Valve Engineer Mike Blumenkrantz Hoping To Accelerate Wayland Protocol Development

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • billyswong
    Senior Member
    • Aug 2020
    • 708

    #71
    Originally posted by oiaohm View Post

    Lets be real we already have cases of gnome application under KDE looking out of place and the reverse. Majority of users would not care if it out of place as long as it equals that the window does not appear without title bar and control buttons.

    So from usability point of view we don't need agreement.

    Lets say someone did a proxy wayland compositor as in like Xwayland except this compositor is just do SSD any application connecting to this compositor just add SSD. This would mostly fix up gnome and weston lack of SSD support. This proxy compositor would be able to crash only taking down the SSD applications.

    In some ways for me it makes sense for the wayland protocol to provide two sockets one of CSD applications and one for SSD applications so that SSD processing and CSD processing can be independent processing so increasing stability.
    I would rather say there are users who care GTK applications and KDE applications don't coexist in the same DE with consistent theming. But there isn't a solution easily accessible to the masses, so most people just live on with this low-priority quirk.

    For the SSD rendering part, it may be possible to defer them to a process separate from the main compositor, such that resource overload by SSD rendering won't pull down the compositor itself.

    Comment

    • hf_139
      Senior Member
      • May 2023
      • 335

      #72
      Wayland developers failed.

      It's good that someone else takes the reign.

      Comment

      • flower
        Senior Member
        • Aug 2018
        • 428

        #73
        Originally posted by hf_139 View Post

        CSD is a security vulnerability. An application moving crucial window elements around, potentially blocking or removing close buttons or making its own to try to get users to click on something they don't want.... we already have that in those malicious browser popups, why would we want that in desktop applications as well?
        we always had that feature as it is needed with many apps. eg things like gkrellm. X11 called it window hints and won't go away. kde on wayland (which uses ssd) also supports it.

        i prefer ssd. but the security argument is nonsense.

        Comment

        • hf_139
          Senior Member
          • May 2023
          • 335

          #74
          Originally posted by flower View Post

          we always had that feature as it is needed with many apps. eg things like gkrellm. X11 called it window hints and won't go away. kde on wayland (which uses ssd) also supports it.

          i prefer ssd. but the security argument is nonsense.
          Waylanders LOVES to argue about security. They think that global hotkeys are a security vulnerability.
          Lets hold them to their own standards. CSD actively makes it possible for desktop applications to scam their users even harder, if they want to.

          Comment

          • oiaohm
            Senior Member
            • Mar 2017
            • 8442

            #75
            Originally posted by Artim View Post
            Not how anything works. Sure, you could just have every application have its own compositor, but I'm very certain this would only complicate things. But you are writing the wrong question. The question is not what apps would go down when the compositor crashes, but why on earth did the compositor crash in the first place?
            Good question how does the crash happen.

            Originally posted by Artim View Post
            I doubt that will increase any stability, you just don't know how anything works.
            I am about to answer that question and when do you will need to take this back.

            There is a little bit of system evil on Linux.


            The out of memory manager. It picks processes to terminate as in kernel causes to crash due to resource usage this is based on oom_score you can cause a X11 windows manager to have a horrible oom_score based on just having a process create and destroy windows repeatedly. The application doing this will have it oom_score growing at a low rate than the X11 Windows manager.

            Maybe if you have a SSD application trying to pull a denial of service attack on you interface to cause you interface to repeatably be terminate maybe you will not be wanting to straight up restart that SSD functionality.

            Dealing with the Linux kernel design of resource management single source of SSD is not a great idea as this create a location that is quite simple to do a denial of service attack on and cause it to be terminated instead of the process causing the denial of service. Yes KDE feature to allow the application to live though compositor death also means an application intentionally design to break the system will also remain alive. So yes KDE design restart compostior then it goes back down because it oom_score is spinning though the roof and the kernel killer keeps on killing it.

            Yes a per application compositor doing SSD the per application compositor would be going down with the application and it would be very clear to the user what application is the problem child. CSD application pulling the same stunt would be spinning it oom score up faster than the compositor so resulting in correct process killed and also be clear to user what one is the problem child.

            This comes about from understanding how to cause the Linux kernel oom killer to take out the wrong targets.

            Comment

            • Artim
              Senior Member
              • Dec 2020
              • 1259

              #76
              Originally posted by deusexmachina View Post

              Of course, if pressed, you always refuse to explain yourself. You disagree with someone politically or scientifically on an unrelated issue and this somehow gives you indefinite carte blanche to do evil acts against this person - and even brag about wanting to do so. Everyone knows this behavior is not fair or even necessarily legal. Taking away someone's freedom and forcibly committing them ("getting 5150'd") for disagreeing with you politically in a domain where they have every right to? This is the ethic to which you adhere and expect everyone to just bow down? Clearly a cult - and people are waking up to it!
              Lol. If you have the unstoppable urge to spread lies and misinformation, that's a well defined psychic condition. The fact that you refuse to acklowledge that says much more about you than it does about me. And yes, he urgently needs to be committed so he can be treated. That has nothing to do with "disagreeing with you politically" but very simple facts, if you like it or not.

              PS: and the fact that you claim "people are waking up to it" when talking about refusing to acknowledge reality says even more about you. So you better get a shrink before you go completely insane and have to be comitted too.

              Comment

              • Artim
                Senior Member
                • Dec 2020
                • 1259

                #77
                Originally posted by hf_139 View Post
                Wayland developers failed.

                It's good that someone else takes the reign.
                If you call the most advanced solution there is a failure, sure. But that way you just call yourself an idiot that has no idea what he's writing about.

                Comment

                • oiaohm
                  Senior Member
                  • Mar 2017
                  • 8442

                  #78
                  Originally posted by billyswong View Post
                  For the SSD rendering part, it may be possible to defer them to a process separate from the main compositor, such that resource overload by SSD rendering won't pull down the compositor itself.
                  Its more what would be happening to the oom killer scores. Yes splitting the SSD rendering way from the main compositor that handling the CSD items directly means the main compositor does not need up on the OOM Killer top pick because of some SSD application doing some stupid things. Yes all the different ways you could cause the Linux oom killer to decide that a person X11 window manager need to be killed as soon as their was a resource usage spike is long and many and SSD bit is a big one.

                  Comment

                  • Artim
                    Senior Member
                    • Dec 2020
                    • 1259

                    #79
                    Originally posted by hf_139 View Post

                    Waylanders LOVES to argue about security. They think that global hotkeys are a security vulnerability.
                    Nobody sane claimed that, ever. Simply nobody has yet finished working on a well thought-through solution for all apps a user may have installed to be able to register global hotkeys without interfering with already existing hotkeys. But that's a question of usability and sanity, not (necessarily) security.

                    Lets hold them to their own standards. CSD actively makes it possible for desktop applications to scam their users even harder, if they want to.
                    It doesn't. You may try to prove that but the only example you have given contradicts your argument as it's pretty much the only case where this actually can't happen, because it's "kinda-SSD", as not the website or pop-up controls the decoration, but the browser.
                    Last edited by Artim; 26 September 2024, 10:19 AM.

                    Comment

                    • billyswong
                      Senior Member
                      • Aug 2020
                      • 708

                      #80
                      Originally posted by oiaohm View Post

                      Yes a per application compositor doing SSD the per application compositor would be going down with the application and it would be very clear to the user what application is the problem child. CSD application pulling the same stunt would be spinning it oom score up faster than the compositor so resulting in correct process killed and also be clear to user what one is the problem child.

                      This comes about from understanding how to cause the Linux kernel oom killer to take out the wrong targets.
                      This concept reminds me of how Google Chrome initiate the trend of putting webpages into their separate processes despite they are being displayed in the same browser and living in tabs within the same browser window. With this implemented, browsers nowadays won't get fully crashed by one erroneous website.

                      p.s. Will DE compositors implement such idea? It probably depends on how big such threat it is in their opinion.
                      Last edited by billyswong; 26 September 2024, 09:46 AM.

                      Comment

                      Working...
                      X