To ensure the vdp_t instance pointers comparison is valid, the old and
new instances must be contemporary. Otherwise, the heap could put both
instances at the same memory address. So keep a reference to the old
device until after the comparison.

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Referenced chapter tracks might not contain chapter only data.

In general, the picture buffer contains, in memory order:
- cropped top scan lines,
- visible scan lines,
- cropped bottom scan lines,
- extra codec lines,
- padding lines.

This order needs to be preserved when flipping vertically or rotating
180 degrees, since filters and displays rely on this. In particular,
the picture plane structure so far has no provisions for non-zero
offsets, so filters expect visible pixels at the top left of each
pixels plane. Effectively, non-zero offsets only work for cropping
after filtering. Lets assume that the number of cropped lines are
identical at the top and bottom, and keep the X and Y offsets as are.

This fixes green lines at the bottom on R180 and VFLIP.

If support for non-centered cropping is ever required, I believe
offsets will need to be added to plane_t. This would break many video
filters though.

Only first frame in a block has timestamp, then (0) last_dts is never valid,
(1) PCR never sent, (2) track never selected by core, goto 1;

So far, the callbacks have no way to access and use the orientation, so
better fix it.

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Earlier commit for dualmono (from 2003) seemed to miss some cases
and reordering the if cases as I assume dualmono isn't most common
case in filter.

Ref #7261

Allow channel upmixing as we have channel mixer that is able to do it.

Fixes #7261

refs #4175

(Note that the pointer position is not reoriented, this comes with a
later patch in the series for all XCB video outputs.)

Modified-and-...
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

This parameter copy was working around broken format-changing video
filters, but it broke filters updating the format correctly
(deinterlace, transform and croppadd).

Refs #10745. Note that transform is still not working correctly: the
aspect ratio is broken for transpositation and quarter rotations. This
latent bug should vanish with the video orientation support later.
Otherwise the proper fix is to allow the filters chain to change the
format.

Unknown columns might occur if the prefs are written
by a newer VLC version.

close #10664

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>