weston: weston randr protocol for testing and configuration

[Bill Spitzak]

There certainly should be no need to worry about multiple threads from 
the same client. I think you can work on the assumption that the 
programmer who wrote the client is not insane! This is the way commit 
for wl_surface works, right?

The per-client pending mode is probably cleanest, but I suspect a single 
pending mode shared by all clients, where a commit from any one causes 
it to be used in whatever state it is in, will work acceptably. It just 
seems like two privileged clients doing this at the same time and not 
being aware of each other is unlikely. Also results in a better final 
state if one of the clients only changes one of the outputs, in that it 
won't "undo" the changes made by the other client to the other output. 
Fixing this per-client would require you to store modified flags, not 
just the final state.