Unfortunately, the thread stack remains allocated as our threading
abstraction does not support detaching.

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>

It avoid an assert if a program was created after the input was paused.

Also updated a bit more against old ffmpeg version.

Add a few more lockings around other playlist data as well. We really need to redo all this stuff.

macosx: volume needs to be stored in absolute numbers instead of percentages. If you don't do this, the user will get a quite different noise experience...

Closes #2805

(cherry picked from commit c01710dd)

(cherry picked from commit 9f71b31e)

This reverts commit 13b7d257.

This is incorrect as it introduces locking the release when this should not be needed.

A memleak in the handling of NSAnimation

Use UDBZ instead of UDZO as disk-image format, which should save us about almost 20% of its previous size

This format isn't 10.3 compatible, but VLC isn't either, so we shouldn't care. Pointed by zacwest on the forums.

Patch by Kaloyan Kovachev

This automagically updates 1.0-bugfix, too. No need to backport.

/me idiot.

misc/mtime.c:59:18: warning: extra tokens at end of #ifdef directive
Grah, sorry about this, Apple parser doesn't see this as a problem, but other platforms do of course.

 I choosed to extract the decoded frame back to memory instead of developping
a new specific vaapi vout. It was easier, allows transcoding and does not break
our current OSD.

 Becarefull, you probably have to applied a pending patch on ffmpeg for h264
decoding.

 You still have to manually uncomment the HAVE_AVCODEC_VAAPI define in
avcodec.h and to fix the build system to test it.