Drew DeVault
I demand that Linux remains exactly the same forever and all volunteer labor to maintain it goes into maintaining legacy software rather than making any new innovations. No, I am not volunteering any of MY labor for that purpose, but thanks for asking.
X11 simps paraphrased
Drew DeVault
Rich Felker
@drewdevault @sobeston Sorry if I wasn't clear. I mean the design around compositing and gpu stuff, which should logically be a computational layer independent of any ui or window system by which applications prepare their own individual display contents, not tied together with how the windowing system works.
Drew DeVault
@dalias @sobeston Wayland did not throw away the X11 model for the sake of eye candy.
Rich Felker
@sobeston @drewdevault Wayland is the result of misinterpreting "Xorg sucks" as "the X model sucks" and throwing away all that was good about it for "zomg compositing eye candy everywhere!"
Kinda like if your response to belief that "glibc sucks" was "let's adopt some new non-POSIX-compatible thing".. 🤔
Rich Felker
@sobeston @drewdevault Xorg (and the implementation it inherited) does an extremely poor job of using that in a way that exposes every client to one another in a way that precludes safe use across privilege domains, but this is a bad implementation choice not a limitation of the model. You can implement the protocol (I believe some Windows ones do) in ways where clients are fully isolated.
Rich Felker
@sobeston @drewdevault Among other reasons, because it throws away the network model. Nowadays that's not likely to be valuable in the sense of physically remote machines, but the key property is that "network" just means "cross privilege domains" (also: "cross ABIs") which is extremely valuable.
Sobeston
@dalias @drewdevault I’ll bite; why?
Rich Felker
@drewdevault No. I'm not going to go on a crusade against it, but I don't and won't use it and when it comes up I will be happy to tell people why.
Drew DeVault
"I don't like Wayland"
That's nice. Did you expect anyone who does like Wayland to stop working on it and maintain X11 for you or something?
Drew DeVault
Virtually all criticisms of Wayland boil down to "it's not X"
Drew DeVault
@dalias yes, it's descriptive enough. For example a pixel buffer accumulates damage indicating which regions have changed and you can send the difference off to the other end of the proxy. But waypipe (Wayland via SSH) takes a more efficient approach of using a video encoder instead
Drew DeVault
@dalias this conversation is what frustrates me about Wayland criticisms. People take misinformation at face value and confidently write it off based on false assumptions and a misunderstanding of the protocol's design and goals. But none of it is true, Wayland does not have the supposed limitations or flaws that people like to cite.
Rich Felker
@drewdevault Ok, that answers my question. I guess the protocol is sufficiently descriptive that the proxy is able to know what parts of the shared memory are allowed to have changed and need to be synchronized? Or does it just have to scan and diff?
Drew DeVault
@dalias if you statically link to libwayland (which makes the file descriptor assumption) you would have to write a proxy but you can definitely write a proxy.
Drew DeVault
@dalias see waypipe.
Rich Felker
@drewdevault The question is: Can you make a bridge by which an existing static linked Wayland-display-using application binary runs on a display that's not on the same logical machine and fully isolated except for some black box communication channel you've been given?
My understanding is no.
Drew DeVault
@dalias not so. The protocol itself doesn't mandate the use of fds, it just defines an out-of-band transport which in almost all cases is a file descriptor. But it does not have to be.
Rich Felker
@drewdevault There's no way to emulate them or put them in different contexts (like isolated VMs, Qubes style) that don't share the kernel. You're limited to container like stuff.
Drew DeVault
@dalias how so? the fds are just shared memory
Drew DeVault
@dalias it does pass fds, yes
Rich Felker
@drewdevault There's your broken privilege isolation.
Rich Felker
@drewdevault That's a network model if there's nothing else it needs. But doesn't it pass more fds over that or something for mapping shared memory etc.?
Drew DeVault
@dalias across the Unix socket
Rich Felker
@drewdevault How do the data/events get from one privilege domain to another?
Drew DeVault
@dalias that's... not true.
Rich Felker
@drewdevault Note that privilege isolated applications you can safely run from low trust sources aren't a thing without a network model for UI.
Drew DeVault
@dalias that sounds like a party trick
Rich Felker
@drewdevault In the same sense that ssh'ing to your box or running stuff in screen is a "party trick" 🤔
Rich Felker
@sobeston @drewdevault Going back to the network model, this is why I was able to test early build of Firefox on Adelie Linux with no invasive system changes. I had Adelie in a chrooted user namespace testing it and just dropped in the xauth and ran it with the network address for the ourside-namespace host.
Drew DeVault
@dalias the video encoder pipeline for waypipe is very clever and extremely efficient, way beyond what X11 is capable of: clients render into dmabufs and send a handle to the server, which on typical use-cases just displays it to a DRM surface on your screen (zero CPU<>GPU copies), and with waypipe it hands the dmabuf off to a hardware video encoder and copies the encoded stream from the GPU to the CPU to send over the network -- which is about as efficient as it is possible to be
Drew DeVault
@dalias so not only can you do network transparency on Wayland, but you can do it BETTER than on X11, to the point where two shitty ARM devices with GPU video encoding pipelines could share windows over a mobile network with high performance.
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.