Re: [Qemu-devel] [PATCH 4 of 5] [UPDATE] DisplayState interface change

[Anthony Liguori]

Stefano Stabellini wrote:
> Anthony Liguori wrote:
>
>   
>
> If am not mistaken this is exactly what Paul proposed.
> As I said before, this model is certainly more elegant, but presents
> some hidden issues:
>
> what do we do with the text consoles?
> Right now a single DisplayState is shared between a graphic console and
> multiple text console.
> Do we really want to allocate a buffer each?
> Who is going to do the console multiplexing?
>   

We really should have a DisplayState that multiplexes multiple 
DisplayStates.  TextConsole should just provide a DisplayState that can 
be multiplexed.

> Even though the QEMUConsole abstraction is not very elegant and should
> probably be improved I believe it is a different and independent issue
> from the one addressed in this series.
>   

I'm not asking you to do that refactoring.  But since we all pretty much 
agree that this is how things should look, I want to make sure that the 
DisplayState API takes all of this into account.

So in that vein, I'd like to see all the things that call 
graphic_console_init changed to not take a DisplayState, but rather to 
have graphic_console_init() return an allocated DisplayState.  These 
functions should then return a DisplayState which can be passed to VNC/SDL.

From what I can tell, this is not a huge change.  You shouldn't have to 
really touch any of the console code.  I think this is important to sure 
up what the DisplayState API looks like and how the rest of QEMU 
interacts with it.

If this seems like a major refactoring, let me know and either you're 
misunderstanding what I'm suggesting or I've miscalculated how much 
change is required here :-)

Regards,

Anthony Liguori