That way, playlist_CurrentInput() can be called from the callbacks.

playlist_CurrentInput() could return the input that the playlist was
destroying. In principles, that was harmless, but silly.

Resetting the playing item only makes sense when there is no already
active input; NextItem() takes care of that from LoopRequest():

This code was redundant since the playlist was made tick-less (the
rebuild date check made no sense anymore).

b_active_videoplayback was not set up correctly when we wanted to prevent sleep.

(forward ported from 2.0 branch)

works with transform filter now too, fixes #7437 again.

This reverts commit 703ecf74.

This should let the kernel do the copy-on-write, only if required.
VLC does not memory copy the frames anymore.

no functional changes

(and revert earlier accidental changes)

The goal of the motion interface is to keep the video upside up. The
angles of the rotate and transform filters are oriented in opposite
directions, yet both filters are treated in the same way by the motion
code: this is incoherent and at least one code path is wrong, regardless
of the orientation of the motion sensors. On my HDAPS T42, the transform
path would turn the video upside down... my take is that its orientation
got flipped somewhere along the way. This fixes that.

This reduces overall CPU consumption.

This enables the normal VLC core resampling.

The VLC core resampling _should_ work now.

PulseAudio underflow is not (usually) an actual underrun. It just means
PulseAudio would fancy more data.

close #7794

Fix #7751
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>