Multi-monitor thoughts

[Alexandros Frantzis]

Hi all,

there are two related but slightly different aspects of multi-monitor
handling.  The first is reacting to HW configuration changes (e.g. user
plugs in/out a monitor). The second is reacting to configuration change
requests from clients.  It's a complex situation, and I am trying to
find some model that will elegantly and simply accommodate both of these
aspects. Here are my thoughts so far...

The display server works with a virtual coordinate space, which is used
for both display and input. Outputs are placed in rectangles in this
coordinate space. A display configuration is a particular placement of
the outputs in the coordinate space (of course, not placing an output is
a valid placement) with valid sizes for each output (corresponding to a
supported mode).

The display server will have a base display configuration. Each session
(aka application, Mir connection) will optionally have a custom display
configuration associated with it. Having such a configuration means that
the session has special modesetting needs, and that this configuration
will be applied when the session is focused (whatever this means in each
case), and reverted (i.e. the base configuration will be reapplied) when
we focus a normal session.

Here is how this would work in some interesting scenarios:

Scenario 1: Two XMir sessions
-----------------------------

We are in the greeter, with the base configuration applied.

XMir session 1 starts up, gets the base display configuration, sends
back a changed configuration and creates/changes its surfaces to match
the configuration. Mir applies the new config since XMir 1 has the
focus.

We switch away from XMir 1 to the greeter, therefore reverting to the
base configuration.

XMir session 2 starts up, gets the base display configuration, sends
back a changed configuration and changes its surfaces to match the
configuration. Mir applies the new config since XMir 2 has the focus.

Now, as we switch between XMir 1 and XMir 2 their respective custom
configurations are applied.

Scenario 2: Two XMir sessions, new monitor plugged/removed
----------------------------------------------------------

XMir 1 is active, a new monitor is plugged. Mir detects the change and
changes its base configuration (without applying it, since a session
with a custom config has the focus). It then notifies all sessions (or
perhaps only those who already have a special config).

XMir 1 gets the new (base) configuration, decides what it needs, sends a
new config back and changes its surfaces to match. Since XMir 1 is the
focused session, Mir applies the new configuration.

XMir 2 gets the new (base) configuration, decides what it needs, sends
new config back, changes its surfaces to match. The XMir 2 configuration
is not applied since XMir 2 is not focused.

Scenario 3: XMir requests new config (e.g. an XRandR request)
-------------------------------------------------------------

XMir sends a new config to Mir, and changes its surfaces to match. Mir
gets and applies the new configuration (since presumably XMir is
focused).

Scenario 4: Mir session/app messes up its custom display configuration
----------------------------------------------------------------------

No problem. When the session/app loses focus (or gets killed) the base
configuration will be applied.

Scenario 5: Native Mir app wants to become fullscreen in one of the
connected screens
-------------------------------------------------------------------

The app gets the base display configuration, creates or changes its
surfaces to become fullscreen.

Scenario 6: Native Mir app wants to become fullscreen in one of the
connected screens *and* wants to have a different resolution there.
-------------------------------------------------------------------

The app gets the base display configuration, changes what it needs (e.g.
changes modes and position in virtual space in one of the connected
screens) and changes surfaces to match. This is the same as the XMir
cases discussed above. Will we allow all apps to change the modes?

Note that for these scenarios to work, we need to allow clients to make
a surface fullscreen for a specific output (they can get output ids from
the configuration).

Thanks,
Alexandros