Currently in most cases the row after the last is returned.
When the last row is visible vbi_draw_vt_page_region() could
crash, because it went over the size of p_page.text and use
invalid values (>=40) as index for p_page.color_map.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Shared cookies are required, for example, by certain HDS and HLS streams
that set a cookie when the manifest is read and expect it to be sent
back on subsequent fragment requests.

The cookie jar is created during playlist initialization. HTTP requests
inherit the cookie jar from the playlist.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

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

VDPAU introduces a delay of one field (one picture if not deinterlacing).
This is normally not an issue as the PTS is preserved. But it does not
work if there is only one picture not followed by another one.

Needed for av_freep

Fix leak on error, fix mismatched free function on success.

A deadlock could occur in an high load situation where the vout and
the decoder are taking excessive amounts of time.

The vout thread repeatedly call ThreadDisplayPicture in its main loop
until it returns an error, while keeping the picture pool locked. If
no picture was recently received, the vout will redisplay the current
picture (a "refresh") by calling ThreadDisplayRenderPicture with
is_forced=true. If this refresh is excessively long, the vout thread
will be stuck in a refresh loop. The decoder cannot make any progress
since the picture pool lock is hold and the vout won't be polling for
control commands, yielding a total deadlock of the program.

This situation can be reproduced artificially by sleeping in the
decoder and decreasing variable VOUT_REDISPLAY_DELAY.

A simple solution to this issue is to exit the ThreadDisplayPicture
loop after refreshing. Since a refresh typically occurs when no new
pictures are received from the decoder, this should not decrease
performance.