> Is there good reason for upgrading our resolver library to bind 8's
> now, or should we wait for bind 9?

since we're caring more and more about threads lately, i think the ability
to call the thread safe irs_*()/nres_*() api from thread aware programs to
parallelize massive dns lookups would be useful to have.  bind9's api will
be a superset of bind8's so there's nothing to be lost by upgrading twice.

this has implications on the "resolver search order" (nis/yp lookups), so
it's not just a dropin.  you'd want to merge our "thread-specific globals"
versions of get*by*() into the non-IRS-style search ordering versions you
have now.  it's not rocket science but it's not trivial either.

(i'm purposely not going to bitch about the non-IRS-style search order code,
since i don't use it on my own netbsd system -- i just dropped in bind8's
logic into the libc, just like i did under contract to bsdi a few years ago.)