duflu | alf_: Is it possible compositor/renderer destruction no longer has a GL context?... https://bugs.launchpad.net/mir/+bug/1489720 | 06:28 |
---|---|---|
ubot5 | Ubuntu bug 1489720 in Mir "[mako] mir_proving_server crashes on screen rotation" [Undecided,New] | 06:28 |
alf_ | duflu: I am not aware of any change that has affected gl context lifetimes. What does eglGetCurrentContext return? | 06:33 |
duflu | alf_: Don't know. I'm not set up on that device right now | 06:33 |
duflu | Although I will be soon | 06:34 |
duflu | Poor mako. Batter just went from full to empty in a blink of the eye | 06:39 |
duflu | I think it was always empty. Because the phone doesn't shut down properly with wily. And even if your force it, it turns itself on again and drains battery | 06:40 |
duflu | alf_: Nope. Context not lost. Everything is sane and in context. Just that glDeleteTextures crashes. Although valgrind reports that android and the graphics driver are full of errors... | 07:17 |
duflu | Might be able to bisect the issue | 07:17 |
=== alan_g is now known as alan_g|lunch | ||
alan_g|lunch | alf_: any idea what's going on here? Weirdness in the mesa stack - I've a workaround, but don't understand why it works. https://bugs.launchpad.net/mir/+bug/1495459 (I'm pretty sure camako saw this last week too, but didn't have consistent results.) | 12:58 |
ubot5 | Ubuntu bug 1495459 in Mir "Segmentation fault in MesaBufferIntegration.*" [Undecided,New] | 12:58 |
=== alan_g|lunch is now known as alan_g | ||
alf_ | alan_g: let me see if I can reproduce | 12:59 |
alan_g | greyback_: please take a look at https://code.launchpad.net/~alan-griffiths/mir/MirBlob/+merge/269949 (I think it is what you're expecting but also see the discussion.) | 13:20 |
greyback_ | alan_g: on my list, will get to it | 13:21 |
alan_g | greyback_: thanks | 13:21 |
alf_ | alan_g: removing config.add_options(x11_displays_option_name, ...) fixes the problem. No idea why. | 13:33 |
alan_g | alf_: yes, I've just discovered removing "boost::program_options::value<std::string>()->default_value(std::string{"1280x1024"})," "fixes the problem too | 13:34 |
alan_g | alf_: This is seriously weird. But likely not mesa weirdness. | 13:35 |
alf_ | alan_g: yes, and in particular it's the default_value() part that triggers the issue | 13:39 |
alan_g | alf_: thanks for confirming that | 13:40 |
desrt | src/platforms/mesa/server/kms/cursor.cpp #includes <drm/drm.h> which doesn't exist on debian. changing that to <libdrm/drm.h> fixed it for me and also seems to work on ubuntu | 13:41 |
anpok | desrt: looking at the current drm package.. libdrm seems to be the install path preferred by libdrm | 13:51 |
desrt | maybe that's why debian doesn't have it? | 13:52 |
anpok | desrt: hm linux-libc-dev provides the other | 13:59 |
desrt | anpok: not on debian... | 14:09 |
anpok | yes we should switch to libdrm | 14:19 |
alf_ | anpok: xf86drm already includes drm.h, so I don't think it's needed at all | 14:51 |
=== dandrader_ is now known as dandrader | ||
=== dandrader is now known as dandrader|afk | ||
desrt | so, having some issues with mir-on-X11. anyone can help with that? | 15:24 |
desrt | kgunn: ^ hi :) | 15:24 |
kgunn | desrt: yo...yes... camako should be helpful ^ | 15:24 |
kgunn | it's his baby | 15:24 |
desrt | that's the name i was looking for. thanks :) | 15:24 |
anpok | camako: knows most about it | 15:25 |
kgunn | he'll be happy to see someone putting it through it's paces | 15:25 |
desrt | camako: hey... having trouble bringing up drm for x11 because gbm: Last dlopen error: /usr/lib/dri/i965_dri.so: undefined symbol: _glapi_tls_Dispatch | 15:25 |
desrt | even though this does get defined (in /usr/lib/x86_64-linux-gnu/libglapi.so.0.0.0, which is loaded) | 15:25 |
desrt | we've sort of been chasing down one problem after another for the entire morning, and now we find ourselves here, and a bit stuck | 15:29 |
desrt | fwiw, it works if i LD_PRELOAD that library, but this seems like a bit of a bad solution :) | 15:30 |
desrt | k. we're gonna break for lunch. bbiab. | 15:34 |
camako | desrt, I haven't encountered this issue before. | 16:12 |
camako | desrt, what do you see when you do ... ls /dev/dri/ | 16:13 |
camako | card0 card1 controlD64 controlD65 renderD128 renderD129 | 16:13 |
camako | this is what I see ^ | 16:13 |
camako | you need the render nodes | 16:13 |
camako | bb after lunch | 16:14 |
=== dandrader|afk is now known as dandrader | ||
desrt | camako: we already hit the render node issues and got past that :) | 16:45 |
desrt | this is a pure dynamic library problem | 16:45 |
desrt | btw: this is debian jessie | 16:45 |
=== alan_g is now known as alan_g|EOD | ||
=== tvoss is now known as tvoss|test | ||
=== tvoss|test is now known as tvoss | ||
=== ahayzen_ is now known as ahayzen | ||
camako | desrt, can you run mir with the mesa-kms? I.e. in a vt? | 18:49 |
desrt | camako: yes. this is an x11 problem. | 18:51 |
=== dandrader is now known as dandrader|afk | ||
camako | desrt, we shouldn't be using gl, only egl on mir-on-x (and gles is used in the compositor) | 19:28 |
desrt | it is using egl | 19:28 |
camako | desrt, but then why the libglapi related error? Did you build the mir binaries in the debian env yourself? | 19:29 |
desrt | yes | 19:30 |
desrt | i'm not sure it's really directly related to mir, in fact, since it's rather a problem with the dri drivers that mir tries to load | 19:30 |
camako | desrt, yeah.. It's curious that mesa-kms works though. Shouldn't be that different. Can you make mesa-x11 the first platform on the 'platforms' list (like in cmake-gui) and then after building, do a 'make test'? | 19:33 |
desrt | i built with only mesa-x11 this time, fwiw | 19:33 |
desrt | i also disabled tests :) | 19:34 |
desrt | i'll switch that back on again. gimme a sec. | 19:34 |
camako | desrt, let's enable and run them | 19:34 |
camako | ok | 19:34 |
desrt | (building) | 19:35 |
camako | desrt, are you using lp:mir or something else? | 19:37 |
desrt | *ahem* something else | 19:37 |
desrt | the branch i am using is based on 365d58f5c662fb67dcfa352d9dcb07bc9b98783e | 19:39 |
camako | lol gotcha | 19:41 |
=== dandrader|afk is now known as dandrader |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!