A couple of ideas from Mike Patton

[Rob Austein <sra@hactrn.net>]

Mike mumbled a couple of things at me on his way out the door (he
drove here from Boston, and wanted to leave tonight to try to beat
tomorow's snow).

1) Reccomended implementation of the "don't follow A6 chains" thing
   ought to be a number indicating how many links in the chain to
   follow.  This number would appear in the config file (or whatever)
   for a full service resolver.  A resolver that only supports the
   degenerate case would only accept a value of zero here.  Setting
   the number to a large number should produce warnings.

2) Suggested way of looking at AAAA: stop thinking of it as an RR type
   per se, and start thinking of it as just a QTYPE (this isn't quite
   right since it would also appear in non-query sections, but close
   enough for the moment).  Stub resolvers that don't want to do the
   work to piece together A6 would just ask for AAAA, done.   Pretty
   much what we were going to recomend for transition, but Mike says
   we should consider leaving this in place forever.  Note that this
   would allow us to replace A6 with something even more horrible at
   some later date without replacing the stub resolvers.

   Mike admits that this is at best half-baked, but we though it worth
   thinking about.