/srv/irclogs.ubuntu.com/2016/04/28/#ubuntu-mir.txt

=== chihchun_afk is now known as chihchun
robert_ancellRAOF, Can Xephyr be run in rootless mode?06:30
RAOFrobert_ancell: I don't believe so, but could probably be made to do so easily.06:30
robert_ancellRAOF, define "easily" :)06:31
RAOFrobert_ancell: Well, you'd get most of the way there by having a root window and just making it transparent? :)06:31
robert_ancellRAOF, but then the window decorations would get very complicated for the internal WM06:32
RAOFYeah, is true.06:32
RAOFSo, my guess is “not *terribly* difficult or invasive”.06:33
RAOFWell, maybe a *bit* invasive.06:33
RAOFSay, less than a month all up?06:34
robert_ancellFor who? I think it would be more than that for me given my experience. Perhaps a RAOF+duflu team would do it in a month..06:35
robert_ancellRAOF, with XMir in rootless mode, how much work does the window manager need to do? The placement should all be done by the parent WM right?06:36
robert_ancellI mean the WM that runs inside XMir06:36
RAOFYeah.06:36
RAOFThe end-state for XMir in rootless mode is that the X11-WM is basically nothing but a proxy to the external WM.06:37
RAOFNeeds to do a bit of translation for resizing and such.06:37
duflurobert_ancell: Xmir -rootless completely throws away the window manager, and implements WM itself Mir-style06:38
robert_ancellduflu, ah06:39
dufluYou could still run one, but it might conflict06:39
robert_ancellSo XMir on X would probably be the most working solution...06:39
duflurobert_ancell: On X?06:39
robert_ancellMir has an X backend right?06:40
robert_ancell(I'm kidding)06:40
duflurobert_ancell: Yes it does exist, but that's fullscreen Mir server in an X window06:40
duflurobert_ancell: Xmir -rootless implements its own XComposite-based window management. I know of no shell that handles it well yet, but also implemented a workaround for that too: Xmir -rootless -flatten  which only asks the Mir shell to implement one window per app (which most Mir shells can handle)06:42
dufluUmm, I mean -flatten uses XComposite06:43
dufluSo Xmir uses Xorg to do the compositing in software with -rootless -flatten06:44
dufluActually since widgets are windows, that's normal06:44
RAOFIt's not using the hardware-accelerated compositing path?06:53
dufluNope. Because that's actually very buggy, and not beneficial to pursue right now either06:54
RAOFThat seems unlikely? We're using the same glamor hardware-composited path on regular X.06:55
dufluRAOF: armhf is set to always software anyway. Desktop is not06:55
dufluThat's why it works06:56
RAOFThat I can see; armhf glamor is unlikely to be awesomely well tested :)06:56
dufluHardware acceleration matters much more for GLX clients. However that's only really feasible with DRI (desktop hardware)07:00
dufluAnd in all fairness, even software buffers are cached in hardware for most of their life, so performance is good07:00
RAOFRight, but GPU composition is vastly faster than CPU composition on all hardware.07:04
RAOFIf you're actually *doing* any composition...07:04
dufluRAOF: True, but "vastly" is still completely insignificant compared to even a single frame of lag. And we have several frames of lag.07:06
dufluSo that gets attention first07:06
RAOFI don't think that's the case?07:07
RAOFAs evidenced by your discovery that your monitor introduces a 20ms delay that you were entirely unaware of.07:07
dufluIt is. Proof: Software rendering completes in well under 16ms even on mobile. Our buffering lag is closer to 90ms07:07
RAOFBut that's two different types of milliseconds :)07:08
dufluRAOF: In user perception07:08
RAOFBut as long as rendering *is* maintaining 60Hz, lag is indeed a more interesting target.07:09
dufluIt is. Software rendering is very fast even on a phone (with the exception of the infamous OTA-10 regression)07:09
RAOFRight. User perception is “30Hz is worse than a consistent 90ms of lag, even though frames containing my input reach the screen faster”.07:10
dufluRAOF: Agreed, but that's not an issue. Also it would be irresponsible to pursue. Do you want Xmir working in 2015 as it was, or wait till 2017/2018?07:11
RAOFYeah, I'm agreeing with you.07:11
RAOFAs long as we're hitting frame deadlines, we don't *really* need to increase rendering performance.07:11
dufluRAOF: Even still, less important than our other current performance focal points07:13
dufluNo one cares if you've made Xmir faster if you had to stop working on improving the performance of all of Unity807:14
zzarrhello! does Mir software rendering work now?07:32
dufluzzarr: In very very limited conditions. Depends what you're trying to do07:34
dufluBut mostly not07:34
zzarrduflu, I'm not trying to do anything right now, it was just a general question07:35
dufluWe are making baby steps in other tasks that will help generic software rendering in future07:36
zzarrnice07:37
zzarrof course I still want to try to get Ubuntu with Mir running om my chromebook07:48
zzarrit's vital that I can turn both the touch pad and keyboard off when flipping it07:49
zzarr(ti's an Asus Chromebook Flip)07:50
zzarrit's*07:50
anpokso you want them to be gone.. so that unity8 discovers that there are not more desktop input device, so it would switch to the tablet ui?07:51
zzarranpok, exactly07:52
dufluDo we need to test of integer wraparounds that will happen after the Sun engulfs the Earth?07:59
dufluI guess we should assume Mir is running outside the solar system07:59
anpokzzarr: hm we should have the necessary interfaces in mir08:01
anpokzzarr: but you would need changes to unity-system-compositor to deal with that... detect the lid switch telling you the flip state?.. hm.08:02
zzarranpok, nice08:02
zzarranpok, I don't know08:02
anpoksounds like hwdb in usc .. dmidecode.. have lots of rules for various devices..08:03
anpokoh i believe we do not expose the input device registy in the mir server api... but thats the smallest problem08:04
anpoki thought about another interface to allow the compositor to filter out input devices..08:04
zzarranpok, okey, sadly I don't know anything about that, I just have a vision that I one day can run Ubuntu/Mir on my laptop08:04
zzarrwell, a script telling mir/unity that it should enter tablet mode and one that tells it to enter desktop mode would be nice08:06
zzarrare the input filtered or are they turned off? (hw)08:07
anpokhm ... if you would tell an InputDevice to stop libinput would still read from evdev08:07
anpokbut thats an implementation detail.. because we use libinputs udev handling.. and turning it off and on again would require to retrigger scanning udev devices..08:08
anpokthe reason for using that are also further implementation details in libinput itself .. because of historic reasons it does certain init steps only when you let it handle udev..08:09
zzarrI'm feeling I'm on thin ice... never mind you answered the question before I asked it08:10
anpokis it important that evdev events are not consumed?08:10
anpokbelieve me there is no ice08:10
zzarranpok, does evdev events trigger text on the screen? (keyboard inputs)08:12
anpokwell input events take a route through the system compositor down to the focused client i noone filters it along that line.. and the rendered results comes back to screen08:13
anpok*if no one filters it08:13
zzarrokey, in that case, yes it is important that it's filtered (the keyboard/mouse pad ends up on the backside when flipped)08:20
anpokoh08:20
anpokI thought the screen is rotated then closed08:20
zzarrthis image shows the form factor http://liliputing.com/wp-content/uploads/2015/03/asus-chromebook-flip_02.jpg08:26
robert_ancellduflu, RAOF, I figure it's about time to propose XMir upstream. Any opposition?08:38
duflurobert_ancell: Ha. I would not have except just landed an emergency fix from Chris Townsend today. Not sure how to test it08:38
robert_ancellduflu, I expect upstream is not going to test it in any detail anyway, so shouldn't be a problem.08:39
dufluUpstream would need Mir to complain about anything. So long as we don't break code outside our own dir08:39
robert_ancellExactly. And it seems to be in well self contained now08:40
* duflu is guessing there are lots of desktoppers in London09:06
robert_ancellPrague09:06
dufluClose09:06
dufluCloser than here09:06
robert_ancellYep09:06
duflurobert_ancell: What venue?09:06
robert_ancellHotel09:06
duflurobert_ancell: Erm, which?09:07
robert_ancellhttps://www.google.cz/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwjHiY7S_bDMAhVqAZoKHXi5DiYQFggdMAA&url=http%3A%2F%2Fprague.boscolohotels.com%2F&usg=AFQjCNGa9EmXsm9WPTU0TKXUdwLjaNbSvQ09:07
robert_ancellah, stupid google links09:07
robert_ancellhttp://prague.boscolohotels.com/09:07
alan_gattente: I'm working through the window management issues in Mir. It looks like gtk-mir might the cause of lp:1445543 - I've updated the description with a specific case that seems to show Mir passing events to the client that are ignored, can you check what happens in gtk?11:14
attentealan_g: hey, sorry. i got your message, but didn't have a chance to check yet11:19
alan_gattente: np11:20
bregmahey robert_ancell I was going to ask if there was any reason (like maybe an Intel incident) why we haven't upstreamed XMir...  I guess my question has been answered11:45
bregmaare you willing to do the work?11:46
robert_ancellbregma, yep11:46
bregmagreat11:47
anpok_bregma: looked at our xkb mapping code..11:50
anpok_the altGr problem could be caused by other problems11:51
robert_ancellhmm, why can't I VT switch out of mir_demo_server anymore?11:51
bregmaanpok_, problems within Mir, or elsewhere?11:51
anpok_we seem to use the api wrong..11:51
anpok_yeah mir client.. i finish up the nested stuff and verify that theory11:51
alan_grobert_ancell: it should be possible - I do it a lot11:51
bregmaanpok_, OK, so still Mir but on the client library?11:52
anpok_yes11:52
bregmak11:52
robert_ancellalan_g, yeah, it used to work fine. But not working for me right now... alt+ctrl+backspace seems to quit it, but not mir_demo_server_minimal11:52
alan_grobert_ancell: minimal doesn't install the quit handler11:53
robert_ancellalan_g, is there any easy way to log the keyboard events?12:01
anpok_MIR_CLIENT_INPUT_RECEIVER_REPORT=log12:01
anpok_oh in a server?12:02
robert_ancellanpok_, yeah, the server12:02
anpok_MIR_SERVER_INPUT_REPORT does some logging.. not sure if it contais key code12:02
anpok_*contains key codes12:02
robert_ancellhmm, doesn't seem to12:03
robert_ancellI'm wondering if it doesn't work because I need to hold down 'Fn' for my F keys12:03
robert_ancellIf I can get a log of keycodes that might help12:04
anpok_hm ok it does .. you should see "Received event .. type=EV_KEY code=<some number> value=0/1"12:05
=== alan_g is now known as alan_g|lunch
anpok_when you set that env var to =log12:06
robert_ancellI got nothing when I set MIR_SERVER_INPUT_REPORT=log12:06
robert_ancellBut mir_demo_server --print-input-events does something, but the output seems to overwrite so I can't read it12:06
robert_ancellhttp://paste.ubuntu.com/16094191/12:07
robert_ancellShould the keycode always be 0?12:08
robert_ancellThe only scancodes seem to be alt, ctrl and backspace. i.e. nothing for Fn or the function keys12:10
anpok_yeah.. no scan to key code translation on the server12:10
anpok_ok thats.. odd..12:11
robert_ancellI wonder if the function keys are done with ACPI?12:12
robert_ancellshowkey suggests F2 is scancode 224, which is KEY_BRIGHTNESSDOWN12:17
anpok_aeh.. hm but neither libinput nor mir filters out keys12:18
anpok_robert_ancell: what does evtest say about the input devices that you have?12:22
robert_ancellah, KEY_F2 does work, but not in combination with alt+ctrl for some reason12:22
robert_ancellok, now with more careful testing, I am getting the event to show with --print-input-events, but still no VT switch12:27
robert_ancellSeems the order in which I press 'Fn' matters12:28
robert_ancellah. I know what this is. The VT switching is done on /dev/console which for some reason stopped working in Xenial. This affected LightDM, so I switched to using /dev/tty0 which is what other things use (GDM, logind).12:31
robert_ancellI'll make a merge proposal...12:32
=== alan_g|lunch is now known as alan_g
=== dandrader is now known as dandrader|afk
=== chihchun is now known as chihchun_afk
=== dandrader|afk is now known as dandrader
=== _morphis is now known as morphis
robert_ancellanpok, https://code.launchpad.net/~robert-ancell/mir/vt-tty0/+merge/29327714:42
=== dandrader is now known as dandrader|bbl
=== alan_g is now known as alan_g|EOD
=== dandrader is now known as dandrader|afk
=== dandrader|afk is now known as dandrader

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