[prev in list] [next in list] [prev in thread] [next in thread]
List: kwin
Subject: Re: Review Request: Finally really fix the XSYNC implementation
From: Thomas_Lübking <thomas.luebking () web ! de>
Date: 2011-08-23 23:41:50
Message-ID: 20110823234150.13227.99232 () vidsolbach ! de
[Download RAW message or body]
[Attachment #2 (multipart/alternative)]
> On Aug. 23, 2011, 6:41 p.m., Martin Gräßlin wrote:
> > very nice :-) Just see the two minor comments below.
> >
> > Do you know whether GTK applications implement sync properly? All GTK apps I just tried (Firefox \
> > (yeah not the best example), GIMP and Inkscape) were still very jumpy compared to the Qt \
> > applications.
No idea whether they implement it "correctly" but they do need the sync request _before_ the XMoveResize \
while Qt doesn't seem to care (XSync? Implementation?)
I wasn't sure what's the "correct" order (isn't specified)
My understanding of a "sync request" was and is: align to the current state and yell when you're out of \
work It's however (on gtk+ at least) rather: "i set a mutex, fire a bunch of instructions (probably only \
"next XMoveResize") and you remove the mutex when you're done"
Whatever. Swapping the two lines will prevent gtk+ applications to rely on the timeout and Qt ones won't \
break. *sigh*
Swapped, fixed commented parts and pushed.
- Thomas
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://git.reviewboard.kde.org/r/102311/#review5949
-----------------------------------------------------------
On Aug. 14, 2011, 9:08 a.m., Thomas Lübking wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://git.reviewboard.kde.org/r/102311/
> -----------------------------------------------------------
>
> (Updated Aug. 14, 2011, 9:08 a.m.)
>
>
> Review request for kwin and Martin Gräßlin.
>
>
> Summary
> -------
>
> <assume some very angry rant about the "protocol", it's authors and the author of the original patch \
> here>
> I finally got along, looking depper into this.
> So far, the kwin implementation did hardly more than wasting resources.
>
> => changes:
> a) ensure to NEVER send a second sync request while one's still pending (leading to the issues back \
> then) b) trigger a resize of the targetted geometry before requesting the sync (so the client actually \
> syncs to the proper size and is there when we release the mutex) c) skip all moveResizeHandle calls \
> while there's a pending sync (the client's not done anyway and we cannot send a sync request) d) move \
> the *request* from "performMoveResize" into handleMoveResize, replace "syncTimeout" (great name) with \
> "performMoveResize" and have it do exactly that. e) don't new QTimer; delete QTimer with every size \
> change - it's on the heap.... f) reduce the timeout to 250ms. waiting 500ms for the client is \
> _completely_ inacceptable. if it cannot keep pace: bad luck. but having the user to wait half a second \
> makes him feel "oh, blast. crashed." g) fix tileset #ifdef (the geometry _has_ to be set esp. if \
> there's no tiling at all)
> ==> todo:
> - I thought about using the compositor to perform a texture scale while we're waiting for the client \
> (but the client/effectwindow would have to keep "realGeometry" vs. "visualGeometry" for \
> this
> - look around whether XSYNC preproc checks are sufficient
> - attempt to drop the second XMoveResize call for the window() WId (since it happened with the sync \
> request)
> - add a checkbox: "don't use xsync protocol" - frankly, this thing is so much wonky by design that i'd \
> nearly say to remove support completely.
> For the records (maybe i should add a comment in the code?)
> - two sync requests on the stack will break it. no way to re-sync.
> - one should aparently trigger the client resize before starting the sync request. otherwise the entire \
> thing is pointless.
> - sending the configure notifications is important. don't ask me why, but w/o the client will stop \
> sending sync events just like in the above case
>
> This addresses bug 160393.
> http://bugs.kde.org/show_bug.cgi?id=160393
>
>
> Diffs
> -----
>
> kwin/client.h 93510ec
> kwin/client.cpp 46c8eff
> kwin/composite.cpp edffac1
> kwin/events.cpp f523bf7
> kwin/geometry.cpp 51ed034
> kwin/manage.cpp 47c8814
>
> Diff: http://git.reviewboard.kde.org/r/102311/diff
>
>
> Testing
> -------
>
> yes.
> no more white edges in konsole (but for timeouts).
> no issues with clemens qclock testcase (which blocks painting for some hundred ms) - scaling is smooth, \
> blocks (as expected) and smoothly continues afterwards.
> nevertheless, i'd like to have this in master as soon as possible (also before "playing around" for \
> performance locally) and hopefully a lot of testing...
>
> Thanks,
>
> Thomas
>
>
[Attachment #5 (text/html)]
<html>
<body>
<div style="font-family: Verdana, Arial, Helvetica, Sans-Serif;">
<table bgcolor="#f9f3c9" width="100%" cellpadding="8" style="border: 1px #c9c399 solid;">
<tr>
<td>
This is an automatically generated e-mail. To reply, visit:
<a href="http://git.reviewboard.kde.org/r/102311/">http://git.reviewboard.kde.org/r/102311/</a>
</td>
</tr>
</table>
<br />
<blockquote style="margin-left: 1em; border-left: 2px solid #d0d0d0; padding-left: 10px;">
<p style="margin-top: 0;">On August 23rd, 2011, 6:41 p.m., <b>Martin Gräßlin</b> wrote:</p>
<blockquote style="margin-left: 1em; border-left: 2px solid #d0d0d0; padding-left: 10px;">
<pre style="white-space: pre-wrap; white-space: -moz-pre-wrap; white-space: -pre-wrap; white-space: \
-o-pre-wrap; word-wrap: break-word;">very nice :-) Just see the two minor comments below.
Do you know whether GTK applications implement sync properly? All GTK apps I just tried (Firefox (yeah \
not the best example), GIMP and Inkscape) were still very jumpy compared to the Qt applications.</pre> \
</blockquote>
</blockquote>
<pre style="white-space: pre-wrap; white-space: -moz-pre-wrap; white-space: -pre-wrap; white-space: \
-o-pre-wrap; word-wrap: break-word;">No idea whether they implement it "correctly" but they do \
need the sync request _before_ the XMoveResize while Qt doesn't seem to care (XSync? Implementation?)
I wasn't sure what's the "correct" order (isn't specified)
My understanding of a "sync request" was and is: align to the current state and yell when \
you're out of work It's however (on gtk+ at least) rather: "i set a mutex, fire a bunch of \
instructions (probably only "next XMoveResize") and you remove the mutex when you're \
done"
Whatever. Swapping the two lines will prevent gtk+ applications to rely on the timeout and Qt ones \
won't break. *sigh*
Swapped, fixed commented parts and pushed.</pre>
<br />
<p>- Thomas</p>
<br />
<p>On August 14th, 2011, 9:08 a.m., Thomas Lübking wrote:</p>
<table bgcolor="#fefadf" width="100%" cellspacing="0" cellpadding="8" style="background-image: \
url('http://git.reviewboard.kde.org/media/rb/images/review_request_box_top_bg.png'); background-position: \
left top; background-repeat: repeat-x; border: 1px black solid;"> <tr>
<td>
<div>Review request for kwin and Martin Gräßlin.</div>
<div>By Thomas Lübking.</div>
<p style="color: grey;"><i>Updated Aug. 14, 2011, 9:08 a.m.</i></p>
<h1 style="color: #575012; font-size: 10pt; margin-top: 1.5em;">Description </h1>
<table width="100%" bgcolor="#ffffff" cellspacing="0" cellpadding="10" style="border: 1px solid #b8b5a0">
<tr>
<td>
<pre style="margin: 0; padding: 0; white-space: pre-wrap; white-space: -moz-pre-wrap; white-space: \
-pre-wrap; white-space: -o-pre-wrap; word-wrap: break-word;"><assume some very angry rant about the \
"protocol", it's authors and the author of the original patch here>
I finally got along, looking depper into this.
So far, the kwin implementation did hardly more than wasting resources.
=> changes:
a) ensure to NEVER send a second sync request while one's still pending (leading to the issues back \
then) b) trigger a resize of the targetted geometry before requesting the sync (so the client actually \
syncs to the proper size and is there when we release the mutex) c) skip all moveResizeHandle calls while \
there's a pending sync (the client's not done anyway and we cannot send a sync request) d) move \
the *request* from "performMoveResize" into handleMoveResize, replace "syncTimeout" \
(great name) with "performMoveResize" and have it do exactly that. e) don't new QTimer; \
delete QTimer with every size change - it's on the heap.... f) reduce the timeout to 250ms. waiting \
500ms for the client is _completely_ inacceptable. if it cannot keep pace: bad luck. but having the user \
to wait half a second makes him feel "oh, blast. crashed." g) fix tileset #ifdef (the geometry \
_has_ to be set esp. if there's no tiling at all)
==> todo:
- I thought about using the compositor to perform a texture scale while we're waiting for the client \
(but the client/effectwindow would have to keep "realGeometry" vs. "visualGeometry" \
for this
- look around whether XSYNC preproc checks are sufficient
- attempt to drop the second XMoveResize call for the window() WId (since it happened with the sync \
request)
- add a checkbox: "don't use xsync protocol" - frankly, this thing is so much wonky by \
design that i'd nearly say to remove support completely.
For the records (maybe i should add a comment in the code?)
- two sync requests on the stack will break it. no way to re-sync.
- one should aparently trigger the client resize before starting the sync request. otherwise the entire \
thing is pointless.
- sending the configure notifications is important. don't ask me why, but w/o the client will stop \
sending sync events just like in the above case</pre> </td>
</tr>
</table>
<h1 style="color: #575012; font-size: 10pt; margin-top: 1.5em;">Testing </h1>
<table width="100%" bgcolor="#ffffff" cellspacing="0" cellpadding="10" style="border: 1px solid #b8b5a0">
<tr>
<td>
<pre style="margin: 0; padding: 0; white-space: pre-wrap; white-space: -moz-pre-wrap; white-space: \
-pre-wrap; white-space: -o-pre-wrap; word-wrap: break-word;">yes. no more white edges in konsole (but for \
timeouts). no issues with clemens qclock testcase (which blocks painting for some hundred ms) - scaling \
is smooth, blocks (as expected) and smoothly continues afterwards.
nevertheless, i'd like to have this in master as soon as possible (also before "playing \
around" for performance locally) and hopefully a lot of testing...</pre> </td>
</tr>
</table>
<div style="margin-top: 1.5em;">
<b style="color: #575012; font-size: 10pt; margin-top: 1.5em;">Bugs: </b>
<a href="http://bugs.kde.org/show_bug.cgi?id=160393">160393</a>
</div>
<h1 style="color: #575012; font-size: 10pt; margin-top: 1.5em;">Diffs</b> </h1>
<ul style="margin-left: 3em; padding-left: 0;">
<li>kwin/client.h <span style="color: grey">(93510ec)</span></li>
<li>kwin/client.cpp <span style="color: grey">(46c8eff)</span></li>
<li>kwin/composite.cpp <span style="color: grey">(edffac1)</span></li>
<li>kwin/events.cpp <span style="color: grey">(f523bf7)</span></li>
<li>kwin/geometry.cpp <span style="color: grey">(51ed034)</span></li>
<li>kwin/manage.cpp <span style="color: grey">(47c8814)</span></li>
</ul>
<p><a href="http://git.reviewboard.kde.org/r/102311/diff/" style="margin-left: 3em;">View Diff</a></p>
</td>
</tr>
</table>
</div>
</body>
</html>
_______________________________________________
kwin mailing list
kwin@kde.org
https://mail.kde.org/mailman/listinfo/kwin
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic