the number of streams can change after we find the stream info

Clang doesn't like that %%rip macro for some reason, so I disabled
it for apple. That MANGLE macro is done based on ffmpeg one and
I added MANGEVARIABLE to work around osx/32-bit compile issues (and
my lack of asm-fu).

The actions of the hotkeys interface all affect the playlist, the
playlist's input, the playlist's input's vout(s), the playlist's aout
or the interfaces (controller). It has no use outside VLC.

Contrary to input_GetAout(), this works even without input thread.

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

And remove the other instance of find-input-callback.

And get rid of the horrible find-input-callback.

The aout can exist even if the input thread is stopped. This makes
audio-related functions work in case where they would previously fail.

Adjust the Android audio outputs to the latest aout core changes.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This reverts commit 346f385b.

Doesn't seem to work in all cases straight of the box

(instead of waiting for play)

If the playlist is activated, then input resources are almost certainly
going to be used. This is a bit simpler.

The stream might not exist.

Regression from e364643f.

In other words, keep the module loaded whenever the aout exists.

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

This keeps track of available output devices. Again, without core
support yet, this is not yet useful in any way.

There is no core support for this as yet, so the crash cannot happen.

This concerns ALSA, AudioQueue, JACK and KAI.

This would enable the input manager to hold an audio output plugin at
all times. Then it would be possible to keep track of output devices,
volume, policy and (JACK) routing continuously. For this to work,
outputs will need to be patched so that device selection and volume
handling works even when playing back.

Note though that this commit only introduces the change in the
audio_output_t interface; it does not implement the features above.