This reverts commit 5086a9ed.

When we moved the code to use the legacy method *or* the new method,
we lost the functionality that actually initialized the memory allocated
for the entire subpicture region (including the part where captions
weren't being rendered.

This was most obvious when regression testing on OSX, where it
resulted in old captions continuing to appear on screen even
after being cleared, or garbage in the render buffer.

Add a line that effectively zeros out the buffer prior to use.
Signed-off-by: Rafaël Carré <funman@videolan.org>

swprintf is not the standard function you are looking for

Since it never built, it was never needed on any platform

'libvpx' decoder doesn't exist anymore but we continue to disable it anyway
for people building and older version

This would need more testing and close look

This needs testing

This would deserve a second look, though

This does not support hot plug and unplug of devices.

Also, it prevented selecting a device not in the choices list due to
checks in the variables subsystem. (At least with ALSA, not all valid
devices are included in the suggestions.)

Far too many bugs in this code path recently.

Using var_Set("device") is wrong: the "device" variable holds the
currently used output device (as set by the aout plugin), not the
intended output device (as requested through the user interface).

Also fix a memory leak.