this doesn't happen on Linux.

Surely I'm missing something, but what?


Thanks,

René J.V. Bertin

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-frameworks-devel/attachments/20151117/c06942aa/attachment.html>