When should a nested server first post a buffer?

[Daniel van Vugt]

I've experienced this for a long time when running mir_proving_server 
nested inside mir_demo_server_minimal. The screen doesn't go grey 
(nested server is drawn) until a client is started in the nested server.

I feel it's a reasonable side-effect of us not drawing any "shell" in 
demo servers. But there is also a reasonable argument that the 
background colour is the simplest form of a /shell/ and it should render 
immediately on start-up.


On 29/05/15 00:35, Alan Griffiths wrote:
> There's a comment in unity-system-compositor (src/window_manager.cpp)
>
>      // We need to wait for the second frame before marking the session
>      // as ready. The first frame posted from sessions is a blank frame.
>      // TODO: Solve this issue at its root and remove this workaround
>
> that led me to do some experimenting. And what I discovered is that a
> nested server posts a frame when the first surface is *created*.
>
> That is...
>
> 1. start a host session: sudo mir_demo_server --vt 2 --arw-file
> 2. Start a nested session that will be obvious: mir_demo_server
> --host-socket /tmp/mir_socket --background-color purple
> --custom-compositor adorning --window-manager tiling
>
>     At this point vt2 will have a black display because the nested
>     server has yet to post a buffer. (Although there's a titlebar
>     visible for its window - which is wrong but not the issue I want to
>     raise.)
>
>
> 3. Now create a surface but don't post a buffer: mir_demo_client_basic
>
>     At this point the purple background from the nested server appears.
>     (And yes, I checked it is the surface appearing that causes the
>     post: "tiling" doesn't paint titlebar surfaces and sticking a
>     sleep() after mir_surface_create() makes it clear this is the
>     relevant call.)
>
>
> This seems inconsistent. I could imagine it making sense to post a frame
> when starting (but the above comment suggests this isn't desirable).
>
> Since we don't post on startup we really ought to wait until we actually
> have content from a client.
>
> Any dissenting opinions?
>
> (And yes, we will need an acceptance for the correct behavior - once
> that is agreed.)
>
> PS This also suggests we have unnecessary compositing cycles whenever
> surfaces are created.
>
>