SPU track works even if it doesn't have something to render at given time.
Fixes soverlay and subtitles in sout.

(Coverity report 1187490)

Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

This is an inclusion of a patch that has been living out-of-tree for some time,
but now as Cubemap 1.0.0 has been released and the protocol has been frozen,
it would sound reasonable to get it into mainline.

Essentially it makes the VLC HTTP server support the Metacube protocol, which
is a framing format to let an external reflector know where the start and end
of the blocks are without having to understand the format itself (similar to
how the VLC HTTP server itself gets to know this information). This lets it
serve the correct header block to new clients joining in the middle of the
stream, and makes sure new clients don't start in the middle of a block,
which is a problem for non-self-synchronizing formats (such as FLV and WebM).
It includes self-synchronization and header checksumming to ensure smooth
recovery from overruns in the server-to-reflector TCP stream.

The only consumer of this protocol currently, and for the foreseeable future,
is Cubemap (http://git.sesse.net/?p=cubemap), a scalable video reflector
designed to be used with VLC.
Signed-off-by: Denis Charmet <typx@dinauz.org>

Getting the current format is unreliable, for instance if the audio
cable is plugged in while playback is running. Thus only the
first original format is trusted.

close #9957

This prevents waiting for the timeout in some situations, especially
if the format was already set to the new value before.

Indeed this can happen rarely.

close #9469

This would result in an unpleasant black or grey area for
a short time, for instance when selecting the media library.

Some OS/2 APIs are not high-memory safe. So there are needs to use wrapper
functions to make them high-memory safe.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

--stats was added at commit c24ea5fb.
So free arguments after this.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

fixes #9919

av_set_cpu_flags_mask() is not thread-safe. So are the first call to
av_get_cpu_flags() after loading libavutil, and the first call to
av_get_cpu_flags() after a call to av_set_cpu_flags_mask(). So call
those under the avcodec lock during initialization.

When converting 1920x1080p mp4 using x264, x264 complains that it failed
to allocate memory. This flags makes it happy.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Files are not added, and a first instance is blocked.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>