input_CreateThread* have been splitted in two functions:
 - input_Create
 - input_Start
to allow attaching to input events and creating settings/variables
before starting the input thread.

For consistency, input_StopThread has been renamed input_Stop.

Initial report from Hugo Beauzee-Luyssen.

Signed-off-by: Christophe Mutricy <xtophe@videolan.org>

Signed-off-by: Christophe Mutricy <xtophe@videolan.org>

Signed-off-by: Christophe Mutricy <xtophe@videolan.org>

Signed-off-by: Christophe Mutricy <xtophe@videolan.org>

 At each serialno changes, the timestamp being non continuous, there
will be an annoying glitch...

 - repacketizing and reconstructing timestamps for a Dirac ES
   suitable for remultiplexing and decoding.
 - sanitizing an input stream that has non-signalled discontinuities
 - handling the basket case AVI fake-pts-is-dts method too:
   For a non well formed input (only has pts or dts), this is
   implemented by simulating the ReOrder Buffer model to resolve
   timestamps.

Updated 20090507 with:
 - use VLC_TS_INVALID rather than private block flags
 - removed useless i_sync_flags
 - add support for next_parse_offset == 0
 - eliminate dirac_FindEOS
 - use block_t's p_priv to avoid dirac_FindPictureNumber
 - remove b_picture_number, add parsecode helpers
 - when stream jumps backwards, generate correct EOS DU
 - dont use block_FindMaskedFlags
 - remove openloop fastsync
 - guarantee synchro sees EOS, remove duplicate EOS DUs
 - dont blindly believe that a data unit is > 1MB
 - fix compilation warning
 - fix issue with unsyncable p_extra
 - rearange file
 - add sanitization of prev_parse_offset
 - do duplicate EOS detection at the right place
 - move all operations on outqueue to Packetize
 - fix stupid offbyone on location of prev_parse_offset
 - sanity check output of sync at input to encapsualator
 - refactor dirac_BuildEncapsulationUnit
 - keep hold of precious timestamps
 - unfortunately block_t has no p_priv
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Laurent Aimar <fenrir@videolan.org>

(Crop is currently dysfunctional, picture offset are missing)

Thanks David Flynn for the remark.

This one is more for consistency with what the core do.

demux_PacketizerNew has been fixed in that regard.

Updated 20090507 with:
 - migrate to VLC_TS_INVALID
 - fix coding style
 - use generic handling of DEMUX_SET_TIME
 - don't add +1 to dts/pts
 - zero point hack while VLC_TS_INVALID = 0
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Laurent Aimar <fenrir@videolan.org>

 On such flag, the bitstream before and after it are not related at all and
so are the timestamp. In order to not confuse the core, do not output
wrong timestamps.

Also, don't reset pts interpolator at discontinuity
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

timestamp will be interpolated at output of decoder instead.
Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Signed-off-by: David Flynn <davidf@rd.bbc.co.uk>
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>