[Spice-devel] [PATCH v9 00/24] Add GStreamer support for video streaming

[Pavel Grunt]

Hi,

I was able to reproduce it also without the replay utility.

On Thu, 2016-01-21 at 17:20 +0100, Francois Gouget wrote:
> On Thu, 21 Jan 2016, Pavel Grunt wrote:
> [...]
> > With "gstreamer:h264" it hangs after a while and I see a few of:
> > ((null):24420): Spice-Warning **: gstreamer-
> > encoder.c:726:map_format:
> > The 8 format has not been tested yet
> 
> This message indicates that videos in the SPICE_BITMAP_FMT_32BIT
> format 
> have not been tested. I would normally expect an eror to manifest
> itself 
> in the form swapped color channels rather than a crash. But maybe I
> got 
> the alignment wrong or some such:
> 
>         // spice_format,         GStreamer format, bpp
>         {SPICE_BITMAP_FMT_32BIT, "BGRx",           32},
> 
> 
> > > With "spice:mjpeg" it works as expected
> > 
> > Unsupported "spice:h264", "spice:vp8" crashes too.

> > 
> > Maybe my usage is wrong and/or the crashes are related to "timing"
> > of
> > the replay utility.
> 
> I have never tested it with the replay functionality so I don't know
> of 
> any compatibility issue. So in theory access to the replay log would
> let 
> me reproduce the crash?

It should, the recording:
http://people.freedesktop.org/~pgrunt/win7record.spice.xz

And I use it with the replay utility:
xzcat win7record.spice.xz | ./spice-server-replay -c "spicy -h
localhost -p 5000" - -S 2 -v "gstreamer:mjpeg"

I have gstreamer1-1.6.2 packages:
gstreamer1-plugins-good-1.6.2-1.fc23.x86_64
gstreamer1-1.6.2-1.fc23.x86_64
gstreamer1-debuginfo-1.6.2-1.fc23.x86_64
gstreamer1-plugins-ugly-1.6.2-1.fc23.x86_64
gstreamer1-plugins-base-debuginfo-1.6.2-1.fc23.x86_64
gstreamer1-devel-1.6.2-1.fc23.x86_64
gstreamer1-plugins-base-devel-1.6.2-1.fc23.x86_64
gstreamer1-plugins-base-tools-1.6.2-1.fc23.x86_64
gstreamer1-libav-1.6.2-1.fc23.x86_64
gstreamer1-plugins-bad-free-1.6.2-1.fc23.x86_64
gstreamer1-plugins-base-1.6.2-1.fc23.x86_64

Regards,
Pavel

> 
> 
> > Have you seen the crash before?
> 
> No. It looks like an assert() in ff_rate_estimate_qscale() but there
> is 
> no line number. There are a couple of asserts related to the image
> type 
> so maybe that's it?
> 
>    av_assert1(s->frame_bits >= s->stuffing_bits);
>    assert(pict_type == rce->new_pict_type);
> 
>