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

Ignore wholly translucent rows at the top and bottom of the subtitle page

Only create an overlay large enough for the remaining rows and only only ask
zvbi to render those rows.

One happy side effect is that now --vbi-text no longer outputs header text
for some teletext streams (although --vbi-text is still unusable imho).

Reducing the size of the overlay allows stutter free playback of teletext
subtitles on significantly slower hardware.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The filter chain can change the format (in particular by changing
colorspace, or by deinterlacing to twice the frame rate), so wait with
initializing the encoder format until the filter chain has been set up.

It might be that the encoder itself demands format changes, e.g. due to
chroma demands or to make the resolution divisible by 16, but we can add
that extra filter to the end of the filter chain after the encoder
parameters have been set up.

Also fix a bug where we'd leak the filter chain when changing it during
the encode.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

The current code doesn't properly take into account that filters
can change the format (including the frame rate) from that of the
decoder's output, and thus uncritically use the decoder's output
format where it should use the output of the decoder -> filter
-> user_filter chain.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

In the video transcoder, call the filters as many times as needed
(second and following time with NULL input) until they stop outputting
frames. This means that frame-doubling filters, such as the yadif2x
deinterlacer, get all their frames output.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

This function was already getting very long and complex,
and we'll soon be moving it into two more for loops.

This also fixes a bug where output packets from the
encoder thread (with --sout-transcode-threads larger than zero)
could get lost, since we could extract the output blocks twice
and thus overwrite the output parameter.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

In the deinterlacer module, when b_double_rate is set, actually
double the frame rate in the output format.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

And don't make use of confusing raised QFrame panels.

This reverts commit c16bc980.
This is no longer applicable with current libav HEAD.

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

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

If the core requests a change of the size of the video window, and
the display plugin returns success, the display configuration stored in
the core goes out of synchronization with the X11 server. This results
in incorrect video placement.

Since it cannot be known if resize will succeed or not (at least not
without patching all window provider plugins to block and cross-check),
just return an error.

...rather than the parent window. This should work better if the parent
window is transparent or has whatever other unusual property. Also pass
a proper visual ID.

Refs: #8579

Close: #8579