> The problem: select on the /dev/sound fd is never saying that data is
> available. Why is this the case? Isn't there always sound available to
> read?

Do you get sound to read at all, say with cat or dd?  If not, it could be
related to kern/26706.  I was on a machine with a neo chip specifically,
but the record-never-wakes-up bug was in audio, not neo.  I think I was
able to get it working in 1.6.2 by changing a >= to a > (the patch is in
the PR) - but the same patch doesn't work in 2.0 and I haven't had time to
pursue it again.  I kind of got the sense that audio *input* might not be as
heavily tested in NetBSD as audio output - I remember looking at the ring
buffer code and thinking "I wonder for how many people this works?"

-Chap