http://bugs.opencompositing.org/show_bug.cgi?id=781





--- Comment #35 from [EMAIL PROTECTED]  2008-04-16 22:59:32 ---
Ok, I got a hold of aaronp from nvidia for a few minutes in #nvidia here is the
discussion (about the window decorations problem):

(03:28:01 PM) aaronp: jdeslip: I haven't really looked for that one, I've been
swamped with other stuff.
(03:28:21 PM) aaronp: However, there were a couple of corruption-causing bugs
that I fixed recently, so hopefully it's fixed already.
(03:29:08 PM) jdeslip: aaronp: were these released in 173.08?  or a future
release?  I don't think this bug is fixed yet in 173.08
(03:29:20 PM) aaronp: No, a future release.
(03:29:29 PM) jdeslip: ah - sounds hopefully
(03:29:52 PM) aaronp: Which window decorator is this?
(03:29:59 PM) jdeslip: in case it is still not fixed, to reproduce it, you
might turn off loose-binding in compiz.  it happens much more frequently
(03:30:13 PM) jdeslip: it happens with both the gtk and emerald decorators -
but only with compiz on
(03:30:31 PM) aaronp: Wait, it reproduces more often with *strict* binding?
(03:30:41 PM) jdeslip: no with loose-binding
(03:30:59 PM) jdeslip: (sorry, I think I misspoke above) much more frequent
with loose-binding
(03:31:00 PM) aaronp: That thread has a lot of conflicting information.
(03:31:15 PM) aaronp: E.g. UseCompositeWrapper vs. no composite wrapper
(03:31:34 PM) jdeslip: I have tried both of those at least, and the bug occurs
on both
(03:31:47 PM) jdeslip: (UseCompositeWrapper or no)
(03:32:34 PM) aaronp: Aargh... why is Fedora 8's window sizing border only 2px
wide??
(03:33:02 PM) jdeslip: What do you mean by window sizing border?
(03:33:14 PM) aaronp: Oh, it was reported against pre-G80 cards... maybe that's
why I can't reproduce it right now... :)
(03:33:18 PM) ***aaronp switches cards.
(03:33:28 PM) zinx: aaronp: no, it happens on my 8600M GT too
(03:33:41 PM) zinx: aaronp: or, at least the shadow thing does.. i -think- the
window corruption happens
(03:34:04 PM) aaronp: Oh, I was wrong, it's only 1px wide.
(03:34:04 PM) zinx: i really should make sure :/
(03:34:05 PM) jdeslip: It seems to happen with varying level of frequency.  For
me, it is about once every 30 minutes or so.
(03:34:19 PM) aaronp: I.e. I have to get the mouse cursor EXACTLY on the right
column before I can resize the window.
(03:34:31 PM) jdeslip: lol, I see
(03:35:43 PM) jdeslip: In other words, once every 30 minutes a window
decoration will become garbled, and if I move the window it is fixed.  So, to
reproduce it, you might have to use compiz for a while.
(03:36:17 PM) zinx: or just change the title of a window a lot
(03:38:09 PM) aaronp: jdeslip: Does rotating the cube fix it too?
(03:38:41 PM) zinx: aaronp: anything that redraws the window decorations fixes
(or breaks) it
(03:38:51 PM) jdeslip: umm, I think so.  Let me try to get it to happen again
(03:39:11 PM) aaronp: It's possible that compiz is reading the window
decoration pixmap before the X server is done drawing to it.
(03:42:52 PM) jdeslip: of course now that I want it to happen, it refuses...
(03:43:08 PM) aaronp: Yeah, it's not happening here either.
(03:43:21 PM) aaronp: I have to work on other stuff, but I'll ping the QA guys
to see if they have time to try to repro this.
(03:43:30 PM) jdeslip: aaronp: if that were the case, could the compiz devs do
something about it?
(03:43:49 PM) jdeslip: ok, thanks 
(03:44:09 PM) aaronp: jdeslip: It's a tough problem, and we actually discussed
it when the Composite and Damage extensions were being developed, but there was
no good resolution.
(03:44:23 PM) aaronp: The problem is that the X server sends damage events
before it's even submitted the rendering commands to the hardware.
(03:44:54 PM) aaronp: If compiz's reponse to the damage events beats the X
server's rendering to the framebuffer, you'll get incomplete rendering on the
screen.
(03:45:28 PM) jdeslip: sounds like a likely cause of the problem
(03:45:41 PM) aaronp: I don't know how compiz interacts with the window
decorator, but if it relies on damage events, that could be it.
(03:45:46 PM) jdeslip: (seeing as it is very machine/time dependent
(03:46:10 PM) jdeslip: I will suggest this to the compiz devs and see if they
have any comments
(03:46:19 PM) aaronp: One trick would be to do the window decorator rendering
with OpenGL, because our OpenGL implementation sends the damage events in
response to the actual damage.
(03:46:43 PM) aaronp: Another would be to read a pixel from the pixmap before
submitting the OpenGL commands, but that'll be a big performance bottleneck.
(03:48:28 PM) jdeslip: hmm, ok (I have to admit, I am personally don't know
much about how compiz works)
(03:48:41 PM) jdeslip: but, I'll pass it along to the compiz bug report

Aaron's last few comments meant very little to me.  But, perhaps they mean more
to you compiz Devs... Could the problem possibly on the compiz end?  Do his
suggestions make sense?


-- 
Configure bugmail: http://bugs.opencompositing.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.
_______________________________________________
Dev mailing list
[email protected]
http://lists.compiz-fusion.org/mailman/listinfo/dev

Reply via email to