ChanServ changed the topic of #wayland to: https://wayland.freedesktop.org | Discussion about the Wayland protocol and its implementations, plus libinput
<danieldg>
hmm, that relies on the transparency being the indicator of a true prompt *and* there being no screenshots available *and* people actually noticing that the transparency is there (vs "oh, it has a distro logo no")
<soreau>
the point is, if a compositor doesn't want to do this, it doesn't have to but it's now considered buggy/broken/insecure because of one misplaced line that oversteps the scope of the protocol
<danieldg>
we can do better than that for secure UI elements
<danieldg>
present a user-chosen or random-assigned image for all secure elements, for example
<outfoxxed[m]>
And if this in particular is such a concern couldn't you just NOOP a client's fullscreen request
<rpigott>
I don't think its a strong argument against dropping it anyway. There's not really any implied requirement of the compositor, so if a compositor wants to be paranoid and has limited screencapture privilege then surely it would be permitted to draw a black bg even wihtout the xdg-shell text mandating that behavior.
<soreau>
if a compositor is concerned with security, this part is only the tip of the iceberg.. it's a very small thing that is irrelevant when considering how many other ways things can get hacked
<danieldg>
a better security proposal would include an overlay at all times that changes if in secure mode
<danieldg>
but that's annoying on its own, I'd only want it if I chose to turn it on
<danieldg>
anyway, allowing the compositor to choose wouldn't make it worse
<emersion>
danieldg: any security feature only makes sense if the app is sandboxed and unprivileged
<danieldg>
for this to be relevant you'd also have to prevent normal xdg windows from covering the whole output
<danieldg>
atm in sway, if you make an excessively sized floating window, it'll end up covering the output. Make it properly transparent and you can fake the prompt without fullscreen at all
<danieldg>
you may be able to do similar with popups
<rpigott>
toplevels can't be reliably positioned in global coords, which makes that also difficult for an imposter client
<danieldg>
basically if you want a 'no faking this security concept' it has to be written as a goal that everything is evaluated against
<danieldg>
rpigott: but popups can be
<danieldg>
they're relative to the parent window, but make it big enough and centered and you'll end up at the goal
<danieldg>
and an attacker is likely fine with a 90% success rate
<danieldg>
(that 90% success could actually be "assume maximized app")
<rpigott>
I mean I still think the text should be dropped I'm not arguing there
feaneron has quit [Quit: feaneron]
rinwa` has joined #wayland
rinwa has quit [Read error: Connection reset by peer]
<zamundaaa[m]>
And I agree, given that not all compositors implement this, I'd prefer changing the existing wording to may instead of pretending the world is different than it is
<zamundaaa[m]>
and add fullscreen requests for transparency or opaqueness + relevant caps, so that clients that rely on either behavior can explicitly ask for it
<danieldg>
just defining it as compositor policy is reasonable imo, since that's what the reality is today even with the protocol as written
<danieldg>
if you're concerned about faking the dialog (and have somehow fixed all the other ways to do it) then you can block transparency
karenw has joined #wayland
<karenw>
So this is a bit of an open ended question, but I'm currently between jobs and looking for something to keep me busy and my github profile green. What projects around the wayland/linux graphics community would appreciate my assistance? I'm familiar with vulkan/OpenGL and wayland/X protocol.
<soreau>
karenw: there are plenty of projects that wouldn't mind a helping hand, it just kinda depends on what you want to work on
<soreau>
there are drivers like mesa, clients like gstreamer and compositors.. wlroots and more
<soreau>
I'd probably try and find some bug(s) that need solving in the area of interest and see about fixing them
lsd|2 has quit [Ping timeout: 480 seconds]
karenw has quit [Ping timeout: 480 seconds]
karenw has joined #wayland
<karenw>
I ask that question and then amdgpu panics and murders my system. There's something poetic there...
<soreau>
that's one that has more recently been piped back to userland - gpu reset robustness
<soreau>
clients and compositors typically just crash into an unrecoverable state in this case, and sadly, so does the gpu :)
<karenw>
Yup, I think it apt upgrading a package that caused my machine to commit sudoku
<karenw>
But unfortunately my knowledge of the graphics stack beyond the surface level of mesa is extremely limited.
<karenw>
The general guidance over on the vulkan discord is similar. "Don't both supporting device loss, just terminate", although that's *mainly* application developers and not compositor/driver/etc. devs.
<soreau>
well the chance to do something is there, even if that means spitting a message, bt or saving something?
<karenw>
Certainly worth a look. yeah.
<zamundaaa[m]>
karenw: I assume that's mostly about games, and not about normal applications?
<soreau>
but I think it's different when you're a game as opposed to compositor
<soreau>
heh
fmuellner has quit [Ping timeout: 480 seconds]
<zamundaaa[m]>
Though for games it might be even more important to save, as they're often the cause of the reset
<soreau>
the holy grail of this is that your compositor and client handle things gracefully and you just see a flicker/blackscreen for a few and then back to work
<karenw>
zamundaaa: Yeah, games or something like CAD applications. The kind of program and end user would expect to go to an app store and install. Not infrastructure like a compositor or display server.
<zamundaaa[m]>
karenw: I wasn't talking about infrastructure, but normal apps
<zamundaaa[m]>
IIRC Qt supports device loss recovery with its Vulkan renderer
<karenw>
I think we are agreeing agressively.
<soreau>
I like to think of compositor code as kernel code - if it crashes, game over.
<zamundaaa[m]>
Certainly a good stance to have as a compositor developer
<zamundaaa[m]>
On the client side, please do assume the compositor may crash and try to recover from that
<zamundaaa[m]>
KWin development has become a lot nicer since Qt gained support for recovering from compositor crashes/restarts
<karenw>
It's pretty reliable on Windows at least to be able to test that by (re)installing your graphics driver.
<karenw>
Is there a reliable way to inject a reset on *nix?
<karenw>
Or just desktop linux I guess
<zamundaaa[m]>
With AMD, sudo cat /sys/kernel/debug/dri/1/amdgpu_gpu_recover
<zamundaaa[m]>
I think Intel had something similar
<soreau>
karenw: sway has gpu reset handling and wayfire needs it
<soreau>
FWIW
<karenw>
Well, that's certainly a material goal, thanks! I will look into wayfire.
sally has quit [Quit: sally]
<karenw>
interesting, I have /0/radeon_gpu_reset and /1/amdgpu_gpu_recover. I assume they both do the same thing, intentionally reset the respective gpu?
<soreau>
there it says `cat /sys/kernel/debug/dri/0/amdgpu_gpu_recover`
<soreau>
I'm not sure if 1 means another device or instead of soft, and hard reset or what
<soreau>
maybe ask #dri-devel
<zamundaaa[m]>
1 means another device
<zamundaaa[m]>
It's the cardN number
<karenw>
Yeah, this laptop is a cursed dual-graphics with the low-power output-connected gpu being too old for amdgpu and therefore using radeon.
<karenw>
While the high-performance gpu is amdgpu.
<soreau>
sounds like double the fun
bindu has quit [Ping timeout: 480 seconds]
feaneron has quit [Quit: feaneron]
Brainium has quit [Quit: Konversation terminated!]
Moprius has quit [Quit: bye]
nerdopolis has quit [Read error: Connection reset by peer]
nerdopolis has joined #wayland
<kchibisov>
zamundaaa[m]: do you happen to know if there's a way to trigger context lost by doing gpu_recover? for me it does nothing other than bliking a screen and showing logs in dmesg.
<zamundaaa[m]>
It's supposed to trigger a context loss
<kchibisov>
will it trigger it if you don't have robustness?
<zamundaaa[m]>
And on my desktop it does, Xwayland goes down and KWin shows a notification that the GPU reset happened
<zamundaaa[m]>
kchibisov: as Xwayland crashes, I'd say yes
<kchibisov>
for me it just never crashes me.
<kchibisov>
maybe on apu it's a bit different.
kts has joined #wayland
rinwa` has quit [Read error: Connection reset by peer]
melnary has quit [Remote host closed the connection]
<outfoxxed[m]>
<RAOF> "The docs for _that_ are not..." <- Would someone mind pointing me to the documentation here? I haven't found anything in or out of mesa apart from the header itself.
leon-anavi has quit [Remote host closed the connection]
Guest5541 has joined #wayland
rasterman has quit [Quit: Gettin' stinky!]
<Guest5541>
With two machines running wayland desktops, are there any programs to have one machine act as a second monitor for the other machine over a network? A VNC (or similar) viewer on the second computer fullscreened would work to recieve & display the picture, but is there a way to have a VNC (or similar) server on the first computer act like an extra monitor rather than sharing the existing one?
<soreau>
wayvnc?
<Arnavion>
Create a headless output and have the VNC server broadcast that. Depends on the compositor
<soreau>
weston vnc backend?
<Guest5541>
soreau: Oh I looked it up it seems like wayvnc may be able to do that. Ill try it
Ermine has quit [Remote host closed the connection]
Ermine has joined #wayland
dri-logg1r has joined #wayland
dri-logger has quit [Ping timeout: 480 seconds]
coldfeet has quit [Remote host closed the connection]
benh has quit []
benh has joined #wayland
agd5f_ has quit []
agd5f has joined #wayland
nerdopolis has joined #wayland
mattst88 has left #wayland [#wayland]
sima has quit [Ping timeout: 480 seconds]
FreeFull has joined #wayland
Guest5541 has quit [Ping timeout: 480 seconds]
lsd|2 has joined #wayland
lsd|2 has quit []
lsd|2 has joined #wayland
d42 has quit [Ping timeout: 480 seconds]
feaneron has quit [Ping timeout: 480 seconds]
nerdopolis has quit [Read error: Connection reset by peer]
nerdopolis has joined #wayland
glennk has quit [Read error: Connection reset by peer]