It allows faster seek and start up time when the access pace can be controlled.

The implementation design was broken. I will fix it once clock/decoder
API is more stable.

It is not perfect as the aout mixer buffers some data too but it is too
late...

No functionnal changes.

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

Have a look at the path
  HKEY_CLASSES_ROOT\Applications\vlc.exe\shell\Play\command
in the registry.

In my registry under Vista 32 EN Business the value is
  C:\Program Files\VideoLAN\VLC\vlc.exe --started-from-file "%1"

Normally, it should read
  "C:\Program Files\VideoLAN\VLC\vlc.exe" --started-from-file "%1"

Windows Explorer accepts both variants but the first one causes problems
when trying to call that command from self-written programs.

Reported by Stefan <vbtricks  gmx.net>

It has little benefits in general, (esp. in our usecase) and huge disadvantages for files on network mounts.
"Basically, IOKit should be so fast and efficient, that there are few usecases left for mmap() on Darwin."

See http://git.videolan.org/gitweb.cgi/x264.git/?a=commit;h=c89bc900a3bf0d4c4c728ad378703970b4f14e18 for more details.

Also #undef some local macros.
Still a lot of potential memleaks.

We don't need an explicit barrier when setting initialized from
vlc_mutex_init(). The caller shall synchronize memory after
vlc_mutex_init() from any thread calling vlc_mutex_lock() anyway,
since locking an unitialized non-static mutex is undefined.

(cherry picked from commit b1aa0a1a872627bb6bf6d94fea8145f4b6c3fccd)