/srv/irclogs.ubuntu.com/2014/05/20/#ubuntu-mir.txt

bschaeferRAOF, hey, how would one force mir to do double buffering? Seeing interesting issues/flashing (looks like the third buffer might be to much?)00:44
RAOFbschaefer: I think you need to twiddle with the code to do that.00:45
bschaeferRAOF, cool, ill dig into that :). Thanks!00:45
RAOFbschaefer: Also, on desktop, bypass *requires* triple buffering. But you don't necessarily care about that if you're just debugging.00:46
bschaeferRAOF, yeah, im just checking if its my code, or the apps code00:46
RAOFbschaefer: You could also check the age on the buffer, client side, to detect if it's being triple buffered.00:47
bschaeferas this is the only time i've seen the need for a frame damage using opengles + sdl00:47
bschaeferthats a good idea as well00:47
bschaeferRAOF, testing qemu out00:47
bschaeferthe current image i see: http://i.imgur.com/RZg6uhx.jpg (it only updates the textures on whats changed)00:48
RAOFWhee!00:50
bschaefer:), it'll be nice to get working haha00:50
RAOFAlso, maybe tiling?00:50
RAOFHm00:51
bschaeferhmm, i think that image is after its freaking out00:51
bschaeferwhen i throw a clear in00:51
bschaeferi can see the textures updating only where somethings changed + width of the window00:52
bschaeferso each frame, it updates a chunk of the window, but adding in buffers it seems to be flashing00:52
bschaeferpossibly an alpha issue?00:53
bschaeferas i was running into an alpha issue with the pixel format before...00:53
bschaefer(as with RGBA8888 selected for the texture + surface, i was getting an empty/grey window)00:54
bschaeferforcing RGBX8888 let me see the textures00:54
RAOFHm.01:00
bschaeferRAOF, ill show you two pictures of one clearing each frame, and one doing it how it normal is01:00
bschaeferRAOF, otherwise ill debug what i can, and show you in malta01:01
RAOFYeah. Yay Malta!01:01
bschaeferits going to be to warm :(01:01
bschaeferbut yay!01:01
bschaeferRAOF, glClear before each from being drawn: http://i.imgur.com/MYUcRda.jpg01:02
bschaeferRAOF, no Clear:  http://i.imgur.com/J4IiccE.jpg01:03
bschaeferbut yeah, ill see what i can find out :)01:03
* bschaefer should have use mir screen cast01:03
RAOF:)01:03
duflubschaefer: I see nice screenshots. The gaps could well be Qemu's fault. But also check you're not relying on buffer-age. Just in case...01:37
bschaeferduflu, sounds good01:43
bschaeferduflu, im not 100% the issue, but ill dig around some more :)01:43
duflubschaefer: Also see if SDL2 on X11 has the same bugs ;)01:43
dufluQemu's SDL2 is new and possibly buggy in damage tracking01:44
bschaeferduflu, works fine on X11 :(01:45
bschaeferwell actually01:45
bschaeferi cant get the ubuntu iso install to come up01:45
bschaeferpossibly a bug there yeah01:45
duflubschaefer: OK then any simple SDL2 demos have similar issues?01:45
bschaeferduflu, nope01:45
bschaeferonly sdl1.201:45
bschaeferwith software rendering01:45
bschaeferwhich i fixed with frame damage01:46
bschaeferduflu, i also dont know qemu very well or how to use haha01:47
duflubschaefer: BTW if the mouse doesn't interact properly try: qemu -usbdevice tablet01:47
bschaefero nice, i've not even gotten to try the mouse out01:47
bschaeferas the keyboard works01:47
bschaeferso thats what i've been tessting01:47
dufluRunning Qemu in tablet mode will avoid it wanting grabs (which Mir lacks)01:47
bschaeferyeah, thats something racarr__ brought up :), my main goal was to make pixels look good01:49
bschaeferduflu, the only way to get sdl2 running as well is disable gtk and set sdl to 2.001:49
* bschaefer hopes its in their damage code01:49
duflubschaefer: You mentioned 1.2... I thought you didn't support 1.2?01:55
dufluIf you do then Qemu's SDL1.2 support is much more mature01:55
bschaeferduflu, i have a branch for it, but they aren't accepting patches :(01:55
duflubschaefer: Oh, fair point01:55
bschaeferduflu, another reason, is they are creating a layer to turn 1.2 SDL API to 2,.001:55
bschaefer2.001:55
bschaefersoo onces that layers there, we'll get all of 1.2 for free :)01:56
bschaeferbut yeah, i should test it out on 1.201:56
RAOFTo the luncheon wagon!02:34
=== chihchun_afk is now known as chihchun
=== chihchun is now known as chihchun_afk
=== chihchun_afk is now known as chihchun
dufluping anpok06:20
anpokpong06:24
anpoki have the same gdb issue you had06:25
anpokor still have?06:25
dufluanpok: Yeah, it's annoying but in other news...06:26
dufluanpok: Can you check this out? Seems to have only regressed last night: https://bugs.launchpad.net/bugs/132107706:27
ubot5Ubuntu bug 1321077 in Mir "[regression] [input] Scroll events are now always zero: event.motion.pointer_coordinates[0].vscroll" [High,Triaged]06:27
anpokoh06:27
alf_anpok: alan_g: Have you seen https://bugs.launchpad.net/mir/+bug/1321215 locally?10:03
ubot5Ubuntu bug 1321215 in Mir "CI failures in CustomInputDispatcherFixture.custom_input_dispatcher_gets_started_and_stopped" [High,New]10:03
alan_galf_: no. But I've not been looking for it.10:05
alf_alan_g: I wasn't looking for it either, it came looking for me :)10:06
* alan_g thinks CustomInputDispatcherFixture.custom_input_dispatcher_gets_started_and_stopped is a redundant name10:06
alan_galf_: let me grab latest updates and try it10:06
=== jono is now known as Guest73754
alan_galf_: this is just running on the phone? Or do you see it on desktop too?10:13
alf_alan_g: I can reproduce it locally on the desktop 100%10:13
anpokalf_: me neither10:13
alan_galf_: I hit it after 232 iterations on desktop10:14
alf_alan_g: anpok: thanks, I have started investigating it, stay tuned10:14
alf_anpok: when/where is InputDispatcher.start()/stop() supposed to be called ?10:15
alan_galf_: while you're there: CustomInputDispatcher.gets_started_and_stopped would be a better name.10:16
anpokalf_: as part of the DisplayServer start and stop proces..10:16
alf_alan_g: ack10:16
anpokoh it seems to not wait for completion10:16
anpokalf_: when I terminate the server process manually it did not fail within 2000 tries (which it did before) but I thougt the text fixture tear down does that already..10:44
alf_anpok: I am almost done with an alternative solution, I will ping you10:44
anpokoh it still fails10:44
anpokjust need to have more load10:45
anpokok10:45
Saviqhey guys, we seem to have an issue on android with the new Qt 5.3: https://bugs.launchpad.net/unity8/+bug/132118911:23
ubot5Ubuntu bug 1321189 in Unity 8 "Launcher is black on Qt 5.3" [High,Triaged]11:23
Saviqfor some reason ShaderEffectSource, which is a component in QML that is, in essence, creating a texture from a QML element, for consumption in a shader11:24
Saviqstopped working11:24
Saviqeverywhere it's used, it goes black11:25
Saviqit's fine on X and on Mir on desktop, but affects all our three android devices11:25
Saviqand we're out of our depth on this now11:26
greybackSaviq: are any shader errors reported on the console?11:26
Saviqgreyback, no11:27
greybackSaviq: and even a simple qml file using a ShaderEffectSource has this problem?11:28
Saviqgreyback, yeah http://paste.ubuntu.com/7492386/11:28
Saviqgreyback, is black instead of red, see https://bugs.launchpad.net/unity8/+bug/1321189/+attachment/4116145/+files/mako.png vs. https://bugs.launchpad.net/unity8/+bug/1321189/+attachment/4116146/+files/mako_good.png11:28
ubot5Ubuntu bug 1321189 in Unity 8 "Launcher is black on Qt 5.3" [High,Triaged]11:28
Saviqfooking barrier11:28
greybackSaviq: first step I'd take: capture an API trace - https://github.com/apitrace/apitrace11:32
greybackis available in the repos11:32
alf_anpok: https://code.launchpad.net/~afrantzis/mir/fix-1321215/+merge/22022811:35
RAOFalan_g: Yo! Are you available to talk about 1hz-rendering?11:44
alan_gRAOF: If you want me to11:45
RAOFI've written a follow-up comment. After you've read that, what would you like to see done?11:46
* alan_g goes to look11:47
RAOFgreyback: There'll be a shiny new apitrace 5.0 in the repos soon, too :)11:51
greybackRAOF: good, because 4.0 chokes on eglBindAPI :)11:51
RAOFgreyback: That might be awkward11:51
* greyback wants to try vogl too11:51
alan_gRAOF: I'm not convinced by the abstraction or the naming. But I do like the overall solution. I could just abstain.12:07
RAOFalan_g: I'll need to fix the merge conflict anyway (tomorrow); perhaps alf_ could weigh in, also.12:08
RAOFalan_g: Adding the last bits of frame dropping policy to FrameDroppingPolicy seems sensible to me, at least.12:09
alan_gMaybe SwapStateObserver <- FrameDroppingPolicy <- TimeoutFrameDroppingPolicy?12:19
RAOFThat would make sense. BufferQueue would still construct a FrameDroppingPolicy from a FrameDroppingPolicyFactory, though. It'd just store it in a std::unique_ptr<SwapStateObserver> rather than std::unique_ptr<FrameDroppingPolicy>12:21
=== alan_g is now known as alan_g|lunch
* RAOF → sleep12:31
Saviqgreyback, so yeah, it choked here on eglBindAPI13:02
Saviqgreyback, pointers?13:02
greybackSaviq: recompile qtubuntu without that call. Frankly, that call should not be happening on arm13:02
Saviqoh ya13:03
Saviqy13:03
SaviqRAOF, anywhere I could get the 5.0 of apitrace already?13:03
greybackSaviq: tried it, chokes there too13:03
Saviqgrr13:03
greybacksorry13:03
Saviqok me builds13:04
kgunnalf_: alan_g|lunch AlbertA ... so do i understand it correct on 1hz-rendering-always MP, we seem ok with the approach for putting policy in bufferqueue, but we would want a follow up second effort for shell to be able to override policy ?13:08
kgunnsorry if i missed any irc chat on this...but i've been keeping an eye on this one, hoping we'll land it soon (as to put some road mileaage on it before promoting)13:09
=== alan_g|lunch is now known as alan_g
alan_gkgunn: There's no real objection to separating a "policy" out - just a lack of clarity about what's separated out and how best to name it. (Although IMO separating out a configurable "policy" is premature until we can see what (if anything) needs separating out.)13:22
kgunnalan_g: ack13:25
alan_gYou could have missed some IRC between RAOF and me about 90mins ago: There'll be a new version his "tomorrow". He's got some conflicts to address before it can land, and I'm happy enough with the general approach not to block.13:30
kgunnalan_g: thanks13:33
kgunngreyback: so when you overrride displaybuffercompositor13:33
kgunnwill you auto-magically also use hwc ?13:33
kgunnor i'm guessing you'll need to "do some effort" there13:33
kgunn(which is hopefully a copy paste)13:34
greybackkgunn: no, effort needed. bypass might work tho13:34
kgunnah...cool13:34
kgunnso we architected it right :)13:34
mterrygreyback, btw, I filed https://code.launchpad.net/~mterry/unity-mir/dont-hide-focused-apps/+merge/220130 yesterday13:59
mterryAbout the hiding of suspended apps we talked about13:59
=== dandrader is now known as dandrader|afk
=== dandrader|afk is now known as dandrader
AlbertAkgunn: not so much sheller but other parts of the system should AUGMENT the policy14:31
AlbertAkgunn: meaning the policy should only be activated on display off/occlusion events14:31
=== alan_g is now known as alan_g|tea
=== alan_g|tea is now known as alan_g
greybackmterry: I added comment on your MR15:01
=== seb128_ is now known as seb128
kdub_greyback, (kgunn) with that convo about getting bypass or hwc to work about 3h ago... would something like this (which works with hwc) get hwc to work for you?15:53
kdub_http://bazaar.launchpad.net/~kdub/mir/future-default-display-buffer-compositor/view/head:/src/server/compositor/default_display_buffer_compositor.cpp15:53
kdub_thats my integrated hwc-working branch (stuck behind some other branches at the moment)15:54
kdub_I also want to rip out that weird lines 60-62 and lines 82-8615:54
greybackkdub_: almost, main issue I see is that your use of HWC is an if/else. I'll probably have to render, but also call post_renderables_if_optimizable(renderable_list) - because the scenegraph contains things that are not Renderables15:57
greybacke.g. a non-fullscreen app, for iteration one, the panel is not in its own surface, but drawn explicitly with GL calls15:58
kdub_greyback, ack, that's the key, is to condense things into renderables where its simple to do so, and use gl where it can't be represented by a RenderableList16:00
=== chihchun is now known as chihchun_afk
greybackkdub_: agreed, but this is iteration 116:00
kdub_because that abstracts the hwc stuff away, and the compositor writer just thinks in terms of 'can this be a RenderableList'16:00
greybackkdub_: so in time, we can move to using RenderableList as much as possible16:00
greybackbut we're not going to be there for a while yet16:00
kdub_greyback, cool, just checking that its still pulling in a sensible direction16:01
greybackkdub_: yep, no objections so far16:01
=== dandrader is now known as dandrader|lunch
=== alan_g is now known as alan_g|EOD
=== dandrader|lunch is now known as dandrader
=== dandrader is now known as dandrader|afk
=== dandrader|afk is now known as dandrader
=== beidl_ is now known as beidl
RAOFSaviq, greyback: If you want it, http://anonscm.debian.org/gitweb/?p=pkg-xorg/app/apitrace.git;a=shortlog;h=refs/heads/master is where you can find apitrace 5.0; git buildpackage will know what to do with it.23:08

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!