[Vorbis-dev] Strange block switching in this sample

[Stephen So]

One of the members from HA.org has produced a rather unusual sample, 
where Vorbis is doing some very strange things. Here is a link to the 
sample:

ftp://ftp2.foobar2000.net/foobar/Bloch_Formule.wv

It would appear that Vorbis at q 4 suddenly decides to use long blocks 
to encode the 2nd half of the sample, which produces some noise.  I've 
uncommented out some of the statements in block.c to print out the block 
types being used and for mose of the sample, Vorbis uses impulse and 
padded blocks, which is a good sign.  But towards the 2nd half of the 
sample, it decides to use only long blocks, which is accompanied by a 
drop in the bitrate, and is noisy.

Is there a bug somewhere?

Link to the relevant post: 
http://www.hydrogenaudio.org/forums/index.php?showtopic=30230&view=findpost&p=261596

Steve.