thomiRAOF: maybe. Unity has binary deps on lots of other things though, so you often can't release libdee (for example), unless you release the rest of the stack at the same time00:03
RAOFOh, that totally makes sense.00:03
robert_ancellRAOF, do you know who I can talk to about lightdm still being in proposed? The one in main has a serious bug01:48
RAOFrobert_ancell: Probably most people in #ubuntu-release?01:49
robert_ancellRAOF, ta01:49
RAOFI'd expect a random archive admin to know.01:49
mlankhorstRAOF: oh btw can lts-raring and mesa in raring be promoted to -updates? :P01:52
duflurobert_ancell: Could you do me a favour?02:12
robert_ancellduflu, sure02:13
duflurobert_ancell: Change the "development focus" to 0.9.11 on https://launchpad.net/compiz02:13
robert_ancellduflu, done02:13
duflurobert_ancell: Thanks02:14
Sarvattmlankhorst: that was extra annoying, finding out it was in -proposed after recommending it to people instead of edgers :)02:17
Sarvattmlankhorst: but infinity was talking about making daily builds work so "any day now"02:18
olliRAOF, hey04:32
RAOFolli: Yo!04:32
ollijust wanted to say thanks for proposing all the patches over the weekend04:32
olliseems like the intel one got the most traction so far04:33
RAOFWell, the most reaction at least :)04:34
olliRAOF, true04:37
olliso, it seems like we have missed all major upstream releases (at least Mesa and X iirc)04:38
ollinot sure if "missed" is the right term04:38
* duflu wanders off to lunch04:39
ollienjoy duflu04:39
olliRAOF, how will we ship these in 13.10 then?04:39
RAOFBy patching?04:39
RAOFSame way we did XI2-multitouch?04:39
olliheh, the amount of "?" indicates I was asking a stupid question ;)04:41
olliwill you have to maintain 2 versions of the patches then, one for Mesa/X/... at 13.10 and one for the respective latest version04:41
olliwill that be a big delta, if at all?04:41
RAOFDepends on how things change underneath.04:48
RAOFThe patch as-of 13.10 though probably won't need _too_ much support that isn't simple backporting.04:49
tvossRAOF, ping05:22
RAOFtvoss: Yo.05:22
tvossRAOF, good morning :) got some time in your voip lair?05:22
RAOFMy new, improved, VoIP lair!05:23
tvossRAOF, \o/ let me grab coffee05:25
tvossRAOF, can you open a hangout and add me to it, would like to join on the ipad05:27
didrockskgunn: just to reash, I'm not thrilled to drop 50% of our testing, and just having one machine with "a workaround" to get Mir starting05:43
didrocksRAOF: hey, you confirm that the ati issue isn't due to lxc, but a real issue, right?05:44
RAOFdidrocks: Yeah, looked like it.05:45
didrocksRAOF: if you need access to the machine, we can provide you that05:45
* didrocks is seeing that we diminish more and more on Mir quality side "just" to ship in universe05:46
tvoss_didrocks, how many machines can we provision in the lab to start mir?06:49
didrockstvoss_: 2 right now06:50
tvoss_didrocks, why not more? it is my understanding that we have 2 ati, 2 intel and 2 nvidia machines06:50
didrockstvoss_: we have that since yesterday06:50
tvoss_didrocks, what gpu is running in the two machines?06:50
didrockstvoss_: we just need to provision the nvidia06:50
didrocksone intel, one ati06:50
didrockstvoss_: the 3 others are for raring06:51
tvoss_didrocks, why do we need them for raring?06:51
didrockstvoss_: because we support our users?06:51
didrocksso we do have SRUs everyday06:51
tvoss_didrocks, sure, but I thought we could just reprovision them with a different Ubuntu version?06:51
didrockstvoss_: no, we need kernel == container06:51
didrockstvoss_: the machines are the same anyway, between the ati, intel and nvidia FYI06:52
tvoss_didrocks, okay, I was told some time back that we have he/le ati, he/le nvidia, he/le intel06:52
didrocksmaybe you are not mentionning the same machines then06:52
didrocksbecause the 3 others arrived this week06:52
didrocksand we only have 3 until then06:53
tvoss_didrocks, that might well be06:53
tvoss_didrocks, so which machine is causing issues? the ati one?06:53
tvoss_didrocks, intel and nvidia come up cleanly?06:53
didrockstvoss_: well, nvidia is on raring (one of the 3 we had)06:53
didrockswe need to wire up the other one06:53
didrockson intel, it was failing as well06:53
didrocksand we needed my workaround06:53
didrocksthe sleep :/06:53
didrocksI got a "we never see it"06:54
didrocksthen, when I gave the description, starting to see devs telling they saw it06:54
didrocksand "just" reboot :/06:54
didrockswe need to change that behavior06:54
tvoss_didrocks, as far as I know we do not see those issues in the cert lab06:54
didrockstvoss_: so, we are inventing the issues?06:55
didrocks"not seen here" -> then, when opening "oh btw, I saw it some times"06:55
didrocksregular rules06:55
tvoss_didrocks, of course not, I'm just trying to track down the issue06:55
tvoss_didrocks, can I get access to the ati machine?06:55
didrockstvoss_: see the bug report, RAOF already dived to it06:56
didrockstvoss_: yeah, once daily are finished, I can give you access06:56
tvoss_didrocks, ack, can you paste the bugreport again?06:56
tvoss_didrocks, my backlog is a bit fucked up after numerous reboots06:56
didrockstvoss_: bug #120307006:56
ubot5bug 1203070 in Mir "unity-system-compositor doesn't start on some ati card (with opensource driver)" [Critical,Confirmed] https://launchpad.net/bugs/120307006:56
didrocksChris told: "it's failed to create the hw cursor buffer, and I don't believe that there are any more relevant logs06:57
tvoss_mlankhorst, ping07:00
tvoss_mlankhorst, hey there :) we are trying to solve an issue with Mir failing to create a hardware cursor buffer on ati.07:08
tvoss_mlankhorst, I wonder if you are aware of any known ati issue in that area?07:08
mlankhorsterm cursor usually has special requirements for buffer07:09
tvoss_mlankhorst, okay, the call succeeds in general, we only have an issue on one specific card. Is there any way to get something like a last error from gbm?07:30
mlankhorstuse the debugger(TM)07:31
tvoss_mlankhorst, woot, thanks for the enlightenment ;)07:32
mlankhorstit's what it's for :P07:32
duflualf__: How do you guarantee that only one compositor buffer is consumed per 16ms in the multimonitor stuff? Is it still partial releases?08:05
duflu... or yet to be figured out?08:07
alf__duflu: there is no guarantee for that atm08:09
duflualf__: OK, I thought I might be too early. When you have an idea, please let me know08:10
alf__duflu: hmm, sorry, there is a partial guarantee08:10
duflualf__: Saved resources?08:11
duflualf__: I would suggest enforcing that compositor_acquire only happens for real on one monitor (which must also have the highest refresh rate of all monitors)08:13
dufluProbably #0. And recycle till that monitor refreshes again08:13
alf__duflu: about partial release... essentially the first release of a buffer marks its end as a preferred buffer for next acquisitions. Since under normal operation this happens once per ~16ms, the consumed once per 16ms is enforced.08:15
alf__duflu: (more or less)08:15
duflualf__: I know but that won't be true under bypass, so I'm hunting for a more robust guarantee08:15
dufluUnder bypass, two compositors are held and one consumed, per 16ms08:16
alf__duflu: Couldn't we just have a differently constructed BufferStream for bypass, that doesn't enforce the MultiAcquisition policies?08:19
duflualf__: Yes. However I suspect that will cause the client to get buffers at Nmonitors*60Hz08:21
dufluWhich is fine for me with one monitor right now. Not fine for working with multimonitor08:21
RAOFDear POSIX: signals are hateful, and08:23
RAOFI hate you.08:23
duflualf__... So before it becomes a problem I would like to see some more explicit distinction between buffer recycling and guaranteed buffer advancement08:25
alf__duflu: I haven't really thought about this yet, but perhaps using a combination of a different back buffer strategy in BufferStream plus a different CompositingStrategy (note, renamed to DisplayBufferCompositor in proposed MP), which we need anyway, perhaps we could reach the needed behavior. Hopefully we won't need to have a single behavior that works for both normal and bypass, if that means making compromises for both.08:25
hikikoalf__, I tried to use the mir buffers as you said but I am not sure if you can fill the ipc package using those buffers08:25
alf__hikiko: what method are you referring to?08:28
hikikofill_ipc_package in platform alf__08:28
alf__hikiko: well, the plan is to relay this request to the native platform to do this for you08:29
RAOFWho thought that having read() able to return EINTR at any point and *partially* consume the buffer was a good idea? >:(08:32
dufluRAOF: Before threads were invented/common...08:32
dufluThough arguably, retrying I/O is less error prone than expecting programmers to do threads reliably08:33
RAOFAnd who didn't want to make libc go to the effort to *not* consume the buffer in that case?08:33
RAOFYeah, but you can't retry IO. You need to keep track of how much was actually read, and then try to read the rest.08:33
RAOFBecause *that's* totally not going to be done badly, and couldn't be sensibly implemented in the goddamned system library.08:34
hikikoI don't create a native and a nested platform at the same time alf__ either native or nested08:34
RAOFNot that this is *necessarily* the cause of output delay, just that I hate noticing things that are going to explode when a SIGIO comes in at an inopportune time.08:38
alf__hikiko: the nested platform should create and use an instance of a specially configured native platform, see second paragraph of mir-on-mir mir-devel email reply08:38
hikikoI forgot to re-add this :p08:40
hikikosorry :)08:40
deathcrawlerCurrent Ubuntu Phone images uses Mir directly?08:40
mlankhorstRAOF: sigh08:44
mlankhorstI'm getting reports about mir now..08:44
ubot5Launchpad bug 1201028 in xorg-server (Ubuntu) "Xorg crashed with SIGABRT in raise()" [Undecided,New]08:44
RAOFmlankhorst: I don't know what you're talking about officer. Mir couldn't cause *any* problems at all. See, we've corraled him in a PPA where he's kept safe from the masses!08:45
mlankhorstthe bugs get filed against xorg-server though, what should I do with them?08:46
mlankhorstI just want them not be against xorg-server for now, i don't care if I close it as invalid or reassign to mir :P08:47
dufluThat's odd. There should be no automatic reports when users (like that one) are using a PPA08:47
RAOFmlankhorst: Feel to reassign to the xmir project. duflu set that one up :)08:48
* duflu assigned it08:49
duflumlankhorst: Launchpad project "xmir"08:49
mlankhorstbut I'll reassign all then08:51
dufluOh, I was reassigning. But OK08:54
alf__deathcrawler: not yet (it still uses surfaceflinger) but we are getting close to switching09:02
=== jono is now known as Guest19501
tvoss_RAOF, disabling sigio event reporting in X is great, works fine here ;)09:20
deathcrawleralf__: Thanks09:47
dufluOh dear. My diff is now over 4kloc to trunk. This is a worry10:42
RAOFtvoss_: Oh, disabling sigio events avoids the bug?10:57
tvoss_RAOF, nope10:57
tvoss_didrocks, alf__ https://code.launchpad.net/~thomas-voss/mir/fix_process_waitpid_approach_to_correctly_setup_tracing/+merge/17636712:17
alf__tvoss_: looking12:18
alf__tvoss_: what does the waitpid (pid, NULL, 0) call do after PTRACE_ATTACH?12:27
tvoss_alf__, it waits for the child process to be stopped as a result to the parent attaching. that should help us avoid a lot of races12:28
ogra_oops, i'm not in -touch, sorry12:34
tvoss_ogra_, still great :)12:34
alf__tvoss_: is this because of  WUNTRACED   also  return  if  a  child has stopped (but not traced via ptrace(2)).  Status for12:36
alf__                   traced children which have stopped is provided even if this option is  not  speci‐12:36
alf__                   fied.12:36
tvoss_alf__, yes, partly and PTRACE_ATTACH really is what we want to do here12:37
alf__tvoss_: I guess my question is what guarantees that we have that waitpid(..., 0) is notified about the stopped child (normally waitpid() only gets termination events).12:40
tvoss_alf__, that's the usual flow of ptrace_attach and then waitpid12:41
alf__tvoss_: ok, I am not familiar with it, and the manpages are not very clear, but if it works then great :)12:42
tvoss_alf__, tests pass and I haven't seen any flakyness locally12:42
mlankhorstoh darn, I had the mir ppa enabled still..13:54
mlankhorstmight have been why my nv50 was still not working after fixing all the card bugs :>13:57
kgunnmlankhorst: are you free to help with ati debug wrt this "unable to provision hw cursor" ?...we have a QA machine on which this is consistently failing 100%, didrocks can help provide access14:10
kgunnits blocking mir entering universe14:10
kgunnrobotfuel: ^ let's see if mlankhorst can help14:11
mlankhorstkgunn: sure ;P14:15
mlankhorstwhat kind of hw btw14:15
hikikoquestion: do we have an env variable for the platform already? I'd like to see if my native platform is gbm or android at runtime without ifdef and maybe getenv is an option..14:16
kgunnmlankhorst: https://bugs.launchpad.net/mir/+bug/120307014:17
ubot5Launchpad bug 1203070 in Mir "unity-system-compositor doesn't start on some ati card (with opensource driver)" [Critical,Confirmed]14:17
kgunntvoss_: ^14:19
tvoss_kgunn, yup14:19
mlankhorstmmm lets see14:28
mlankhorstkgunn: that's only starting with unity-system-compositor from ppa:mir-team/unity-system-compositor right?14:30
kgunnmlankhorst: well, this may be a didrocks build, via his daily ppa (not necessarily exactly system-compositor-testing)14:32
mlankhorstmm lets try anyway..14:32
didrockskgunn: mlankhorst: to ensure to run the same version, use those from ppa:ubuntu-unity/daily-build-next14:32
mlankhorstok, upgrading14:34
tvoss_mzanetti, ping14:35
mzanettitvoss_: pong14:35
tvoss_mzanetti, soemthing weird going on here: https://jenkins.qa.ubuntu.com/job/mir-android-saucy-i386-build/1415/console14:35
mzanettitvoss_: hmm... seems to be the same as last time14:36
mzanettitvoss_: I wonder why this is build on ps-android-sandybridge. That seems wrong14:36
mzanettitvoss_: as that's the machine where the devices for testing are attached14:36
mzanettitvoss_: yeah. this definitely seems wrong to me. who sets up Mir jobs usually?14:40
kgunnhikiko: should be easy to add...i would think you could check if gralloc is present14:41
tvoss_mzanetti, same happens for clang: https://jenkins.qa.ubuntu.com/job/mir-clang-saucy-amd64-build/1300/console14:42
tvoss_mzanetti, not sure, thomi perhaps?14:43
tvoss_alf__, ^ can you help here?14:43
mzanettitvoss_: hmm... not sure why the one on kinnara fails. that should work.14:43
mzanettitvoss_: I contacted fginther. he's in a phone call right now. when he's free again we'll try to resolve it asap14:44
hikikoyes I've added kgunn I just wonder if we already used one because I couldn't find any14:44
alf__hikiko: why do you need to get the native platform type?14:44
hikikobecause my nested platform has a pointer to the native platform in order to call native functions when necessary14:44
hikikoand I have to initialize that pointer to a new GBM or a new Android platform14:45
hikiko(the pointer is mg::Platform)14:45
tvoss_hikiko, I thought the idea is that the nested platform does not need to know those details?14:45
hikikoit won't know14:46
hikikothen it will use the mg::Platform functions14:46
hikikobut at some point I have to initialize the mg::Platform14:47
tvoss_hikiko, isn't it like auto platform = mg::NestedPlatform()?14:47
alf__hikiko: you can do it that the same way we get the platform currently, by exposing a symbol in the native platform library (e.g. create_native_platform_nested(...))14:47
hikikoI call create_platform_nested() :s14:48
hikikobut even the current create_platform14:48
hikikoreturns a GBMPlatform in GBM case14:48
hikikoand an Android in case of android14:48
hikiko    return std::make_shared<mgg::GBMPlatform>(report, vt);14:49
hikikoyou specify that your pointer is mgg::GBMPlatform14:49
mlankhorstkgunn: hm tons of fun spam in dmesg......14:50
alf__hikiko: Isn't that what you need? The native platform knows what to do when you call create_native_platform_*(...). It creates an appropriate mg::Platform pointer and sends it back to you.14:50
mlankhorstkgunn: http://paste.debian.net/18047/14:50
mlankhorstthough no idea why....14:50
hikikoyes but in create_platform* you specify what type of platform this is14:51
hikikoyou can't have a mg::Platform pointer without initialize it with new GBM or new Android14:51
kgunnmlankhorst: do we need to contact radeon guys?14:51
mlankhorstbut I guess that's what causes your issue here14:51
hikikoyou either have to do ifdef "GBM"...14:51
mlankhorstkgunn: I think so..14:51
hikikoor get the platform at runtime14:51
hikikousing getenv for example14:51
kgunnmlankhorst: do you have a contact ? (i'm supposing you already have :)14:52
mlankhorstkgunn: that was with drm-next + drm-fixes of today :P14:52
hikikocreate_platform returns a different pointer in gbm and in android case14:52
mlankhorstI guess #radeon would work, brb buying some food14:52
alf__hikiko: that's what create_native_platform_nested() will do also, no difference in the mechanism, just in the way the native platform is set up internally14:53
mlankhorstthat was a merge of drm-next + drm-fixes, dno how representive that is :P14:53
mlankhorsthm i should probably try with a cleaner kernel later just in case14:53
hikikoalf__, I am talking about the create_platform_nested implementation14:53
hikikoI mean that14:54
hikikoeven if create_native_platform_mnested14:54
hikikolives in GBMPlatform14:54
hikikoit has to return a GBMPlatform pointer14:54
hikikoah :) sec14:54
hikikolol ok :) mg::Platform lala = create_native... yeah... got it... :P14:56
hikiko<-tired :p ok, ignore :)14:56
hikikoI wrote the func and forgot to use it 2 times today :p14:57
tvoss_greyback, ping15:05
greybacktvoss_: pong15:05
mlankhorstkgunn: ugh, was lacking pm firmware15:27
kgunnmlankhorst: pm == power management ?? (e.g. pmic )15:28
kgunnmlankhorst: oh...yeah...so hw mouse would be like one of the first hw accelerators i suppose15:28
kgunnin terms of sw trying to use it15:28
mlankhorstso I guess you should check dmesg, just in case15:29
kgunnmlankhorst: so, where does that lie in the field of responsibility ?....is that a ubuntu saucy foundational problem ?...or?15:30
mlankhorstkgunn: heck check dmesg on that machine, then you know for sure15:30
kgunnmlankhorst: mmm, but X worked ?15:30
kgunngotta run15:31
mlankhorstx can fallback to unaccelerated15:31
kgunnmlankhorst: ok...back, so "we" mir team need to add in fallback for sw cursor for xmir/radeon ?15:52
kgunnbschaefer is this something you could look pre-australian time ? ^16:00
kgunndidrocks can give bschaefer access to machine if needed16:00
bschaeferkgunn, whats going on? The sw cursor thing?16:01
kgunnbschaefer: yeah16:01
ubot5Launchpad bug 1203070 in Mir "unity-system-compositor doesn't start on some ati card (with opensource driver)" [Critical,Confirmed]16:01
kgunnbasically we know the problem....we need to create a solution & test it16:01
bschaeferkgunn, possibly, i've a few other things ive to work on :)16:03
didrockskgunn: sure, tell me once you need the access so that I can prepare16:03
kgunnbschaefer: no prob...if you're helping robotfuel keep doing that16:03
kgunnno pointing in switching just for the sake of it16:03
bschaeferkgunn, no, i do unity-maintenance under bregma and i've some new ibus changes I need to get landed16:03
kgunngreyback: curious...you said unity8 now (w/ some caveats/bugs) runs on mir....do the instructions on the wiki need to change ?16:03
kgunnor does it all just magically stay relevant ?16:04
bschaeferkgunn, if I can finish that up quickly I could possibly take a look :)16:04
greybackkgunn: which instructions? Let me check through them16:04
kgunnbschaefer: ah...ibus...yeah...kinda critical :)16:04
kgunngreyback: https://wiki.ubuntu.com/Touch/Testing/Mir16:04
bschaeferkgunn, yeah, as some would like the new version landed, but that SW fallback cursor is also critcal!16:05
greybackkgunn: hmm, so that URL to s-jenkins is an internal url only, so no community people will be able to use it.16:05
* bschaefer goes to work fast16:06
greybackkgunn: other bits could be updated, yes. I'll add it to my list16:06
kgunngreyback: doh...good point16:06
kgunnnp...in due time16:06
greybackkgunn: ack16:06
kgunngreyback: crap...can you shoot me the link to your sketchpad of "todo's" closed it before saving link16:08
greybackkgunn: sure: http://studio.sketchpad.cc/UFFAX8fxc216:09
kgunngreyback: one more interrupt, does "Use actual surfaces instead of screenshots for window management" actually address my "cut corners" blueprint ?16:10
greybackkgunn: I don't think so. That proposal is for shell to animate the actual application surface, not a screenshot of the application (as we're doing right now)16:12
greybackkgunn: nothing to do with single-surfac shell16:12
kgunngreyback: got it...texture source for the single surf...not actual compositing...16:14
kgunndidrocks: can you paste the dmesg output from the ati-trouble machine into the bug ? https://bugs.launchpad.net/mir/+bug/120307016:29
ubot5Launchpad bug 1203070 in Mir "unity-system-compositor doesn't start on some ati card (with opensource driver)" [Critical,Confirmed]16:29
tvoss_mlankhorst, ping16:31
didrockskgunn: I just pasted the relevant time log: https://bugs.launchpad.net/mir/+bug/1203070/comments/516:38
ubot5Launchpad bug 1203070 in Mir "unity-system-compositor doesn't start on some ati card (with opensource driver)" [Critical,Confirmed]16:38
didrocks(the lightdm segfault)16:38
kgunndidrocks: ??....lankhorst indicated the pm firmware not being installed as the reason for hw mouse failing....this is somewhere after that16:40
mlankhorstkgunn: pong, might be related, maybe not16:41
didrockskgunn: I don't see any logs for it, does he know what we should install? (even on the image by default)16:41
mlankhorstcan i start unity-system-compoistor alone16:41
didrocksmlankhorst: it seems lightdm is setting up an environment for it16:41
didrocksso stopping/start lightdm16:41
mlankhorstwithout lightdmm...16:42
didrockswould be a question for kgunn and robert_ancell I guess :)16:42
kgunnmlankhorst: sure.. (in xmir config i assume you mean)16:42
kgunnwhich should just kick of mir16:43
kgunntvoss_: ^ sanity16:43
kgunnbut obvioulsy...no mir clients (as xmir has failed by that point)16:43
tvoss_kgunn, ENOCONTEXT16:44
kgunntvoss_: could one, after a failure to boot in xmir, kick off the u-s-c process manually16:46
tvoss_kgunn, hmmm, no, you need lightdm and the env for that ...16:46
kgunntvoss_: is the need for lightdm because we cannot have a "clientless" mir ?16:48
* didrocks wasn't on crack then :)16:48
tvoss_kgunn, of course not, it's because the usc needs t obe passed some file descriptors16:48
tvoss_kgunn, and a vt to operate upon16:48
kgunntvoss_: ack16:49
* kgunn was insane...16:49
bschaeferkgunn, ping18:15
kgunnbschaefer: yo18:21
bschaeferkgunn, hey, so ive some time now, but I don't have an ATI card :(18:21
bschaeferi've a large desktop I can install ubuntu on but it'll take a bit of time18:21
kgunnbschaefer: yeah....it'd be best to get access to didrocks machine18:22
kgunnbut he's not on18:22
kgunnat least for the moment18:22
bschaeferdang and now hes gone...18:22
bschaeferkgunn, well I should have mentioned this acouple hours ago :)18:22
kgunnhe might show back up18:22
bschaeferyeah, Ill try poking some QA people in the meantime18:23
kgunntvoss_: robotfuel: do you know if there is someone else who might be able to provide us access besides didrocks to the ati machine failing to boot ??18:24
tvoss_kgunn, not sure, sil2100 might be able to help18:25
robotfuelI don't have access to the ati machine yet, jibel might?18:25
robotfuelkgunn: ^18:25
kgunnjibel: could you help us ?18:29
bschaeferkgunn, I think I found one18:30
kgunnbschaefer: i'm gonna break for lunch...18:31
bschaeferkgunn, cool, enjoy your lunch!18:31
tvoss_xjunior, hey there :)18:50
tvoss_xjunior, still busy debugging18:50
xjuniortvoss_: nice man :) good luck with that!!18:51
xjuniorI believe it's a hard piece of code to debug18:51
bschaeferkgunn, yup, got access to a ati machine, thanks for the suggestion :)18:51
tvoss_xjunior, it is, raof is looking into it, too18:51
xjuniorfortunately it's not Xorg :P18:51
xjuniorI still can't get into Xorg, btw. I believe it's a intel driver issue. It's crashing and I remember it was updated a few days ago18:52
tvoss_xjunior, did you pin the ppa?18:52
bschaeferxjunior, something like this: https://bugs.launchpad.net/xmir/+bug/120377618:54
ubot5Launchpad bug 1203776 in xserver-xorg-video-intel (Ubuntu) "X crashes w/ latest xserver intel driver from System Compositor Testing PPA" [Undecided,Confirmed]18:54
tvoss_xjunior, just to make sure: you have got the system-compositor-testing ppa?18:54
* bschaefer was getting an intel crash on normal X as well this morning18:54
racarrI am trying to process the scene graph discussion email19:05
xjuniortvoss_: yea yeah, absolutely19:05
xjuniorI guess that's what's happening to me :)19:05
racarrand I think what Alan is describing, and what I've started to see we need through some of the pain in resolving races like in session-transactions19:06
racarris not actually a scene graph in...any particular sense19:06
racarrone of it's responsibilities (it's interface to the compositor) coincides with some of the responsibility of a scene graph19:07
xjuniorThat's exactly the situation I have here, tvoss_. I +1'd that bug19:07
tvoss_xjunior, thx19:07
racarrbut it contains assosciations like session->surface, etc, which aren't actually 'scene' assosciations19:08
racarrI guess I feel like the problem we are trying to solve, is that we have dataa and state distributed over the system so we want to integrate it in to a single multi index store19:08
racarrand not19:09
racarrwe need a "scene graph"19:09
racarr*trails off in to mumbling*19:10
racarrI am just thinking outloud :)19:10
bschaeferracarr, so your trying to come up with a data structure that can store all of the session/surfaces while having random access while having some sort of graph structure?19:14
* bschaefer could have read that wrong19:14
* bschaefer looks up scene graph...19:16
bschaeferwhich looks just like a B-Tree19:16
bschaeferbut a bit crazier...19:18
racarrbschaefer: Hmm, thats probably part of the data structure. but I think finding a data structure once we know exactly what we need wont be so bad19:19
racarrI guess to me the thing is, state is distributed all over the system19:20
racarri.e. the input state is held in some data structure that basically builds an input view of the surface stack19:20
bschaeferright, reading scene graph wiki was kind of funny: The definition of a scene graph is fuzzy19:20
racarrthe shell bit with the session assosciation, and shell state lives in the shell, and acts as sort of another view/controller to the surface stack19:21
bschaeferisn't input in a different thread as well?19:21
racarrlikewise, the compositor gets a view of the surface stack, and treats it like a scene graph19:21
racarrso with all the state in different bits19:21
racarrsynchronization gets19:21
racarrincreasingly difficult19:21
bschaefergeez, thats going to be hard to get that all sync...19:21
racarrand, another thing is pretty much all state19:21
racarrchanging in mir19:21
racarrneeds to support some pattern like19:22
racarrnotify the shell of this and potentially allow it to interfere, etc.19:22
racarrand without one authoratative source of...what is what when19:22
racarrthis becomes kind of difficult and we've ended up with this weird pattern where the shell will do things like subclass one of our factory objects19:22
bschaeferyou'll get lost...and out of order :(19:22
racarrand interfere and then call up to the base class19:22
racarrso that's what I am thinking about :p19:24
bschaeferthat is an interesting problem to solve...19:24
racarrand we have always called this a scene graph, because at a very high level19:24
racarra bunch of components manipulate it, then the compositor visits it to render the scene19:25
racarrwell, at one very high level XD19:25
bschaeferyeah, that sounds like a fun problem that almost seems like it needs to be broken down a bit haha19:25
racarrbut I think, "scene graph" doesn't actually capture the problem we are trying to solve19:25
tvoss_racarr, what you are describing is the mvc, isn't it?19:26
bschaeferwell, i don't a single data structure can capture that...scene graph is pretty high level thats just a bunch of other graphs/trees19:26
bschaefers/dont/dont think19:27
racarrtvoss_: Well...it's some kind of MVC right! But it has a very particular structure19:27
racarrand has to provide a few seperate views19:27
racarrand I think part of what it needs to do is19:27
tvoss_racarr, right, but mvc can account for that ;)19:27
tvoss_racarr, gerry is merely pointing out that we can leverage an existing implementatoin quite easily. and I do strongly agree with that19:28
racarrsupport some system of19:28
tvoss_most likely, the glue layer I'm mentioning in the mail will allow for querying the views (compositor, input, focus) that you have in mind19:28
racarrtransactional/atomic operations19:28
racarrbut I guess I am saying that is the real problem at hand19:29
racarrand if we knew these interfaces19:30
racarrthe data structure is entirely trivial19:30
racarrmaybe Qt through OpenSceneGraph can help us on the rendering end19:31
racarrbut that's closer to a view on this structure in the compositor not the structure itself I think19:31
racarrright I mean this is our central data store, not our "rendering pass optimizer"19:32
racarrI honestly still dont understand what it would mean though, using the qt scene graph19:35
racarrdoes it mean ms::Surface implements QSceneGraphItem?19:35
racarrAre we talking about just reusing their data structure? or using the renderer too?19:38
racarrI don't think it understands that each display element has its own opengl context that its children are drawn in19:38
racarrso I don't understand how that can be integrated with the compositor exactly19:39
racarr / at all.19:40
racarrSorry :) I don't mean to be negative19:40
racarrbut I literally don't know what the first step towards doing this would be19:40
kgunnracarr: what is the first "it's" in "one of it's responsibilities (it's interface to the compositor) coincides with some of the responsibility of a scene graph"19:47
kgunni/f to compositor ?19:47
bschaeferkgunn, hey, been rebooting this machine for a while now and xmir starts up each time :(19:49
bschaeferusing : 01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Caicos PRO [Radeon HD 7450]19:49
* bschaefer keeps rebooting19:49
kgunnbschaefer: hmmm....didrock's was a Cedar series...which i think is Radeon HD 5000 series of gpu19:51
bschaeferkgunn, right, i should go check my desktop really quick...as if its a cedar series it might be worth getting xmir installed on it19:52
kgunnbschaefer: also..19:52
kgunnjust thinking19:52
kgunnit wasn't so much the fact it was cedar series....19:52
kgunnthe current theory of suscpicion was that19:52
bschaeferthe power management?19:53
kgunndidrock's machine didn't have pm firmware installed19:53
bschaeferyeah, hmm I wonder if I can uninstall that...19:53
kgunnso when the gfx driver tried to use hw cursor it couldn't19:53
kgunn...and then got pissed19:53
bschaeferbut you would think things would fail softly...19:53
bschaeferinstead of bringing everything down19:53
kgunn (that's U.S. pissed not U.K. pissed)19:53
* bschaefer is from U.S19:53
* bschaefer just learned there was a difference19:53
bschaeferkgunn, would you happen to know how to remove pm firmware?19:54
kgunnoh...thot you were in canada...and you never know which they favor with speech :) z==zee vs zed19:54
kgunnbschaefer: no ide19:54
bschaeferkgunn, haha yeah, from Seattle :)19:54
bschaeferkgunn, alright, well ill look into if thats even possible to remove, and if removing it should yield a working desktop or not19:55
kgunnbschaefer: kernel folk might know....19:56
bschaeferkgunn, would you happen to know a channel for that? When/if i get stuck at lease...19:56
racarrkgunn: This data structure that we are debating in the email thread19:57
racarrif I understood your question19:57
racarrthat point was just supposed to say that while it mimicks a scene graph in one scenario19:58
kgunnbschaefer: took a guess...lots of people in #ubuntu-kernel19:58
racarrI dont think that's the right way to come at it19:58
bschaeferkgunn, great guess :), thanks!19:58
kgunnracarr: ack19:58
kgunnbschaefer: high five!19:58
racarrI mean in the most literal sense19:58
* bschaefer high fives back19:58
racarrit could be called a graph of the scene19:59
racarrbut it could also be called a list haha19:59
racarrso I am trying to find the right direction to zoom in on it from20:00
kgunnbschaefer: hmm, could it disabling the "APM" option referred to here be the trick http://how-to.wikia.com/wiki/How_to_configure_the_Linux_kernel/Power_management_options_(ACPI,_APM)20:01
* bschaefer takes a look20:02
bschaeferpossibly, I was thinking I could move the firmware/driver out of its dir in the kernel somewhere20:02
kgunnbschaefer: maybe...something easier first...20:03
kgunnwhy not change the xorg conf to just rely on sw cursor20:03
bschaeferhmm so right now its relying on the hw cursor?20:03
* bschaefer looks at the xorg conf20:04
kgunnbschaefer: so i guess any xorg driver actually reads the xorg conf file....20:10
* kgunn realizes how little he knows20:11
bschaeferkgunn, as do I :), im looking at turning some things off in the BIOS at first possibly that'll cause the firmware to not load?20:12
bschaeferkgunn, also the machine im on doesn't even have an xorg conf file20:12
bschaeferbut I did fine a bunch of binary in the kernel that looked like it was dealing with power...but I have a feeling moving those will cause other problem s:)20:13
bschaeferkgunn, how did we come to the conclusion the we were missing pm firmware?20:13
kgunnmlankhorst saw dmesg's go by saying so20:14
kgunnbschaefer, but i think you should be able to simply add a xorg conf file with over-riding values for options20:14
bschaeferright, i need to look at those man pages :)20:15
bschaeferlots of things to read :)20:15
kgunni only think this because of sna wiki (https://wiki.ubuntu.com/X/Testing/IntelSNA) where it talks about downloading their pre-formed version for sna20:15
bschaeferyeah you can add an xorg conf file... im just wondering why there wasnt one...20:19
bschaeferim also trying to wrap my head around why not having a pm firmware loaded would cause the hw cursor to not work20:20
kgunnon the pm part...hw cursor is actually a seperate piece of silicon, that requires to be "powered" up at some point, so something probably inside the hw mouse driver20:24
kgunnqueried pm, it failed, so he failed..."sorry i ain't got no power"20:24
bschaeferthat does make more sense, im waiting on this machine to do something20:25
bschaeferkgunn, sorry, seem to have broken to machine, hopefully I can get it up soon :)20:38
kgunnbschaefer: thanks for the effort20:38
bschaeferthe machine*20:38
kgunnand the update20:38
bschaefernp, I guess you can't just disable all PM  and expect it to turn on :)20:38
bschaeferalso forcing a SW wont help much atm, as im trying to reproduce the problem atm :)20:44
bschaeferalright, have the machine back...21:11
xjuniortvoss_: hey man, did have success21:12
tvoss_xjunior, how?21:12
xjuniorany news on debugging the slowness on XMir21:13
tvoss_xjunior, ah, that was a question?21:14
xjunioryeah :) sorry21:14
xjuniorI ate some words apparently21:15
tvoss_xjunior, no worries :) know the symptoms21:15
tvoss_so I have an instrumented build of X now, will give that a spin tomorrow. Its 11pm here :)21:15
xjunioroh, gotcha :) Europe?21:18
tvoss_xjunior, yup, Germany21:18
xjuniornice :) have friends there21:20
kgunnrobert_ancell: so looks like the ati bug is the last hurdle (at least for universe)21:20
xjuniorAlright, good luck tomorrow. Hope you solve it.21:20
kgunnrobert_ancell: so some debug mlankhorst did that didrocks machine didn't have pm firmware installed...21:20
mterryrobert_ancell, you asked about u8-greeter and running as mir server?21:21
kgunnso when hw cursor tried to init...it failed....and xmir didn't fall back to sw cursor21:21
mterryrobert_ancell, I was thinking we wouldn't need to run as mir server anymore21:21
kgunnwhereas standalone x does (or that is the assumption)21:21
kgunnrobert_ancell: so bschaefer is in the midst of trying to "simulate" that problem...since didrocks is sleeping :) and we can't get to the machine21:22
bschaeferyeah, and so far xmir is working, sooo trying to remove pm firmware...but just got the broken machine back...21:23
robert_ancellmterry, I'm just getting LightDM to support Mir sessions/greeter with VT switching since we don't have nested Mir support yet. Just wondering if I can run the u8 greeter for now like that or support has been dropped21:23
kgunnrobert_ancell: so i'm hoping this is either a bug or missing feature we have to fall back to sw cursor21:23
robert_ancellkgunn, who let didrocks go to sleep? ;)21:24
kgunni know... right ?21:24
robert_ancellkgunn, yeah, wonder if it's just never been noticed because everything does the fallback21:24
robert_ancellseems like it should be fixed in the driver depending on how widespread it is21:24
kgunnrobert_ancell: yeah...its one of those bugs that might have been lurking...also, just in case...didrock card was  a Cedar (5000 series)21:25
kgunnjust in case....as in bschaefer is toying around on a 7000 series21:26
bschaeferit would be strange that a specific ATI card was doing this vs PM ... but what do i know haha21:26
kgunnrobert_ancell: if you, duflu or RAOF have a 5000 series it might be worth it just to enable swcursor in xorg.conf and test it21:27
bschaeferthey are under different classes though...21:27
robert_ancellkgunn, I just have the one intel/nvidia card21:27
kgunnit would at least help in confirm/deny21:27
mterryrobert_ancell, my branch is still using an internal mir-server21:27
bschaeferim on a Caicos PRO21:27
kgunnthat this whole issue is related to swcursor21:27
mterryrobert_ancell, i.e. I haven't made it a pure client yet21:27
robert_ancellmterry, cool, can you keep it like that for testing purposes until we have the whole stack ready?21:28
robert_ancellor support both?21:28
kgunnmterry: now has 3 optional configs he needs to support :))21:28
mterryrobert_ancell, uh, I can keep my current split as is, yeah.  Let me know when you don't care about mir-server mode anymore21:30
robert_ancellmterry, will do :)21:30
kgunnrobert_ancell: also...would you mind dispositioning the MIR feedback ? i think your best to speak to it21:37
kgunnrobert_ancell: one note...for all the "need to assign a bug owner"21:37
* robert_ancell looks up dispositioning21:38
kgunnolli chose mir-team for the moment :)21:38
robert_ancellkgunn, that works for now21:38
kgunnyeah...he promised to seek another victim...oops i mean owner21:38
robert_ancellwiktionary says "Present participle of disposition.". Now I have to remember what a present participle is :)21:39
kdubEventSink hitch-hikes through a lot of constructors21:39
kgunnrobert_ancell: whenever RAOF or duflu come on, just check if they've got a cedar/5000 series ati card to try (w sw cursor).....since that bugs our #1 right now21:40
robert_ancellkgunn, ok21:41
robert_ancellmterry, oh, how did you remove the u-s-c task from the mir MIR?21:47
mterryrobert_ancell, there's a little red minus symbol next to the tasks21:55
robert_ancellmterry, huh, I totally missed that21:55
mterryrobert_ancell, way nicer than marking invalid in terms of bug spam21:57
robert_ancellmterry, in the MIRs you refer to needing a dev team subscribed to bug reports. Is that the "bug supervisor" in https://bugs.launchpad.net/mir?22:08
mterryrobert_ancell, yeah, but for the ubuntu package side, not the upstream one22:14
mterryso ubuntu/+source/mir22:15
robert_ancellmterry, k22:15
mterryI guess that's mir-team22:15
robert_ancellmterry, so, Unity doesn't seem to have a subscriber https://bugs.launchpad.net/ubuntu/+source/unity22:16
mterryrobert_ancell, that would be good to add too22:22
mterryrobert_ancell, we didn't use to *require* it, only strongly suggest it22:23
mterryrobert_ancell, but no one did it, so we are bumping up our language22:23
robert_ancellmterry, yeah, I'm trying to find *any* project that has a team subscribed to bug email22:23
racarrkdub: It's not strong coupling, all the classes are just really good friends and having a party.22:24
mterryrobert_ancell, most of the gnome desktop ones have a team subscribed22:25
robert_ancellthe dialog for this is really bad. The list of teams is not sorted alphabetically22:25
robert_ancelland you can't seem to tell who is subscribed by looking at the project page22:28
kdubracarr, haha22:42
robert_ancellRAOF, did you see the traceback about cedar/5000 series ATI cards22:47
RAOFrobert_ancell: I did, yes.22:47
robert_ancellRAOF, got any?22:47
RAOFI do not.22:47
RAOFI've got a 4535 or somesuch, and a 7870 that I don't have a desktop box to put into.22:48
bschaeferRAOF, i've been trying to reproduce it as well, on a 745022:48
bschaeferbut xmir is working perfectly22:49
bschaeferRAOF, it was mentioned that this could be due to the power management firmware missing, but it doesn't seem to be easy to just remove that to test if thats the problem...22:50
RAOFbschaefer: I don't know about that - on 3.11 kernels you need more firmware than we've got in linux-firmware to *load* radeon.ko, but (a) we're running a 3.10 kernel, and (b) Mir would fail in a different way if that were the case.22:52
bschaeferyeah... hmm I wonder if its the radeon family (as the 5000 is in the evergreen series) drivers then...22:54
bschaeferRAOF, shouldn't mir attempt to play nice when it can't load the HW cursor?22:54
RAOFMaybe? It currently doesn't, though.22:54
bschaeferwell you would assume if this happens: std::exception::what: failed to create gbm buffer22:55
bschaeferno other gbm buffers are going to work...22:55
RAOFYou know what? I should look in the xf86-video-ati source and see whether they have any specific hacks for that card & the hw cursor.22:56
RAOFUnfortunately, not. The hw cursor bo is super-specific - failing to allocate it doesn't mean that you'll fail to allocate anything else.22:56
bschaefero hmm thats interesting22:57
bschaeferwell im sure you've seen the list family for the ATI cards but: http://xorg.freedesktop.org/wiki/RadeonFeature/22:57
bschaeferas the card im using is under Nothern Islands...idk if that even matters though vs the driver is self for each card22:58
RAOFIsn't a 7450 SI? Stupid damned marketing department.23:14
bschaeferit should be, looking at the number but the machine im on gives me:23:16
bschaefer01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Caicos PRO [Radeon HD 7450]23:16
bschaeferwhich Caicos is a NI card...possibly the  PRO is throwing it off...23:16
bschaeferbut looking here as well: http://wiki.gentoo.org/wiki/Radeon#Feature_support23:17
bschaeferand it  has the 7450 under NI23:17
RAOFYeah, it the radeon page is probably right.23:18
bschaeferhmm my desktop that i've not used in a while had an evergreen card23:18
RAOFI just hate that the card marketing generation is not the same as the card chip generation.23:18
bschaefera redwood one23:18
* bschaefer starts downloading13.1023:19
RAOFI've got an evergreen (although I can't quite remember the specific model) too, but that worked fine last time I tried it.23:19
* RAOF will be slow for the next hour or so, as he's minding Zoë.23:19
bschaeferwell it'll take  a bit for me to get my desktop up with xmir.. 1-2 hours...23:20
bschaeferit'll be strange if its just specific to CEDAR23:20
bschaeferbut then again this could be normal, as I haven't done video card stuff much before :)23:21
bschaeferhopefully duflu will come on a reproduce the problem23:22
RAOFHey, is there any way to get a buffer ID that crosses the IPC boundary? I want to check that the buffer I most recently submitted from XMir is *really* the buffer that Mir's displaying.23:44
olli_bregma, can you pls ping the relevant armhf issues to slangasek?23:48
olli_or robert_ancell ^23:48
kgunnRAOF: ok....any guesses on why robotfuel might see openarena values higher on Xmir vs X ?23:52
kgunnkgunn: the results are still strange with the monitor attached x is 44.47  frames per second average and xmir is 136.9723:52
kgunnis it because there's no governing ?23:53
kgunne.g. no more vsync on the bottom of x (moved to the bottom of mir)23:53
kgunnrobotfuel might take a trained eye....but do you see any tearing ?23:53
RAOFvsync would be my immediate guess23:54
robert_ancellkdub, around?23:54
RAOFXMir clients are *not* throttled to vsync23:54
robert_ancellkdub, I've just forwarded you an email regarding armhf tests and android. Perhaps you can help understand what lp:mir expects and what is happening23:55
kgunnrobotfuel: does it also provide time to complete ?....and is that different on x vs xmir ? (ignoring the fps for moment)23:56
kgunnRAOF: right...so clients can go nuts and swap buffers as they like...which means, the truly displayed frames is much less than that23:56
kgunnor "allegedly" is much less than that :)23:57

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