Remodel auxiliary properties data flow

[romangg]

Problem statement

Certain properties we have and need to evaluate in the backend are not meant to be passed on through the windowing system. At the moment these are:

• Scale value on X11 and KWinFT backends.
• Replication source on any backend.
• In the near future likely additional flags for scale, resolution, refresh rate that tell if the property was set manually to some specific value or an optimal one should be chosen by the system.

In practice the values of such properties are currently instead saved and loaded by KDisplay directly to so-called control files. But in any case the API of Disman becomes less intuitive and functional because of that:

• A different set of values is relevant when sending a configuration to the backend as when receiving one.
• The logic to save and load the auxiliary "control" values is currently not in Disman.

Solution ideas

Document API

Improve documentation of Disman API so it is clear some values must be side-loaded by the consumer in one direction.

Disman Control

Move auxiliary data storing from KDisplay to Disman and declare a new type Disman::Control that can be composed into a Disman::Config and side-loads the relevant additional values through one or two backends (for example via file reads/writes directly or a central control manager process via D-Bus if D-Bus is available).

Overview

[romangg]

Following concept:

• Control file handling moves from KDisplay into Disman.
• If D-Bus is available we use Disman's out-of-process backend (and don't remove it as was discussed in #2) as a single contact point for consumers and a broker between them and the hubs.
• If D-Bus is not available we fall back to using Disman in-process with file watchers to poor-man-synchronize access to the control files like we do at the moment.

[romangg]

Current progress is at:

• https://gitlab.com/romangg/disman/-/commits/auxiliary-properties
• https://gitlab.com/romangg/kdisplay/-/commits/auxiliary-properties

The overall concept was applied but the "hubs" have been less drastic of a change as they look in the diagram above. Basically the prior backends have become the Window System hub and a new class Filer_controller has been introduced that does the Control hub job.

Open tasks:

• [x] Output::enforcedModeSize can be removed. Was replaced by auto_mode.
• [x] Fix multi-output file save in Disman (multiple entries for regenerated ids, Replace output id by hash?).
• [x] Remove one of the hash functions.
• [x] Show all refresh rates in combo box again now with auto mode in place.
• [x] Increase initial size of KCM
• [x] Move control elements in KCM: resolution not important (scale, rotation, resolution, refresh rate)
• [x] Fix X11 when no output is connected
• [x] Make System settings not crash, see https://gitlab.com/kwinft/disman/-/issues/10#note_391849014.
• [x] Replica in Wayland is not fitted to the display, fixed by https://gitlab.com/kwinft/kwinft/-/merge_requests/34
• [x] Move config handling completely over to Disman
• [x] Merge Disman::Config::connectedOutputs and outputs
• [x] Replace auto rotation radio buttons with switch.
• [x] Generator::bestScaleForOutput should do fractional scaling when supported.
• [x] Remove config key-value pairs when default value instead of writing it down. With the config completely saved by Disman it's better to keep the file (that way Disman knows the configuration is already known and does not try to optimize the current one). Also this way allows simple manipulation of the file.
• [x] Make backend operational without any frontend: should be doable just by executing disman-doctor for example once. Works by calling disman-doctor -o
• [x] Test laptop lid behavior (KDisplay): Showed problems in tests. Decided to disable the functionality for now and later on either fix it in KDisplay or directly move at least the laptop-lid and laptop-docked logic into Disman.
• [x] Check KWinFT wl_output scale update for clients on change seems to work fine, event is received according to WAYLAND_DEBUG=1 output

Postponed tasks (will go into follow-up issues):

• Rename disman-doctor to dismanctl. Add an explicit watch mode.
• Fix KDisplay config control and maybe move the functionality at least partly into Disman.
• Make KDisplay KCM independent of kcmshell5 (see KDE Connect's standalone app as an example for that).
• Better names for outputs in Wayland session, see https://gitlab.com/kwinft/disman/-/issues/10#note_391988021.
• Automatic fallback to in-process mode when D-Bus not available.
• Convert all Qt smart pointers to std ones.

[romangg]

The branches are now in a good state and I'll create merge requests for them later today.

I decided to ultimately give up on any file system watching in the beginning. I.e. Disman can only work with its full functionality when D-Bus is available (or as a single process without automatic changes).

The reason is that otherwise we would have a second update channel and must synchronize the file read on change with potential later windowing system notices for a change. This might be doable with a timer so it could be revisited later.

[romangg]

mentioned in commit d5ee6c7d47db94348bafc67053c7f37add2b702f