Shared publicly  - 
 
Playing with Klyde for #hackweek9 - making a lightweight desktop based on a Plasma Workspace. Made a 'light' version of systemsettings, hiding almost half the settings modules and categories by default. See http://paste.opensuse.org/24387480 for a taste of a cleaner systemsettings :D
17
1
Shalok Shalom's profile photoJos Poortvliet's profile photoWill Stephenson's profile photoVictor Vicente de Carvalho's profile photo
33 comments
 
I'd like to see that and man, it'db great, if the dependencies of the whole KDE package could be somehow cut. E.g. I don't use Kmail and Kontact, but I can't remove it, cause it would remove other dependencies too...
 
Finally. I installed 12.3 with KDE and once again saw the clutter and ran screaming. 
 
while the idea is good, you should really coordinate that with the KDE developers. A good example of something done "wrong" is using Kubuntu's low fat settings. They disable compositing which is honestly rather stupid thing to do - and yes I complained to Kubuntu devs about it.
 
I want to help with some design, if needed! :D Let me know. I tried adding myself to the Trello board but I guess I have to be invited lol. Stay in touch guys!
 
+Andy anditosan +Will Stephenson can probably add you. Realize that we need extremely simple design for performance reasons. We're using Plastique for the style and window decoration and a flat color background... For Plasma, we use Aya as that follows the desktop color scheme and is relatively simple.

If another style would lead to better performance and/or lower memory usage, it'd be an option but otherwise, these are the choices we have to go for.
 
+Jos Poortvliet  as there is no "Plastique" window decoration... I assume you try using the Plastik decoration. Which is nowadays written in QML and probably for some systems rather expensive - as said get in touch with the developers. Only we can tell you what is a good combination and what not. Plastik in combination with OpenGL compositing is great, but terrible (on 4.10) without compositing or XRender.

So don't just pick stuff, but do proper benchmarking. Or write new stuff - we have been waiting for someone to write a really simple decoration for years.
 
+Martin Gräßlin thanks for that feedback. We are testing stuff, using smem -u on a freshly started user account with various settings. I haven't tried different window decorations yet so thanks for pointing that out.

I don't think we have time to write a new windeco in 3 days as I wouldn't know where to start and Will is busy getting kdeinit to work with systemd and writing stuff for systemsettings... But maybe Klaas Freitag might have interest...

If you can tell me what windeco would have the lowest performance overhead, I'd love you even more ;-)
 
Get rid of redundant clutter and I'll be very happy. 
 
The performance of the decoration pretty much depends on how KWin is used. Main factors here are which compositor and which Qt graphics system are used (in combination with GPU and driver). So there is no magic "this one is fastest". It really depends on how you want to combine such a system.

From my testing: gradients are evil and all decorations shipped in KWin use gradients, but Oxygen allows to turn them off. Nevertheless Oxygen would probably still be the most expensive one (shadows, pixmap cache...).

What it needs is a new decoration designed for low performance. Sure probably nothing for in three days, but anything else doesn't make any sense. It ends up as a random pick of stuff for reasons nobody can understand and gives everything else the branding of being slow. That doesn't help anyone and just creates work for upstream as we would have in future to justify why we use Oxygen over $FOO.
 
I used a qtcurve theme modeled after Adwaita from Gnome. It performed nicely, was clean, and treated GTK applications well. 
 
+Martin Gräßlin as I can't really judge drawing time and stuff like that, as I said, I use smem. The whole endeavour is unscientific anyway...

A reasonably clean session with laptop windeco is about 4-6 mb smaller than the same session with plastic, Oxygen is in between. Just looking at kwin (change deco, run kwin --replace, see memory usage), kwin with oxygen is 200kb larger than with the laptop deco (8.5 vs 8.7 mb) and plastik is indeed 12.3 mb, much bigger. Looks like I should go for Oxygen, for now. It's not as braindead as Laptop (what a horrible windec that is...).

Thanks for the tip. Any more are welcome...
 
one more point about benchmarking window decorations: some decorations use caches (especially Oxygen). So if you test with one or two windows the memory usage will be high, but once there is a large number of windows that really shows off and can than easily outperform even a minimalistic deco which doesn't use caches.

Also the memory usage doesn't say much. The difference you see for Plastik is most likely the usage of QtQuick, which hits you as soon as you use Alt+Tab or anything else with QtQuick anyway.

The problem is that it's unscientific benchmarking and you know what I think about that ;-)
 
+Martin Gräßlin sure, I know... But the goal is to get lower memory usage and we ARE getting there, I think. Scientific or not. Let's see by the end of the week...
 
+Jos Poortvliet honestly I doubt that it's possible to get the memory usage down when not doing it scientific. If it's not done scientific it's just like what Kubuntu did - let's disable random crap which sounds like being heavy. Whether it is or not doesn't get evaluated.

Don't get me wrong: I think it's a great idea and something very useful. But if it's not done correctly it causes more problems than benefits.
 
+Martin Gräßlin problem is that being really scientific about it would probably require a lot of time, using various combinations of hardware etcetera. So instead, we use educated guesses & quick benchmarking. There are obvious savings (after the changes, a fresh login takes quite a bit less memory than before) so I don't think it causes more problems than benefits.
 
+Jos Poortvliet at KWin we call that "to phoronix". Just think about how wrong you have been with Plastik. And the "educated guesses" result in things like "Reduce selection of Desktop Effects". Apart from technically not possible it would not reduce memory usage or startup time.

That's the point about contacting the developers. They can provide you at least with more information than "educated guesses".

Whether it causes more problems than benefits depends on how you market it afterwards. What Kubuntu did with the low fat settings caused problems for us in KWin as their "educated guesses" about Compositing being "fat" was wrong. So if you do a blog post "look we made KDE fast" and it's just based on "educated guesses" which would at least partially be wrong it would create problems - for both upstream and downstream. It's a topic I would not want to have on slashdot for example.
 
If I may stick my oar in here, Jos is mixing up different efforts.  See the colour coding on the trello, "Reduce selection of Desktop Effects" is part of the "simple config" effort, ie. not aiming to affect memory usage or startup time - just not presenting Desktop Cube, wobbly windows etc on the default install.  'Do proper memory measurement is still a todo, and until I've figured it out I'm not posting any numbers.
 
+Will Stephenson doesn't change a thing about getting in contact with the devs. E.g. the "Reduce selection of Desktop Effects" isn't possible for multiple technical reasons.
 
I know, we talked about it about a year ago, remember? You pointed out to me that the effects are statically compiled in for performance, and that +Aaron Seigo should do the same thing for the base plasmoids.
 
They are all compiled into one plugin - that's one of the reasons why it's not possible.
 
What reqts do you have of a 'really simple decoration'?  Is it possible to do simple and scalable?
 
+Will Stephenson I was just searching for the bug report about it, but cannot find it at the moment. The requirements would be "don't be too ugly, and work well with network transparency". Thought the last one becomes a little bit tricky because with Qt5 there is no way to get it working with network transparency. A solution to that would be going down to X11 directly, but that would conflict with Wayland plans...
 
Thank you so much, this is exactly what I thought KDE needed some weeks ago. Now if other things (i.e. Amarok's ui) would be simplified too, that would be amazing.
Also, a look and feel comparable to W7 (not differentiating betweens running applications and launchers) would probably be welcomed by users who hate W8. Can be done by doing some tweaking, but it would be nice if there would be just some ready-to-use present.
 
KlyDE finally on the right track. Activities for example I have never needed. And all these annoying animated boxes. Quite nasty is the menu with its delay, which until now I still can not leave.

In the configuration I like the division into:
Beginners and professionals

A 'set it to default' button is not bad either.

I look forward to KlyDE :)
 
And the KDE devs working to gether with th razor qt ones, so maybe KlyDE can offer KWin with razor-qt better than yet ;)
 
Yes, i just read, that kde and razor-qt gets more cooperate in the future, so maybe we can integrate razor-qt directly (as optional DE) in KLyDE?
 
+Martin Jü
There is the Icon Only Tasks Plasmoid - you can design them (in his settings) like Win7 or Unity. And YaRock is very nice ;)
 
Display manager is a question for upstream and the distro. If upstream KDE switches to SDDM or LightDM we'll follow that. 
Add a comment...