[PATCH weston v4 00/15] Add a fullscreen shell protocol

[Andrew Wedgbury]

Hi Jason,

On Tue, 11 Mar 2014, Jason Ekstrand wrote:
> >
> > Thanks for this. I've been putting together a prototype using our VNC
> server - I've implemented the required interfaces along with your new
> fullscreen shell, and I'm launching it in place of weston from your
> screen-share module. It appears to be working well.
> 
> Glad to hear it!  Not sure if it's helpful, but you might want to checkout
> libwlb:
> 
> https://github.com/jekstrand/libwlb
> 
> I'm currently in the process of getting the latest version of
> wl_fullscreen_shell implemented in it.  Unfortunately, it's not API-stable
> yet, but I think it's getting close.

Thanks, I'll take a look.

> 
> > One query - I notice you've added a "sprawl" option in compositor-wayland,
> > which creates an output for every output in the fullscreen shell
> > compositor. This makes sense in the "system compositor" use cases. But for
> > screen capture you really want the opposite - i.e. the fullscreen shell
> > compositor needs to create an output for each output of the real
> > compositor. I guess we could just connect back to the real compositor and
> > get the output configuration, but there still needs to be some way of
> > connecting this up in screen-share. (At the moment I'm simply creating a
> > single output in my fullscreen shell compositor)
> 
> Yes, this is an issue.  The wl_fullscreen_shell protocol only solves half of
> the problem.  The other half is, as you've mentioned, is setting up
> everything.  There are multiple reasons why this is a challenge.  First is
> the question of how slave compositor (RDP or VNC server) gets launched
> connected to.  One way to do this would be to develop a priviledged client
> interface whereby a client can ask the server to connect to it for
> screen-sharing and hand it the connection fd.  However, that's kind of out
> of the scope for the fullscreen shell.  For the sake of my little demo, I've
> made weston launch it.  (Note that the --width and --height parameters that
> get passed should not be needed.  I have those in there because the RDP
> backend for weston can't properly resize without crashing.  Ideally, they
> would not be there)

Yes, we're looking to support several different scenarios:
1) "user mode", where a logged-in user can start and stop screen sharing 
from their session. 
2) "service mode", where the screen is always shared, regardless of who is 
logged in (or not).
3) "virtual mode", where we launch a separate virtual session for a user.

In 2 of these scenarios, we would need to be able to start screen sharing 
when the compositor starts up without requiring any interaction (because 
the machine may be unattended in case 2, and because screen sharing would 
be the only way of interacting with the session in case 3).

> 
> The second issue is configuring outputs.  I don't know much about VNC or RDP
> but I know that they have some mechanism for multiple outputs and we would
> like to expose this.  One option would be to have the VNC server connect as
> a client to the compositor.  If we go this route, you have to be very
> careful that it doesn't dead-lock anyware because of the client-server
> loop.  There are also a lot of race conditions there between the server
> creating an output and the slave compositor creating one.  Even if we solve
> those problems, this doesn't immediatly provide a mechanism for associating
> the outputs that the VNC server sees as a client with the ones it provides
> as a server.
>
> Another possible way to solve this is to have an "add_output" request. 
> Obviously, many compositors won't allow this.  However, I am already
> planning to add a set of capabilities flags so the compositor could easily
> advertise whether or not this is supported.  If we did this, we would
> probably also want to add some sort of move_output request to allow the
> server to tell the slave compositor where the outputs are.  I talked to
> Jasper a bit about this and I'm not 100% sure what I think of it.

At the moment we tend to combine multiple outputs into one large 
framebuffer for remoting, but even for this we still need to know the size 
and position of each output.

Would it make sense for the fullscreen shell interface to have the 
"add_output" request? You could have a separate "move_output" request, or 
simply specify the size and position when adding. Then screen-share can 
create outputs to match those of the real compositor and present surfaces 
for them. I guess you'd also need a "remove_output" request.

---
Andrew Wedgbury <andrew.wedgbury at realvnc.com>