Re: [Qemu-devel] [PATCH/RFC 0/7] Screendump to UNIX socket & in PNG format

[Stefan Hajnoczi]

On Mon, Mar 12, 2012 at 7:27 PM, Marc-André Lureau
<address@hidden> wrote:
> On Mon, Mar 12, 2012 at 7:06 PM, Stefan Hajnoczi <address@hidden> wrote:
>>
>> I think ppm for lossless and png for small file size are good options to
>> have.  Beyond that it's up to the caller to deal with the screenshot,
>> which might be scaling, detecting diffs, encoding in a fancier format,
>> etc.
>
> Daniel Berrange suggested we use gdk-pixbuf for saving into various
> formats and scaling support. If we make it optionnal, it brings a few
> more dependencies although the library itself is fairly small (140k,
> but adds gobject), and we would gain optionnal support for other
> formats (noteably jpeg) and scaling.

Yeah it wants png, jpeg, tiff.  What I'm concerned about is that QEMU
should be doing emulation/virtualization, not processing images for an
end-user.  Its main loop is not structured to do utility tasks
unrelated to running the guest.  If implemented properly QEMU isn't a
good place to host this processing, and if done badly we block the
guest and make it unresponsive.  This is a general architectural
problem, not something specific to your proposal but this is why I'm
against putting this into QEMU.

If you send a ppm over a file descriptor, what's the fundamental
limitation which requires you to add this code to QEMU?  Perhaps
simply supporting file descriptors is the right level of support from
QEMU.

Stefan