If we didn't have enough data to fill the buffer, fill the rest
of it with zeros. This is better than playing back whatever happened
to be there from before.
Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

This (more or less) reverts commit
2b7e0168.

Stopping the aout from within the callback like this could
lead to deadlocks, where AudioOutputUnitStop in the callback thread
and AudioOutputUnitStart in the audio decoder thread blocked each
other (noticed at startup of playback on a 3GS with iOS 6.0).
Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

Apparently this is the new fad and comes in several length-flavors

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

This is the _VLC_ source code.

The buffer from decoder_NewAudioBuffer() was only used if the audio
output format was planar. Now it is allocated only in that case.

This also remove leaks when decoder_NewAudioBuffer() failed.

Fix cid #1048854

In case of a wrongly formated file, the strings where never dealocated. Same if
some strings where empty.

This has two benefits:
 - allow zero copy in audio decoders,
 - allow negotiation of the output formats (potentially for S/PDIF).

This makes the android vout work on Android 4.3.

This API is available since Android 2.3/Gingerbread (api level 9).
It's a public, official API, contrary to the earlier hacking by
accessing the internal C++ Surface class directly. In most cases,
the ANativeWindow API is a very thin wrapper doing pretty much
the same as the existing Surface class entry points we've been using.

In Android 4.3, the C++ Surface class lock function we've been using
was renamed again, but using the new corresponding function as
we've done before doesn't work any longer.

Therefore, use the public API where possible (loading it dynamically,
to not break compatibility with pre-2.3 devices).

The only non-public API used in the vout is now the YV12
pixel format, which doesn't work on all devices.
Signed-off-by: Martin Storsjö <martin@martin.st>

Fixes #8760

The ALSA channels maps only indicate the order(s) of channels supported
by the audio adapter. They do not imply that all channels are actually
physically wired.

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

The original code was non-working because the notication was never
setup correctly and resulted in a DSERR_INVALIDCALL. A proper fix
would look like:

         if( sys->p_notify )
         {
+            IDirectSoundBuffer_Stop( aout->sys->p_dsbuffer );
             DSBPOSITIONNOTIFY notif = {.dwOffset = aout->sys->i_write, .hEventNotify = sys->hnotify_evt } ;
             if( IDirectSoundNotify_SetNotificationPositions( sys->p_notify, 1, &notif ) ==  DS_OK )
             {
+                IDirectSoundBuffer_Play( aout->sys->p_dsbuffer, 0, 0, DSBPLAY_LOOPING );
                 WaitForSingleObject( sys->hnotify_evt, INFINITE );
                 IDirectSoundBuffer_Stop( aout->sys->p_dsbuffer );
             }

But fixing it makes the problem worse since the notification is setup
too late and thus the buffer will replay a full loop of the back
buffer during a drain.
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>

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

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

With the reference counting stuff, the AVFrame structure is gone or
unreachable by the time the release function is called.

This will simplify later commits.