This reverts commit 13268071.

It is wrong as -1 and 0 does not have the same meaning for picture_Export.
(I will document it).

This reverts commit 260dd07f.

 It is useless, the variable cannot be uninitialized, and does not
generate a warning for me. The patch could prevent finding real
bug if the code is modified.

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

VLM : move input event callback deletion. In order to receive all inputs events it's better to delete callbacks just before releasing the object.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

...and fix a memory leak in LUA

This will avoid adding plenty of config_GetFoobarDir exports later.
Also make config_GetHomeDir return a heap-allocated string.

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

This patch does the following :
- as a vout window provider, make sure skins functions run by vout thread(s)
  don't issue any calls to the GUI. (skins on Linux runs fine at last !)
- prepare support for SET_VOUT_SET_FULLSCREEN new control
- implement hotkeys (no longer available at the vout level)
- modify a show() method not consistent across platform
- cleanup (dead code)

skins2: avoid deadlock when audio-visual is used (accounts for vlc often but not always hanging in this case)

Audio-visual holds the audio_mixer lock during its initialization phase which includes initialization of the vout.
On the other hand, skins polls every 10ms to check change in the audio volume (this also requires holding the audio_mixer lock)
This leads to a frequent deadlock at the skins level.

The patch stops polling w.r.t volume and rather reacts on change signalled by the core via a callback. This makes deadlock occurence almost nil.

A long term solution could be that 1/ audio_visual doesn't keep on holding the lock during vout initialisation and 2/ skins stops polling every 10ms for a whole bunch of things (rather use callbacks as qt4 does)

This problem appeared in vlc1.1.
setting volume from core or from the user slider must be differentiated, as sending back a zero value (mute) tells the core to really set value to zero (thus erasing the saved volume)

this patch reuses what is already done for position slider (similar issue)

This does not properly account check changes, so this is usual that this breaks.

This is ugly, until jb finds "a clever solution."

Fix for [1f579327] for Windows

[62bd7af8] and [182692ca]