[...]
   > primitives, 3) have enough things implemented, and 4) conform to an
   > existing layout.

   [...], but it does seem a bit gratuitous to limit the sort of
   information you can get to the fairly limited set of things the IETF
   saw fit to put into the common MIBs.

You're substituting `straightjacket us' for `conform'.  For things
which don't fit in MIB II, we'd obviously have to extend the MIB
ourselves.  There's a standard way to do this, and it's not a big
deal.

   The other notion, though, of having the queries go to an snmpd, or
   something like one, still seems rather bad,

As I said, this can be optional, but we should definitely have the
ability of doing remote queries.

   and Jason has made the
   interesting point that you can't run netstat or whatever on a kernel
   dump if it uses sysctl(), and one does need to be able to do such things...

First of all, that's an argument for why we should do nothing, and I
think it's fairly clear that's not a viable option.

Secondly, Jason's point is not necessarily true.  One could imagine
designing a tree data structure in the kernel that could be groveled
manually if necessary.  This seems fairly easy.

Lastly, you seem to be ignoring the point that creating a random new
interface does not solve this problem, and in fact creates a variety
of new problems.