/srv/irclogs.ubuntu.com/2014/04/15/#ubuntu-mir.txt

kgunnRAOF: hey, do you happen to know if/how xmir install/enabling changes with unity8-desktop-session-mir ?00:12
kgunnracarr_: ^ ?00:12
RAOFkgunn: Yes; is there anything in particular that you want to know?00:12
kgunn....so i'm just wondering, unity8-desktop-session-mir installs unity-system-compositor....00:13
kgunnand installing u-s-c was the mechanism to install xmir00:13
kgunn..i assumed it still was ?00:13
kgunnbut guess it must not be true00:13
kgunn...wonder, now, one must go modify /etc/lightdm/lightdm.conf.d/unity-system-compositor.conf ?00:14
kgunni was thinking to test this...00:15
kgunnwondered if you knew off the top of your head00:15
racarr_kgunn: I think ubuntu-desktop-mir enables xmir00:20
racarr_err00:21
racarr_which is pulled in by unity8-desktop-session-x11 what?00:21
racarr_so lets hope not lol00:21
kgunnracarr_: ah-ha...unless, you already had a usc conf file...? which had #type-unity modified :)00:21
racarr_kgunn: I dunno, packages can either override or not override conf files...would have to look deeper00:23
RAOFkgunn: Yeah, if you already had a usc config file lying around it wouldn't overwrite it.00:24
RAOFkgunn: Assuming that the config file that the package wants to install is exactly the same as the config file at the time you modified it.00:25
kgunnok...gonna just see if it still works, rebooting00:27
RAOFThis bodes.00:35
kgunnuh well...that was unpleasant...00:45
kgunnusc log says https://pastebin.canonical.com/108519/00:46
kgunnRAOF: ^ in case something comes to mind...it could easily be something of mine00:55
kgunnsetup incorrect...00:55
kgunnracarr_: totally...bag o candy, 8o clock...i got a sick belly now :(00:56
* RAOF will check just as soon as his SSO 2fa device finishes being unresponsive.00:57
RAOFkgunn: Hrm. Looks like we failed to do drm setup there, which we should be able to do. I wonder if things like plymouth are being silly?01:05
kgunnRAOF: wanna join meeting ?01:06
kgunnit'll be quick one01:06
RAOFSure.01:06
RAOFWas rather expecting it to be at 13:0001:06
racarr_kgunn: Oh no :(01:15
racarr_oh01:15
racarr_did I miss01:15
racarr_meeting01:15
racarr_I guess not that much to say01:16
racarr_RAOF: I am writing X Input driver01:16
racarr_I wont be able to get back to it until wednesday but its almost all done01:16
RAOFracarr_: Hurray!01:16
racarr_was going to test some basics of it but then had intel driver mis match...and01:16
RAOFracarr_: Also, meeting is on now. Come on down!01:16
racarr_I dunno probably just built wrong stuff during sprint01:16
kgunnthomi: are there specific instructions to the "Run the autopilot functional and unit tests" for desktop ?02:01
kgunnor just read the readme?02:02
thomikgunn: I can show you here, if you like02:02
thomiyou have the packages installed from the landing silo?02:02
kgunnthomi: just upgrading...will install as soon as that completes02:02
thomiok02:03
thomiwell, when that's done, opena  terminal and run:02:03
thomiautopilot run -o ap_2_results autopilot ; autopilot3 run -o ap_3_results autopilot02:03
thomiyou'll want to double check that all the AP-related packages from the silo have been installed02:04
thomiespecially: python-autopilot python-autopilot-tests python3-autopilot python3-autopilot-tests autopilot-desktop02:04
kgunnthomi: ok...will do that02:05
kgunnRAOF: duflu...weird you both expected the meeting to be later, but i updated and saved it earlier today....hmmm02:06
kgunnis there a google-american-aussie-daylightsavings-bug somewhere?02:06
kgunnRAOF: any other ideas on my xmir failing to start? (i'm updating just in case)02:08
* kgunn wonders if there's a handy drm debug wiki02:08
duflukgunn: Well my timezone never moves. Only the meeting did. Not sure why02:12
RAOFkgunn: Can you switch to a VT and try restarting lightdm? Can you ssh in and do the same?02:13
* duflu is actually mostly impressed with being awake in the *morning* at all02:14
kgunnRAOF: i will try...kinda got too many spinning plates atm :)02:15
* duflu mixes fresh XMir with fresh coffee02:15
AlbertAkgunn: just fyi I built against mir devel r1550 for the mir-0.1.9 branches I sent you02:43
duflukgunn: Yeah using that PPA XMir doesn't start (normal X starts fine instead)03:02
dufluProbably because USC is never started03:03
RAOFHm. This test is becoming more complex than the code it's trying to test. Time to rethink.03:06
kgunnthomi: so...i ran autopilot run -o ap_2_results autopilot03:28
kgunnand it ran a bunch of stuff...03:28
kgunnno failures03:28
kgunnbut no summary either03:29
kgunnnormal ?03:29
thomikgunn: it'll be in the ap_2_results file03:29
thomithat's what the -o bit does03:29
kgunnthomi: thanks (i should've read the command line and thot a little)03:30
thomikgunn: :)03:30
kgunnall pass btw03:30
thomikgunn: sweet - now you gotta repeat with 'autopilot3' :)03:30
thomikgunn: I gotta head out for 45 minutes or so. If you need anything in the mean time, I'msure veebers can help you in #ubuntu-autopilot03:33
kgunn_RAOF, racarr_ interesting...so i'm on my clean machine, installing unity8-desktop-session-mir, which in turn installs u-s-c, did not create a /etc/lightdm/lightdm,conf.d/10-unity-system-compositor.conf file03:33
duflukgunn_: That would explain why XMir wasn't running either :)03:46
kgunn_duflu, nope, completely different machine...other machine, i had a 10-usc.conf file....03:49
RAOFkgunn_: Got ubuntu-desktop-mir installed?03:51
kgunnRAOF: on the new machine...no03:52
RAOFkgunn: That is the package with 10-unity-system-compositor.conf in it.03:52
kgunnhuh...not installed on this machine either03:53
dufluI knew that was true. But it's a problem for Unity8 it seems03:53
kgunnhmmm....ok, so unity-system-compositor no longer installs ubuntu-desktop-mir ?03:53
RAOFApparently so.03:53
dufluMaybe?... what does "ubuntu-desktop-mir" mean? Does it mean XMir or "XMir or Unity8"?03:54
kgunnduflu: afaict, unity8-desktop-session-mir....means unity803:54
kgunnon mir03:54
kgunnbut unity7 still on x (no mir)03:54
duflukgunn: So we either need to depend on ubuntu-desktop-mir or move the conf file to another package...03:54
kgunnwhereas ubuntu-desktop-mir means xmir (guessing there)03:54
kgunnor just change the instructions ?03:55
dufluAh but ubuntu-desktop-mir isn not XMir-specific, is it?03:55
kgunnRAOF: ^ ?03:55
* duflu checks package contents03:56
RAOFkgunn: Dunno; I didn't do those packages :)03:56
duflukgunn: The only purpose (only content) of ubuntu-desktop-mir is to provide 10-unity-system-compositor.conf. Unfortunately it depends on xserver-xorg-xmir, which I think we should uncouple03:57
duflu... and then unity8 can cleanly depend on ubuntu-desktop-mir. As should xserver-xorg-xmir (i.e. reverse the current dependency)03:59
* duflu now wonders why the USC conf is not included in USC packages04:00
dufluThat would make more sense -- I have USC installed and therefore I expect it to start\04:00
RAOFI don't know if unity8-desktop-session-mir is expected to work on a raw VT or not.04:01
kgunnso on my clean machine, just installed ubuntu-desktop-mir...still no conf file installed04:01
RAOFI don't know why unity8-desktop-session-x11 depends on ubuntu-desktop-mir, though.04:01
kgunnok..i'll try again tomorrow...i need to tie loose ends and go to bed04:01
kgunnRAOF: duflu ...btw, silo3 ppa may actually build (...making it sucessfully thru usc now)04:01
kgunnso if you wanted to test xmir with https://launchpad.net/~ci-train-ppa-service/+archive/landing-00304:01
kgunnand send a mail...that'd be great04:01
duflukgunn: I think we should abolish ubuntu-desktop-mir. It's confusing and it's only contents (the conf file) would be better placed in the USC package04:01
duflu-it's +its04:02
kgunnduflu: might file a bug and flag bregma to stay in sync04:02
RAOFThat seems sensible.04:04
dufluWell as usual it's a simple change to make... if only I can find the branches (often the greatest challenge)04:04
dufluOn second thoughts, I don't know lightdm that well. Might be missing something04:06
dufluRAOF, all: Please?https://code.launchpad.net/~vanvugt/mir/version-0.1.9/+merge/21580405:24
dufluThere's more to come too05:25
alf_Saviq: Hi! Where does unity8 keep its log on the phone (e.g. if I print to stderr from the unity8 process)?09:45
anpokalf_: ~/.cache/upstart/unity8.log.*10:11
=== alan_g is now known as alan_g|afk
=== alan_g|afk is now known as alan_g
=== alan_g is now known as alan_g|lunch
=== alan_g|lunch is now known as alan_g
=== greyback_ is now known as greyback|lunch
=== greyback|lunch is now known as greyback
=== dandrader is now known as dandrader|afk
=== dandrader|afk is now known as dandrader
=== alan_g is now known as alan_g|tea
=== alan_g|tea is now known as alan_g
josharensonWhen I try to cross compile mir (in my own feature branch) get an error about the boost libs not being found. Compiling any other branch works just fine... Any pointers? I have changed a lot of cmakefiles, but nothing that should have affected libboost15:52
alan_gjosharenson: are the boost libs in your cross-compilation environment and not being found or are they not being installed there?15:54
josharensonalan_g not being found I believe15:55
josharensonhttp://pastebin.ubuntu.com/7256002/  shows the first of several errors15:55
alan_gjosharenson: You're setting CMAKE_PREFIX_PATH to somewhere useful?15:58
alan_gOops PKG_CONFIG_PATH15:58
josharensonalan_g never had to do that before... should I try setting it to /usr/bin ?15:59
alan_gNo, to wherever you're installing your target libraries15:59
josharensonok16:00
alan_gs/libraries/packages/16:00
alan_gE.g. from cross-compile-chroot.sh : export PKG_CONFIG_PATH="${MIR_NDK_PATH}/usr/lib/pkgconfig:${MIR_NDK_PATH}/usr/lib/arm-linux-gnueabihf/pkgconfig"16:00
josharensonalan_g yes I tried that manually and via the script16:03
josharensonre-downloading some deps over airplane wifi right now.....16:03
josharensonhummm seem to be getting 404s for all armhf repositories16:07
=== dandrader is now known as dandrader|lunch
=== alan_g is now known as alan_g|EOD
=== dandrader|lunch is now known as dandrader
kgunnracarr_: kdub ...so that whole "scene observer" discussion going on, is that going to be another hit on server api ?19:26
kgunni'm just starting to wonder if we should just plan on cherry picking the non-blocking egl (...feels like an endless api update exercise otherwise)19:26
racarr_kgunn: Yeah it would be19:35
kgunnracarr_: so do you think we're killing ourselves trying to work this particular item at high priority while also keeping step with all the other churn ?19:37
kgunnAlbertA: thots ^ ? guess you suffered y'day...19:39
AlbertAkgunn: I've rebased the branches to tip again locally19:39
AlbertAkgunn: well actually only papi was affected19:40
AlbertAkgunn: trying to cross-compile to test19:40
kgunnAlbertA: on papi was it the removal of the placement strategy ?19:41
AlbertAkgunn: yeah19:41
racarr_kgunn: Ah sorry. It's19:46
racarr_server ABI19:46
racarr_but not sever API19:46
kgunnAlbertA: ok..just worrying there's more to come e.g. "scene observer" , cursor stuff, input dispatcher,19:46
racarr_cursor stuff also breaks ABI but not API19:46
kgunnracarr_:  true..that makes it less churn-y19:46
AlbertAkgunn: for 0.1.9?19:47
AlbertAkgunn: when do we expect to tag?19:47
kgunnAlbertA: yeah...since its not really tagged19:47
kgunnwell...that's just it, if we take a "wholesale" approach...we need to tag _after_ we get mir good to go for non-blocking swap19:47
kgunnif we take a cherry pick approach...it'll just be rework pain for alf19:48
racarr_ok well all this stuff doesnt have to land quite let19:49
racarr_yet19:49
racarr_no one minds MPs sitting around if everything is an approve19:49
kdubkgunn, I have something for the android overlays to work on that shouldn't hit much server api, I can switch to that for a bit so there's less churn from me for a few days19:49
racarr_imo19:49
AlbertAkgunn: yeah theres's only 4 MPs ready to land which should not break api19:50
AlbertAkgunn: I'm holding off on the one that does (add uid/gid) after tagging19:50
* kdub to lunch19:54
kgunnthanks AlbertA, kdub, racarr_ ...if we can keep it a bit more "calm" for some days it'll help for sure19:55
kgunni might still get my arm twisted19:55
kgunnby rel team to cherry pick19:55
racarr_ok well keep it cool20:02
racarr_hmm20:48
racarr_this scene observer + register surface observer approach ends up being really annoying for the compositor20:49
racarr_because its hard to unregister20:49
racarr_oh nvm I can just move the20:49
racarr_started/stopped check in to the callback, it already owns the right lock20:49
racarr_so its the same20:50
racarr_lalala20:50
AlbertAracarr-21:13
AlbertAracarr_: note that logic is about to change a little bit21:13
AlbertAracarr_: the compositor start/stop state21:13
racarr_AlbertA: Mm..I think it will be fine21:15
racarr_we will find out soon :p almost to that part21:15
AlbertAkgunn: there's a deadlock issue with non-blocking21:26
kgunnAlbertA: awesome :-/21:26
kgunnhow is that possible ?21:26
kgunn....its non blocking :)21:26
kgunnlol21:26
AlbertAkgunn: he yeah It does sound ironic21:26
kdubhaha21:26
kgunnat least it made me laugh a little...21:26
AlbertAkgunn: if you turn off the screen21:27
AlbertAkgunn: and try to turn it back on it will never turn back on21:27
AlbertAkgunn: it's waiting for allthe compositors to stop21:27
AlbertAkgunn: but at the same time, the display is waiting on a power on  condition so it can unblock21:28
AlbertAkgunn: so deadlock21:28
AlbertAkgunn: what was the issue alexandros was looking at? that or something else?21:29
kgunnAlbertA: so is it basically that it needs your powerd change to land first?21:31
AlbertAkgunn: no it's real deadlock21:31
AlbertAkgunn: just trying to understand how we can fix it21:31
kgunnAlbertA: he was looking at 2 things...1 was apps in background still blocked21:31
kgunnand then 2 some apps seemed to not render (or render beneath the shell)21:32
AlbertAkgunn: ah I see what's happening...21:34
AlbertAwe turn stop the compositors, turn off the display, start the compositors again21:35
AlbertAthe display buffer compositors then try to post to display buffer21:35
AlbertAwhich will be locked21:36
AlbertAwaiting for a power on change21:36
AlbertApress the power button again, we try to stop compositors21:36
AlbertAand we wait forever since one of them is blocked21:36
kgunnyou mean even with the change we "stop the compositors" ?21:37
AlbertAthe display buffer compositor yeah21:37
AlbertAwhich makes sense21:37
AlbertAwell21:37
kgunnb/c they can't touch FB's at that time ?21:37
AlbertAkgunn: yeah21:38
racarr_RAOF: Can you review cursor spike phase 1 again when you have time?21:38
kgunnAlbertA: i'd agree with that...yeah, you gotta stop, to just "release" the FB's21:38
AlbertAkgunn: I guess I can fix it in usc21:39
kgunnAlbertA: i guess its "after display off, compositors start again"...why does it try to post to FB ?21:40
AlbertAkgunn: if we are going to power on, power on display first21:40
kgunnAlbertA: oh, you're assuming a rapid sucession of power button striking...e.g. this specific case is its "going to on"21:40
AlbertAkgunn: why? because the multi threaded compositor has no idea about the display buffer power state21:41
AlbertAkgunn: so it just receives start/stop21:41
AlbertAkgunn: this is specific to going off then going on, it happens all the time for me21:41
AlbertAkgunn: I don't see how it cannot actually21:42
AlbertAkgunn: unless you are really fast yeah...then you shoulnd't see it :)21:42
kgunnAlbertA: oh wait...you mean its not rapid...reaching steady state for off21:42
AlbertAkgunn: yeah just a normal power off21:42
AlbertAkgunn: then a casual power on21:43
kgunnAlbertA: so when pressing the button to go to "on"...why do we try to "stop the compositors" again ?...e.g. aren't they in effect already stopped ?21:46
kgunni guess state-wise, they're technically not21:46
kgunn?21:46
AlbertAkgunn: yeah that's what I'm changing now21:47
AlbertAkgunn: if on = just turn on the display21:47
kgunnright21:47
AlbertAkgunn: if off = stop compositors, change display state, turn compositors back on21:47
AlbertAkgunn: that will mean that we'll still have at least one stale buffer though21:50
AlbertAkgunn: but I suppose we can work on that issue21:51
AlbertAkgunn: later21:51
kgunnAlbertA: yeah...its no worse than what we have at the moment21:51
kgunnwhich is like a whole pipeline :)21:51
AlbertAkgunn: ok this works21:53
AlbertAkgunn: anything in particular I need to test for non-blocking?21:53
kgunnAlbertA: hit power button, screen off with music playing, try volume21:54
kgunnif volume keys work, then gui thd is svcing events21:54
AlbertAkgunn: send mp3 through adb?21:55
kgunnAlbertA: shamefully, i dunno :)21:55
AlbertAkgunn: well volume keys not working21:59
kgunndamnit22:00
AlbertAkgunn: btw yeah sending mp3 through adb to /home/phablet/Music worked fine22:01
kgunnAlbertA: this all seems strange...i don't think alf_ & greyback ever saw the display off deadlock, and the volume keys worked22:01
AlbertAkgunn: strange..let me see the mir history22:02
kgunnjust thinking with them specifically testing screen off....they would see that22:02
kgunnalmost sounds like the compositors didn't "turn back on" in the sequence of comp-off, change-dpy-state, comp-on22:06
AlbertAkgunn: I was testing with the fix proposed today, let me back that up22:08
AlbertAkgunn: well I don't know what I'm missing...22:16
AlbertAkgunn: I'm using the updated 0.1.9 branches for umir/usc/papi22:16
AlbertAkgunn: and mir r1557 with alf's patch on top22:17
AlbertAkgunn: oh I think I may need a unity change no? I think by default they pause when they receive the display off22:20
AlbertAnotification22:20
AlbertAShell.qml?22:20
kgunnAlbertA: yeah...actually i thikn greyback commented out his changes in qtubuntu right before he EOD'd22:30
kgunnhe said something about it not doing what he thot22:30
AlbertAkgunn: well I took out the greeter.show after a power off is received22:31
kgunnhmmm...22:31
kgunnhttps://code.launchpad.net/~gerboland/qtubuntu/surface-visible-hidden-side-channel/+merge/21588422:31
AlbertAkgunn: that did the trick22:31
kgunnhe has attempt 222:31
kgunnoh cool...22:31
AlbertAkgunn: in Shell.qml22:32
AlbertAkgunn: I guess that triggers Qt's window management22:33
kgunnAlbertA: does it do that pre-emptively ?22:34
kgunnoh...yeah, maybe ?22:34
kgunnmterry: might know ^22:34
* mterry looks22:34
kgunnmterry: mainly, the greeter.show from shell in a display off case22:35
mterrykgunn, er, what problems does the greeter.show cause?22:36
AlbertAmterry: the volume keys didn't work with the non-blocking egl swapbuffers branches22:37
AlbertAmterry: after greeter.show22:37
AlbertAmterry: if I take it out, during screen on, the volume keys work while playing music22:37
AlbertAsorry screen off22:38
mterryAlbertA, OK...  so the shell is trying to render frames while screen off, and it can't take input as a result...22:38
mterryAlbertA, I wouldn't say that would involve Qt's window management, unless you mean a low-level window.  The greeter is just a Qml object being slid offscreen and on22:39
AlbertAmterry: but it should be able to render now with the non-blocking implementation22:39
mterryAlbertA, right...  Is it greeter specific?  Like, if you show something else..  Like maybe the launcher or something22:40
AlbertAmterry: you mean do that in Shell.qml?22:41
AlbertAmterry: of just leave the launcher on, hit power off22:41
mterryAlbertA, yeah, keep greeter.show() commented out, but maybe do launcher.show() instead22:41
AlbertAmterry: let me see22:41
mterryAlbertA, just as a test to see if it's just anything with frames, or if greeter is doing something insane22:42
AlbertAkgunn: I'm wondering now if eglSwapBuffers is blocking at the driver this time22:42
kgunnAlbertA: very well would....at the driver level it would22:43
kgunnmost likely22:43
AlbertAkgunn: oh...maybe we are hitting autosuspend22:45
AlbertAkgunn: nah but then the music wouldnt' play22:46
racarr_*not entirely sure I undewrstand the context*22:46
racarr_but in this case eglSwapBuffers is mir egl swap buffers right so22:47
racarr_I mean it only blocks if we block it?22:47
AlbertAracarr_: it still goes through the driver first though22:47
kgunnAlbertA: the music can play ok...it just won't advance to the next song or vol change22:47
kgunngotta restart....22:48
AlbertAmterry: launcher.show() doesn't do anything, It doesn't show the launcher22:49
mterryAlbertA, oh really?  let me see Shell.qml22:49
AlbertAmterry: should that me showNow()?22:49
mterryAlbertA, how about launcher.tease() then22:50
mterryAlbertA, show() is not used, because the launcher is always dragged, not programmatically shown22:50
AlbertAracarr_: so it goes through the driver then the driver calls us through AnativeWindow methods for android drivers22:52
racarr_yes. I guess I just assumed it doesnt do any direct22:52
racarr_hardware access itself or22:53
racarr_try and do anything fancy but who knows22:53
AlbertAwell you never know, it's an egl context after all22:53
AlbertAmterry: yeah that worked22:53
mterryAlbertA, volume worked too?22:53
AlbertAmterry: nah same issue22:54
AlbertAmterry: which means we really do need Gerry's side channel branch22:54
mterryAlbertA, OK, well good that greeter isn't doing something weird.  Just general qml stuff then22:54
AlbertAmterry: right22:54
AlbertAkgunn_: so it looks like it still blocks when qml tries to render things22:56
AlbertAkgunn_: maybe I'm not pushing the correct libraries for mir nested22:58
AlbertAkgunn_:I 'll check that22:58
kgunn_ack22:58
AlbertAkgunn_: well that wasn't it, I can see in the log the fake compositor is posting23:03
AlbertAkgunn_: both in usc and unity823:03
AlbertAkgunn_: I guess time for gdb23:04
kgunn_man....23:07
kgunn_definitely leave the findings in a mail for gerry & alf23:07
AlbertAkgunn_: the good news is that it doesn't seem to be blocked at the driver23:16
AlbertAkgunn_: it's blocked at the client side waiting for the next buffer23:16
AlbertAkgunn_: yeah I'll put all this on an e-mail23:16
AlbertAkgunn_: ah I think I know why23:17
kgunn_....cliffhanger....why ?23:18
AlbertAkgunn_: well since the display compositors are blocked23:18
AlbertAkgunn_: they are never going to release the buffer they are holding23:19
AlbertAkgun__: so eventually we consume all buffers from client but we have to return in sequence23:19
kgunn_so it just blocks on buffers being full...23:19
AlbertAkgunn: and that only happens when all compositors release their reference23:19
AlbertAkgunn: so we probably need a way to just start the fake compositor only23:20
AlbertAkgunn: at display off23:21
kgunn_this is what duflu was hinting at i think...wrt unsync'd swapinterval023:21
AlbertAkgunn: uhhh, yeah we have to guarantee sequence23:22
AlbertAkgunn: but that's ok, the fake compositor will guarantee that23:23

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