cpearce (Chris Pearce)

Removed invalid assertion.

Don't assert that vorbis packets don't span pages, they technically can, even though most encoders won't output them as spanning.

Index chopped theora streams correctly; we were assuming that the first data packet's packetno is 3, but in chopped streams which include the remnants of a preceeding packet in the first page, that's not always the case.

delete[] what's new[]ed.

Ensure fisbone packet is always the correct size, rather than assuming the content type message header is always 28 bytes.

Fix theora granulepos calculation for 3.2.1 streams, and handle case where we're capturing theora granulepos and we encounter a keyframe between capture and the first buffered frame.

Fix iterative backoff to find last page with time when getting end time.

Updated readme, deleted old spec.

Updated spec to skeleton 3.1 keyframe index spec.

Build validator with -O2 and witout -g.

Initialize VorbisDecoder::mNextKeyframeThreshold to negative value, ensures we always index keyframes for vorbis!

Fix theora granulepos correction when preceeding keyframe.

Warn when index packets have no keyframes.

Corrected granulepos calculation in theora case where frame is before a keyframe.

Ported to linux.

OggIndex now produces valid indexes as per OggIndexValid. I was off by one in theora tracks, and I changed OggIndex to only consider the first keyframe in any page for indexing. Reimplemented -k, -d and -p command line arguments.

OggIndexValid now checks that keypoint presentation times are accurate. I had to rewrite OggIndex in order to make it more accurate. Restructured slightly as well.

Refactored out index verification code, and created separate OggIndexValid program which determines whether the offsets and checksums of an index are correct.

Ported to Linux.

Option to change min time between indexed keyframes.

Fixed typo in spec.

Added paranoid null checks and made serialno unsigned instead of singed long.

Fixed compile error and warning on Linux.