[Spice-devel] New xrandr multi-mon / arbitrary resolution support issues

[Hans de Goede]

Hi,

On 08/27/2012 12:09 PM, Alon Levy wrote:
> On Mon, Aug 27, 2012 at 11:20:09AM +0200, Hans de Goede wrote:
>> Hi,
>>
>> So I've been playing around with this more, and except for the issues
>> when going fullscreen with multiple monitors (the one where the 2 windows end up
>> on the same monitor), things look good.
>>
>> There are some issues with the agent which need some looking into though IMHO:
>>
>> 1) Sometimes messages like this one show up in the logs:
>> Aug 27 10:52:50.318895 spice-vdagent[1064]: err: delete_mode of in use mode, setting crtc to NULL mode
>
> You cannot delete an in use mode, that's all the message is about - It
> should be a warning, not an "err:".

In that case it should not even be a warning but a debug message.

I must say even with that said, that I'm still not thrilled by the existing behavior
though. The setting of the crtc to NULL feels like a hack to me.

To me a better solution would be to:

1) Use <width>x<height> as name for the mode
2) This won't cause conflicts if the same mode is used on multiple monitors, since
find_mode_by_size will find it, avoiding it getting added twice
3) When changing to a new mode, don't delete the current mode before creating the new
one, but delete it after changing the mode to the new mode. Note this requires tracking
of modes added by the agent, so as to not delete standard modes, as well as refcounting
of the tracked modes, as one mode can be used by multiple monitors
4) Even delete the custom previously in use mode, when the new mode is a standard mode,
something which we currently don't do.

>> 2) Some weird vdagent-# modes show up in xrandr output (may be related to 1):
>> qxl-0 connected 1712x962+0+0 0mm x 0mm

>>     1712x962       60.0*+
>>     2560x1600      60.0
>>     <snip>
>>     vdagent-0       0.1
>
> The wierd thing is that there are two modes, one named 1712x962 and
> another named vdagent-0. I don't remember writing code to create the
> first.


Yeah, that is weird, and I cannot find anything in the code explaining the vadagent-0
mode showing up twice. Anyways conclusions:

1) What we have now works, but could do with some improvements
2) Since that what we have now works, I'll go and do a new upstream spice-vdagent release
soon (likely tomorrow) and use that for F-18, etc. for now.

Then later when things are a bit more quiet wrt other things needing our attention.
we can try to improve things.

Regards,

Hans