This module is based on the old native audiotrack but uses the public Java
AudioTrack API via JNI. All JNI operations are done in a separate thread in
order to avoid the overhead when attaching and detaching the current thread to
the Java VM.

News since native audiotrack:
 - Flush discards audio immediately (if not wait).
 - It can handle a mediaserver crash and restart (for Android 5.0 and after).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Remove this module to leave room for new Java AudioTrack module.

Indeed AudioTrack private API is not usable after Android 4.4, since there are
some virtual methods that are only defined in private headers. Therefore, these
methods won't have symbols and dlsym won't work.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This reverts commit 812425ba.

Also fix dependency on codec/h264_nal.lo

Sounds like a broken encoder (growing frame size from N-1 to N twice)

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>

refs A-codecs/DTS/dts/dtswavsample16.wav

refs A-codecs/DTS/dts/dtswavsample16.wav

This reverts commit 1f0c9de8.

This comes from src/misc/messages.c with new code.

IDirect3DDevice9_Present() creates an event like this:
\Sessions\1\BaseNamedObjects\DwmDxBltEvent_a7045e

But it isn't closed with IDirect3DDevice9_Release(), only explicitely
destroying the video window closes the event handle.
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>

This is what the tests expect, and was the final status before the
merge of ABORT and DEAD input events.

In fact, libvlc_Stopped does not really correspond to anything: it has
no STOPPED_S equivalent in the input thread state. The only way to
reach it is to call libvlc_media_player_stop() explicitly.
Regardless, lets keep it for backward compatibility.

This potentially allows generating the plugins cache on-the-go - there
is no more need to know the total number of plugins ahead of time.

Also fix module cache loading if int and uint32_t have different sizes.

(as opposed to reaching premature end-of-file)

To initialize logging, we typically need configuration and command line
parameters. And using modules would also be nice. But we also want to
emit log messages while initializing the configuration and the modules
bank. This is a catch-22.

To work around the problem, we store the early log messages in a
temporary list, and drain the list later once logging is configured.

(Also fix a minor integer overflow involving atoi().)