It fixes a regression with interpolated timestamps for some
mpeg-I/II and for h264.

Force new not to throw. Also drop a separate declaration of a static function that's called exactly once, and mark it inline instead.

It is needed to ensure valid buffer alignment.

And don't ignore errors

You need a compiler.

Also use autoreconf where possible
And no need to reun automake after autoreconf

This reverts commit f001a156.

This is usually a good practice. Moreover compare to locking and unlocking the
potential overhead is really small.

Signed-off-by: JP Dinger <jpd@videolan.org>

Signed-off-by: Rémi Duraffort <ivoire@videolan.org>

Signed-off-by: Rémi Duraffort <ivoire@videolan.org>

reset the current status item.
(cherry picked from commit 6d28f6de)
Signed-off-by: Rémi Duraffort <ivoire@videolan.org>

It seems that avformat does not recover from a byteseek when it does not know a duration for the file. It continiously spits out data with pts values of AV_NOPTS_VALUE after the seek.

Tested with mxf_original_NOB_format.mxf

When we set the video track, we probably be wanting to change video-es vars instead of audio-es

contribs: Use GCC 4.2 instead of LLVM-GCC 4.2 to compile the 32bit contribs on Darwin like we already do for the 64bit ones

LLVM introduced to many runtime bugs for now. Updated bin package for x86_64 is already available, PPC and i386 will follow within the hour.

Try TI_1_Einfuehrung.avi to trigger the bug.

(cherry picked from commit d0e16efd)