Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

(There does not appear to be much logic there, this might be a driver
issue or confusing documentation.)

The loop parameter is always 0 nowadays.

Always call the dequeueOutputBuffer method again if it returned
changed output buffers or changed output format.
Signed-off-by: Martin Storsjö <martin@martin.st>

Previously we would only return one frame even if multiple output
frames were available, dropping all but the last one, while waiting
for a free input buffer. Instead return the first output buffer
and don't consume the input buffer at all in that case.
Signed-off-by: Martin Storsjö <martin@martin.st>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Adds the optional skeleton index for faster seeking.
Switches skeleton generation to Seekable input only case,
as we have no way to know values for fishead page before
the end of the encoding (total size, data start).

Ogg streams were invalid.
This patch ensures all streams have a way to end independently and
no other stream could start at arbitrary places. Also simplifies
a bit some code.

The ogg muxer suffers from an insufficient muxer cache value, and usually
starts streams before all streams have been declared.
The previous behaviour ( closing and restarting a new group of streams ) is kept
and fixed. But this is really not optimal.
We might need to consider increasing that cache for ogg.

It also handles correctly chained streams.

This patch adds Skeleton V4.0 metadata for multiplexed ogg streams.

Note this does not adds skeleton indexes to streams, as we need to
have a way to get sout position to reseek and rewrite that header from
keyframes.

Ref #7241

Sometimes, f_seglen ended up being garbage and resulted to a very long duration for the first segment in the m3u8 index file

Signed-off-by: Olivier Langlois olivier@trillion01.com
Acked-by: Ilkka Ollakka <ileoo@videolan.org>
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Checks support for a EGL client extension.

This is required to support non-native EGL platforms.

EGL needs to know the kind of window at run-time.

If EOF happens during buffering, the buffered frames will be flushed
and never played back.
This should only be visible for very short files such as:
http://streams.videolan.org/issues/9195/73ms-VLC-KO.wav

Note, pulse will lose the audio for such short files, alsa and afile work better

Closes: #3152

This reverts commit 72999af2.

This module seems to work fine, let developers enable this verbose
message again if they wish.

Only check samplerate and physical_channels for now