Currently we have a whole daisy-chain of startplasmawayland-> kwin ->
startplasmawaylandsession -> launching the session.
This was needed as we needed environment variables which previously we
only knew after kwin started.
After some kwin refactoring this is no longer the case, and
kwin_wayland_wrapper will sync it's environment variables to the
appropriate startup environments.
This allows us to move starting kwin_wayland to be systemd managed,
brining resource control, alongside it's classic counterpart
plasma-session.
Previously the call to ksplash was being emitted by plasma-session. This
isn't used in the systemd startup so it was introduced in a somewhat
messy script.
This script has an issue:
If ksplash is disabled the bus name won't be registered. Typically this
isn't an issue. However to avoid a race condition ksplash registers
itself as being DBus activatable, so that the dbus server holds
dispatching the message.
This combo means ksmserver our unit is considered as not running until
our script times out. This then defers other things in the chain that
depend on ksmserver.
Moving the call into the relevant piece of code into ksmserver unifies
things into a single path and gives us more control.
BUG: 432364
This means on logout we'll be gracefully closed down.
All of these are closing down properly anyway, but this makes things a
bit more explicit in case you managed to log out at an awkward point.
Summary:
This brings numerous advantages such as:
- easier admin configuration with drop-ins, overrides and multiple
hooks throughout (and only having to learn one tool)
- session cleanup on exit, avoiding that occasional part where shutdown
hangs
- startup that actually knows when things are up
- race free autostart and DBus activation at once
- logs that rotate are split by service and usable
- resource management through slices and cgroups (the part I want)
Over the past 2 years I've been trying to tidy up and encapsulate the
relevant parts of startup into the binary plasma-session so that we can
just runtime swap out that one part and supporting both paths will be
easy.
Support is toggleable via cmake flag, as it seems like it should be a
distro decision, especially as we will require a specific systemd with
the xdg-generator.
KDED/kwin/other services are attached to the relevant repo.
Task T11914
Differential Revision: https://phabricator.kde.org/D28305
squash
David Edmundson