Originally posted by anda_skoa
View Post
AF_UNIX and AF_UNIX.side-channel kept on coming up in the Linux core kernel developers mind every time they looked at what kdbus and bus1 was doing. There is something different about Linux that kept this coming up as well
Linux also supports an abstract namespace which is independent of the filesystem.
Varlink slightly extended can do basically everything dbus use to do without being dependent on a single service/broker.
Remember anda_skoa AF_UNIX was established IPC before dbus existed yes this includes linux having abstract namespace. Lot of the reason why dbus is the way it is happens to be so it cross platform. Yes the idea of running gnome and kde desktops straight on windows and other insane options is why dbus is the way it is.
This is the same with X11 server basically being a full OS duplicating everything the OS kernel does these days.
Yes it would be possible to increase the stability of xdg desktop portals and the like moving over to something that is brokerless.
Yes trying to force everything though the wayland compositor or x11 server means those items are now being a broker coming a single point to overload. Yes the reason for dbus in the first place was pushing everything though the xfree86/x.org x11 server was causing it to stall out. Yes this was just move the stall out point not eliminate it.
Some cases the old Unix designs are 99% right.
Yes calling dbus cancer could be correct but this also means we need to replace it. Removing dbus and running everything though the x11 server or wayland compositor is going to move the reason why dbus plays up to the X11 server or wayland compositor. Varlink displays clearly you can do a brokerless item that does the job of dbus using existing AF_UNIX sockets and AF_UNIX socket side channel.
Comment