Re: [Qemu-devel] [RFC 5/7] qxl-render: call ppm_save on callback

[Luiz Capitulino <lcapitulino@redhat.com>]

On Wed, 22 Feb 2012 14:22:50 +0100
Alon Levy <alevy@redhat.com> wrote:

> On Wed, Feb 22, 2012 at 11:17:17AM -0200, Luiz Capitulino wrote:
> > On Tue, 21 Feb 2012 19:40:16 +0200
> > Alon Levy <alevy@redhat.com> wrote:
> > 
> > > On Tue, Feb 21, 2012 at 09:15:45AM -0700, Eric Blake wrote:
> > > > On 02/21/2012 01:19 AM, Alon Levy wrote:
> > > > 
> > > > >>>  (2) Async monitor command.  Keeps interface and works nicely.  A bunch
> > > > >>>      of QAPI bits tickled into master meanwhile, so we could look at
> > > > >>>      this again.  Luiz?  What is the status here?
> > 
> > The qapi infra is already in place for sometime now, but it doesn't support
> > async commands yet. However, we're accepting a combination of command + async
> > event on completion for commands that have to be async.
> > 
> > We'll eventually have a good interface for async support, but it's not likely
> > we'll have it for 1.1 (possible, but unlikely).
> > 
> > I think item 2 above is a good way to go, considering it will add a new command,
> > of course.
> > 
> 
> Ok, so that sounds good: I'll add an event, and later libvirt/autotest
> can use it. But in that case I'll need to introduce a connection between
> the command and the event, some id. That id will have to be generated by
> the command issuer, so a new command with event id + complete event?

That's a good question.

The only events we have today which are actually a response to an asynchronous
command are the block streaming API ones and they don't include an id.

Honestly, for this particular case, I'm not 100% sure that having an id is
_required_, as I don't expect a client to submit multiple screendump calls
in parallel and we don't "officially" support multiple QMP clients either.
Also, having the screendump filename in the event will serve as a form of
identifier too.

Btw, are you planning to add the event to the already existing screendump
command? Adding a new command that doesn't use the monitor async API and
is truly asynchronous wouldn't better?