[Spice-devel] [PATCH 0/2] RFC: handle startup race for monitors config

[Jonathon Jongsma]

On Wed, 2015-04-01 at 18:02 +0200, Marc-André Lureau wrote:
> 
> On Wed, Apr 1, 2015 at 5:41 PM, Jonathon Jongsma <jjongsma at redhat.com>
> wrote:
>         This is a potential solution to the bug described in
>         https://bugzilla.redhat.com/show_bug.cgi?id=1200750. It seems
>         to fix the
>         problem, but if anybody has a suggestion for a more elegant
>         solution, I'd be
>         happy to hear about it.
>         
> 
> 
> Is the issue related to the client reconfiguring the window monitors
> on startup after initial server MonitorConfig and sending back that
> old configuration? If so, is it possible to investigate how to filter
> the resize triggered by the application itself as opposed to resize by
> user? (I would consider the timer hack as the last possibility)


What seems to happen is the following:

(assume that the guest currently has 2 displays enabled, and the client
machine has 1 monitor)

- client connects to server, main channel / vdagent connects
- client sends monitors-config message to server to set displays to the
same resolution as fullscreen local display (e.g. display #0 = 1600x900)
- server receives monitors-config messages and forwards it to guest
vdagent to handle
- Display channel gets connected, and server sends current monitor
configuration to the client (e.g. display #0 = 1024x768, display
#1=800x600)
- client creates two windows in response to the message from the server
- the updated monitors-config message arrives from the server, and
display #0 gets resized to 1600x900 and display #1 gets disabled.
- Because display #0 gets resized, it triggers the client to send down a
new monitors config message, but due to
https://bugzilla.redhat.com/show_bug.cgi?id=868970, the second display
also gets re-enabled.

In general this is an area that's tricky to get right. We recently added
some code to try to avoid sending down monitor-config messages from the
client when they're not necessary (e.g. when they would be set to the
same dimensions that it already is). But that already created a couple
of minor regressions, so I'm a little bit reluctant to do too much more
there.

Jonathon