[ogg-dev] possible regression in libogg

Monty Montgomery monty at xiph.org
Tue Dec 3 14:49:41 PST 2013

Any time code is stuck in a loop calling ogg_sync_pageseek() or
ogg_sync_pageout(), you'll see most of the time sunk into checksum
calculation.  It's the only thing in libogg that _can_ take much time.
 This is not likely a libogg bug, but a bug in the code calling libogg
in a tight loop.

Looking at the file listed in the bug, there's a page discontinuity at
the very beginning; after the header packets/pages, there are several
missing pages before data starts normally.  Whatever generated or
trimmed this file did a ragged job at the beginning.  This isn't an
error, strictly speaking, but it's not completely correct either.
First thing I'd check is whether or not this has tripped up the
seeking code in gstreamer...  Or actually, _any_ kind of analysis of
what gstreamer is actually doing.

A profiler is not a debugger!  It's just waving a dead chicken at this
problem, then trying to read the tea leaves.  As of yet we have no
concrete evidence of what is happening aside from gstreamer
calculating a ton of Ogg checksums.


More information about the ogg-dev mailing list