Here is the updated patch with the modified configure.ac.
I have also added some comments and cleaned up a little bit of code.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

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.