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

The MTA is fully taken care of by mmdevice.c.

vout_display_opengl_Delete will call OpenglLock, so it is a bad idea to previously release sys->glView

Close #7292

MSDN mostly fails to explain COM requirements and totally fails to
explain thread-safety requirements. This patch takes the most
pessimistic approach.

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

VLC_GENERIC was particularly bad for RISC processors.

Actual support for changing device requires (yet another) addition to
the audio_output_t callbacks set, I think.

Apparently, there is no need for COM to do that.

This reverts commit 7f807d9b.

It fixes skins2 crashing when an input is started.

Ref #7842

Since NO_NEON is not used anymore, leaving it undefined on ARMv6 will result in HAVE_NEON showing up, resulting in crashes.

ARMv7 Android builds are now built to tegra2 with a special dynamic NEON loader for libav.
Signed-off-by: Konstantin Pavlov <thresh@videolan.org>

Patch taken from Gentoo portage.
http://sources.gentoo.org/cgi-bin/viewvc.cgi/gentoo-x86/media-libs/libmad/files/libmad-0.15.1b-gcc44-mips-h-constraint-removal.patch?revision=1.1

Signed-off-by: Konstantin Pavlov <thresh@videolan.org>

Ref #7838

Remove libvlc hack, and fixes missing input state changes

- Run the volume/mute handling on the COM thread,
  removes remaining volume hacks.
- Spawn session manager from device rather than client (fixes #7809)
- Insulate session management from output API (refs #7394)

A library should not call exit(), but I do not see any easy way around
it: if libvlc_new() returns NULL on error so real errors cannot be
distinguished from --help stuff. Also --daemon already calls exit()
(through daemon()).

This introduces an O(1) memory leak when calling --help. It could be
fixed by merging libvlc_InternalCreate() and libvlc_InternalInit().

(If daemonizing succeeded, daemon() will call exit(0) internally.)