[00:01] in this little demo shell the application switcher event filter is constructed from in the_event_filters from, but needs to get the_frontend_shell() hich needs the_input_focus_selector() which needs the_event_filters() ;) [00:02] hmm. I don't think I have moved more than 15 feet since tvoss went to sleep [00:13] ugh - as an interim hacky solution, benchmark data is being stored here: https://code.launchpad.net/~mir-team/+junk/benchmark-graphs - so please nobody delete that branch ;) [00:53] weird problems with my demo shell server configuration...seems the contents of my std::initializer_list are being lost in passing around [01:08] fbnativewindow defeats me another day :/ [01:09] aww [01:40] hmm, neat http://codicesoftware.blogspot.com/2013/04/put-your-hands-on-programming-language.html [01:40] language aware merge tool [02:11] Hm. Interesting. Build Mesa with clang and Mir segfaults. [02:23] Woot! [02:23] We draw! [02:30] RAOF: Still building Mir with gcc? [02:30] Yes. [02:31] Fun fact: A radeon 5350 can't render mir_eglplasma at 1920x1200 at 60 FPS [02:31] RAOF: My first guess is accidental writes to const variables (which are memory protected under clang) [02:31] RAOF: I'll look at optimizations :) [02:41] RAOF: Oh. eglplasma is trying to do 1.3 billion single precision cosines per second :/ [02:41] That's… quite a lot :) [02:42] RAOF: I've been thinking for some cases (like this) it may make sense to create a low-res surface and flag it as needing upscaling. The other case I thought of was simple games [02:43] That's a part of the "fullscreen" request, isn't it? [02:43] You say "make me fullscreen" with flags such as "and please change the output's resolution" or "upscale me if necessary" or etc. [02:44] RAOF: Yeah in the real world. Though some of our work-items are marked as DONE without such forethought :/ [02:44] I've been guilty of similar [02:44] Oh, have we marked fullscreen as done? [02:44] RAOF: No, that one's in progress pending my state work [02:45] [02:45] MY NEMESIS! [02:47] RAOF: What was your frame rate BTW? [02:47] 40 FPS [02:48] Which is not bad for a two-generations-ago low-power GPU. [02:48] RAOF: Hah. I call that a great success then [02:52] Hmm, we're out by two orders of magnitude. Best I can find, AMD advertise the 5450 as "Processing power (single precision): 104 GigaFLOPS" [02:53] Though it's likely that cosine takes many FLOPS [02:53] This might also be a limiting factor: Pixel fill rate: 1.6 Gigapixels - 2.6 Gigapixels/sec [02:54] I can't find specs for 5350. Seems to happen with all my graphics cards too. They're always so low-end that the manufacturer does not publish their existence [02:55] You can find it on http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units - it's an RV710 [02:55] Which has specs around what you've listed. [02:58] Hm. With -O0 I can see why clang's failing - the glFlush pointer appears to be garbage. [03:00] glFlush has a "pointer"? [03:00] To the context? [03:01] The dri2_drv structure has a glFlush function pointer. [03:01] ...which is populated via dri2_drv->get_proc_address("glFlush"), which is presumably the thing that's broken. [03:02] ...which, in turn, is dlsym()'d from _glapi_get_proc_address [03:03] And clang does throw a warning somewhere in the mapi/glapi build, so that might be it. [03:04] And that suggests that the problem is not in *my* code, and we build mesa with gcc, so I'm going to call it good :). [03:07] Sounds a bit odd. Plain old C issues with clang should be easy to resolve. I surprised Mesa is not yet clang-friendly [03:11] Oh, it's Wednesday! [03:12] That means it's SRU DAY! [03:19] * duflu wonders if the SRU/SRY pun has been overdone yet [04:19] behold: mir/glmark benchmark graphs: http://people.canonical.com/~thomir/ [04:19] errr... kind of hacky, I know [04:19] now the preasure is on RAOF to fix that bug in mir :P [04:19] err, mesa [04:22] The pleasure's all mine! [04:28] https://bugs.launchpad.net/mir/+bug/1167133 this stuf is weird [04:28] Launchpad bug 1167133 in Mir "Input Filters fail to work when overriden through server configuration" [Undecided,New] [04:29] thomi: I see only 59 fps on one of the tests [04:29] I guess someone is to be fired. [04:29] isn't fps a bad metic ? [04:29] :p [04:29] yes [04:30] espescially when its capped to vsync ;) [04:30] right, so let's fix the vsync issue [04:30] I thought I read this article on arstechnica about measuring frame time [04:30] it's just a start [04:30] if anyone has any better ideas, I'm open to suggestions [04:30] racarr: the clear answer to this one is to just make mir hot-patch monitor firmware to make them refresh faster [04:31] "hey guys" [04:31] "my monitor goes at 50,000fps" [04:31] smspillaz: What do you think mir is doing no?? [04:31] w* [04:31] "all thanks to mir" [04:31] we stole the code from beryl [04:31] racarr: you stole code from norwood? [04:31] err...^H^H^H [04:32] oh sorry, norwood is the protocol [04:32] I meant northfield [04:32] you stole code from northfield [06:04] duflu, you seem to have a very unreliable computer! [06:04] robert_ancell: It was working fine. Then I upgraded gtalk [06:04] alf_, racarr, meeting? [06:05] robert_ancell: ah, sorry I was waiting for a link on irc :) [06:05] alf_, np :) [06:06] kdub, ^ [06:12] Aaah, hangouts :) [06:22] robert_ancell: sorry, network died for a moment there [06:22] alf_, :) [06:24] thomi: btw, glmark2 prints frametime if you prefer to parse and display that (although you could also calculate it yourself easily) [06:24] Only if we're not vsyncd :) [06:24] Argh. Stupid audio subsystem. Now looks like system-wide, my audio goes to/from the wrong devices and ignores my selection of devices to use. It worked last week... [06:25] Wheee! [06:25] It's even broken in gnome-control-center [06:27] Oh, and my other headphones that would have worked recently disintegrated from age [06:55] alf_: yeah, I have frametime - if you think that's a better metric to use it'd be simple to switch [08:46] alf_: Do we need nouveau support for use-dma-buf before landing it? I've got radeon support. [08:47] Eh, I'll just implement nouveau support as well. It's not like it's hard. [08:47] erm [08:47] didn't I do that? :P [08:47] mlankhorst: That was in xmir; it needs mesa as well. [08:47] oh [08:47] glxgears seemed to have worked though [08:48] You probably didn't do that, because the gallium changes you need only just landed on github :) [08:48] ok [08:48] Yeah glxgears will work. es2gears_mir won't :) [08:48] RAOF: Feel free to add nouveau now or leave it for later, I don't think it blocks anyone. [08:49] it blocks meeeeeeee and I'm not a mesa developer [08:49] RAOF: ^^ well I guess you have your answer ;) [08:50] alright alright I'll poke mesa some [08:50] mankhorst: Ok. I'll do nouveau as well and land use-dma-buf tomorrow. [08:50] RAOF: I was sarcastic :P [08:51] Probably a better idea anyway; landing in the morning means I can put out fires at my leisure during the day. [08:51] i'll fix up nouveau and any bugs I come across myself [08:52] just put it somewhere [08:54] RAOF: ^ [09:19] mlankhorst: github.com/RAOF/mesa - egl-platform-mir-dma-buf and mir-ppa. [09:20] kk [09:21] You need to implement whandle->type == DRM_API_HANDLE_TYPE_FD in bo_from_handle and bo_get_handle. See also: https://github.com/RAOF/mesa/commit/4deaa1bd5377e6224eb266f6a5c8e602f666aeff [09:22] (You might notice the stunning kludge in that commit ) [09:23] aw, installing some armhf packages removes some amd64 ones [09:27] libboost doesn't like multilib, it seems :( === tvoss is now known as tvoss|test === tvoss|test is now known as tvoss === tvoss is now known as tvoss|sick === pete-woods is now known as pete-woods|afk [11:50] looks like my nexus 7 doesn't like mir either, I tried the instructions, but mir just crashes [12:03] done with nouveau I think, I'm ignoring the GART stuff as it's unneeded with the kernel patches.. [12:03] mlankhorst: for nexus 7, ping kdub, he should know more about what's up there [12:04] mlankhorst: something about libhybris mods required only for nexus7 & only to egl...something about pthreads....platform team is workin' on it [12:04] w kdub [12:04] sounds about right === ogra_` is now known as ogra_ [12:27] RAOF: mir_egltriangle seems to work for me(TM) [12:28] http://people.canonical.com/~mlankhorst/mir-mesa-nouveau.patch [12:43] it's against the dma-buf branch but with debian from your ppa-dma-buf branch, which is why i disabled the egl-platform-mir patch [14:59] Morning! [15:02] good morning [15:05] racarr: kdub: Good morning! [15:10] status: Proposed VT switching MP, investigating potential performance improvements [15:12] status, working on stabilizing my fb branch, i suspect i'm not using the hwc right somehow [15:30] * kdub likes how a fresh day brings fresh perspective on problems [15:30] mlankhorst, with the nexus 7, hybris at the moment, does not support process-shared pthreads (which the nvidia drivers need for sharing buffers with clients) [15:59] morning racarr [16:06] Hi Katie! [16:07] racarr, how's it going? do you have time for a brief hangout? [16:15] katie: Sure lets do it! [17:23] hmm xkbcommon-mapper test hangs on jenkins [17:23] for 120 minutes until jenkins quits ;) [17:44] comcast business = faster downloading of chroots woo === UbuPhillup is now known as UbuPhillup_ [18:56] kdub: Did another around of updates on enable-inprocess-egl [18:59] racarr, will look it over shortly [19:34] kdub: more to xkbcommon-mapper too (I think the latest jenkins error there is a network quirk should be fixed soon) [19:34] fixed build in with FindXKBCommon.cmake :) [19:54] morning all [20:01] morning [20:59] racarr, hello [21:35] robert_ancell: hey...how did the team meeting go? [21:35] robert_ancell: i literally woke up at 12:50 with no alarm....like an internal clock went off [21:36] kgunn, we had alf, RAOF, duflu me and thomi there [21:36] robert_ancell: i thot about joining....then thot better [21:36] mostly did a round of "what we're doing" then talked about some CI stuff [21:36] kgunn, good idea :) [21:36] robert_ancell: bummer...no racarr or kdub? [21:36] kgunn, no :( [21:36] * kdub is here [21:37] kgunn, we found it funny that we moved the meeting to a better time but had less people! [21:37] kdub: past tense [21:37] robert_ancell: and alan was out [21:41] robert_ancell: so did you know what to do after leaving that prd review with richard? [21:42] kgunn, I didn't think we had any actions other than to keep him up to date on planning changes [21:43] robert_ancell: i suppose we did agree to split out mir into pieces...like, input support, system compositor, app copy/paste etc [21:43] kgunn, the PRDs not editable by us right? [21:43] kgunn, and those splits are reflected in the blueprints [21:45] robert_ancell: prd isn't editable but the m2m planning one is.... [21:45] i'm gonna take a crack [21:45] i'll let you correct the damage :) [21:46] kgunn, heh :) [21:47] I just finished configuring jenkins to graph mir autolanding build times. The result is: http://static.inky.ws/image/3841/showChart.png [21:47] tl:dr: it's getting longer [22:01] robert_ancell: got a minute? [22:01] thomi, sure [22:02] robert_ancell: so WRT the lightdm autolanding, we need to hack the packaging stuff a bit. Lots of the PS projects have started using inline packaging, which makes it easier to keep the packaging & source code synced [22:02] so I could either propose a MP that adds the debian/ dir to lp:lightdm, *or* I could make a new branch: lp:lightdm-team/lightdm/trunk-packaging [22:02] it's up to you :) [22:04] thomi, yeah, I'd prefer the packaging was separate as lightdm is used outside of ubuntu [22:04] robert_ancell: sure, OK. [22:06] I think there's some script that strips out the debian/ dir for "upstream" releases for Unity... but that's some deep juju that I have nothing to do with [22:10] robert_ancell: hey is duflu going to propose his surface state stuff this week? (curious) [22:10] kgunn, I though he was but I'm not 100% sure. I'll ask today [23:01] robert_ancell: any chance you could add me to the lightdm-team so I can push these packaging branches to be owned by that team? [23:01] thomi, what's you LP id? [23:01] thomir [23:02] thomi, done [23:02] thanks [23:10] PSA: I've pushed the use-dma-buf mesa changes and marked use-dma-buf as approved. In the period while these things are not in sync, Mir EGL will not work. [23:10] And by that I mean "Mir clients won't be able to use EGL" [23:18] robert_ancell: Merr...sorry...forgot about IRC. [23:18] ripping the InputDispatcher controller stuff (for focus) out of the InputManager to fix a focus bug to make my demo shell work :) [23:21] I made a demo shell that fullscreens everything and adds and alt tab keybinding but it crashes when you cycle focus because [23:22] the second time a window is made focused the way we recycle the android window handles in set_input_focus_to [23:22] causes the InputChannel to be destroyed and recreated...closing the fd [23:22] triggering an assert! [23:23] So it seemed like a good time to just go ahead and bite the bullet and implement the proper focus selection where the dispatcher controller maintains a list of all the window handles (this is what is needed to enable picking for touch input as well) [23:23] but then it became a refactoring spiral [23:24] haha [23:24] almost done [23:41] RAOF: Do you think you could do another review of https://code.launchpad.net/~robertcarr/mir/enable-inprocess-egl/+merge/155888 to override Alf's need's fixing (was fixed yesterday) [23:41] kevin approved so I think if you are ok with it we can land it [23:41] racarr: Sure [23:41] Would like to land it so I can do a software cursor example [23:41] (a really funny one ;)) [23:43] kdub: Does this mean anything to you https://jenkins.qa.ubuntu.com/job/mir-android-raring-i386-build/337/console [23:43] err, loaded question maybe :p but you know what I mean [23:44] racarr, no unfortunately :/ [23:45] hmm [23:45] ill try a rebuild in an hour or so and maybe it will fix itself [23:45] if not investigate deeper