On Mon, 2 May 2016, Christophe Fergeau wrote:
[...]
> > + if (GST_MESSAGE_TYPE(msg) == GST_MESSAGE_ERROR) {
> > + GError *err;
>
> 'err' must be set to NULL here.
>
> > + gchar *debug_info;
> > + gst_message_parse_error(msg, &err, &debug_info);
Done. Here is the explanation for reference: gst_message_parse_error()
cannot return an error (it returns void) but internally it calls
gst_structure_id_get() which returns a boolean to signal
success/failure. That information is lost and it would make sense for
gst_structure_id_get() to not initialize the output parameters when it
fails. Thus we must initialize the parameters ourselves so we can check
if they got set. Yuck! That's not a very clean API.
> > +#ifdef HAVE_GSTREAMER_0_10
> > + GstAppSinkCallbacks appsink_cbs = {NULL, NULL, &new_sample, NULL, {NULL}};
> > +#else
>
> This bit can go to the last patch
[...]
I moved those two.
> > +#else
> > + gst_bus_set_sync_handler(bus, handle_pipeline_message, encoder, NULL);
>
> Using _sync_handler which runs in a different thread seems quite
> complicated. Have you tried using gst_bus_add_watch() (we have a
> mainloop in the display thread where this code runs if I'm not
> mistaken), or even gst_bus_pop_filtered() if all that you need is to
> catch encoding errors?
I tried various alternatives but they did not work. encode_frame() gets
called from the main loop and then calls gst_app_sink_pull_sample(). But
if an error occurs the buffer gets dropped and we never return from
gst_app_sink_pull_sample(), never get back to the main loop, and thus
our gst_bus_add_watch() callback will not get called. Deadlock.
Also I'm not convinced we really have a glib main loop in Xspice. If we
did I would expect APIs like g_idle_source_new() to work. See:
https://lists.freedesktop.org/archives/spice-devel/2016-March/027502.html
--
Francois Gouget <fgouget@xxxxxxxxxxxxxxx>
_______________________________________________
Spice-devel mailing list
Spice-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/spice-devel