This reverts commit 4332c070.

A higher value creates problem with OSD. This will need more
thinking.

 Demuxer running their own thread sending data to es_out will have to
properly handle es_out (reset).

Normal stream layer should be about the same latency now.

It removed the need of access_t.info.b_prebuffered
It reduces latency (removing the need for stream immediate)
It may increase a little the CPU usage but only for demuxer that
already do too much small stream reads.

 It is still needed as filter change need to be checked and do not use
this wait/lock (they are about picture fifo).

forward-port [578af5df020a05e7deb7f4b0fdf5b6c5d402814e]
Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

100 200 300 750 2500

For http/rtsp caching is doubled, for mms its 10x that (maybe too much on that one).

The schroedinger project and vlc module are much faster dirac decoders;
that should be used in preference to libdirac[-research]
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Laurent Aimar <fenrir@videolan.org>

 - OggDirac support using revised granule_position mapping:
     http://www.diracvideo.org/specification (20081109)
 - Uses p_extra for BOS page
 - Flushes Ogg page after each OggDirac packet
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Laurent Aimar <fenrir@videolan.org>

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

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

Used vlc_fourcc_t.
Protected a macro.
Used var_CreateGetNonEmptyString where needed.
Cosmetics.

 The rawvideo stream may not be CFR but has a timestamp for every frame.
 Anyway I prefer that we try to display the stream even with small glitches
than rejecting it.

demux:
 - Use accurate timestamps (date_t)
 - Signal frame rate in fmt_out
 - Fix aspect ratio for YUV4MPEG2
 - Reoganise input options to allow sensible overrides
   Ie: extension gussing < yuv4mpeg < cmdline

codec/packetizer:
 - Fix incorrect assumption of 25 fps video
 - Use accurate timestamps (date_t)
 - Copy input format as output format (to preserve frame rate)
 - Set progressive flag per picture

TODO:
 - Have module option for interlaced
 - Detect interlaced in demux for YUV4MPEG2
 - Have codec read set the interlaced flags in the video
 All seems rather tricky for the moment (well, YUV4MPEG2 bit does,
 since the demuxer strips the headers that reveal the interlace
 mode).
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Laurent Aimar <fenrir@videolan.org>

Iinvalid on POSIX (pthread_t is not scalar), crashing on Win32

Signed-off-b: Rémi Denis-Courmont <rdenis@simphalempin.com>

Pointed-out-by: Tobias Klein <tk@trapkit.de>

Pointed-out-by. Tobias Klein <tk@trapkit.de>

 Do not use vout_Request in aout_filter_t anymore but aout_filter_RequestVout.
 They have the same semantic but it will allows to control every vout creation
from aout.

NB, this highlights some reference counting strangeness.
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Laurent Aimar <fenrir@videolan.org>