On Sat, Apr 09, 2016 at 09:08:30PM -0400, Cole Robinson wrote:
On 04/09/2016 05:23 PM, Pavel Hrdina wrote:
> On Fri, Apr 08, 2016 at 01:40:21PM -0400, Cole Robinson wrote:
>> On 04/04/2016 09:20 AM, Pavel Hrdina wrote:
>>> If in qemu.conf vncAutoUnixSocket is enabled remove all other listens
>>> parsed from XML config.
>>>
>>> Signed-off-by: Pavel Hrdina <phrdina(a)redhat.com>
>>
>> Sorry I should have read further before commenting on patch #15 :)
>>
>> This isn't in line with how we handle qemu.conf vnc_listen, which is
'only use
>> the qemu.conf setting if the user doesn't specify any explicit
<listen> config'
>>
>> I think that's the more sensible behavior so I'd rather see the code
fixed to
>> abide that instead. So, the change I suggested in patch #15, but maybe it
>> needs fixes elsewhere too
>
> I dislike this behavior too, but I'm not sure whether we can change it or no
> since it was already released and someone can count on it and libvirt has a rule
> that we don't change anything. My intention in the following patches is to
make
> the address/socket attribute of listen elements optional and if only
> <listen type='address'/> is present libvirt will use a config address
and for
> <listen type='socket'/> libvirt will auto-generate the socket path.
The
> qemu.conf will be used only to force all graphics to use only socket if that
> option is specified.
>
I think this is a perfectly acceptable thing to change behavior for. I really
doubt anyone is dependent on libvirt's behavior of throwing out other
graphical config and forcing the socket bits, and this change would unify
behavior with vnc_listen. Plus in my mind it's much more intuitive
I've checked the qemu.conf again and even the comment for that option says this:
"This will only be enabled for VNC configurations that do not have a hardcoded
'listen' or 'socket' value. This setting takes preference over
vnc_listen."
So yes, this is a bug and should be fixed. I'll include updated version of
15/18 patch in the next patch series.
Pavel