Content-Type: text/plain; charset="utf-8"
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
X-RT-Original-Encoding: utf-8
Content-Length: 1954

> $ CHARSET=ASCII dig axfr . @k.root-servers.net | grep '^xn--' | wc -l
> 576
> $ dig axfr . @k.root-servers.net | grep '^xn--' | wc -l
> 0

Ah you're right, that does work with IDN2003.  I was testing the input
side: "dig straße.de", and that's not working.  Here's what I'm seeing,
marked with (FAIL) where it didn't match expectations:

   No IDN support:
     "dig straße.de" sends a query for "stra\195\159e.de."
     "dig axfr ." contains xn-- names

  idnkit:
     "dig straße.de" sends a query for "strasse.de"
     "dig axfr ." contains names in non-ASCII charsets

  idnkit (with IDN_DISABLE):
     "dig straße.de" sends a query for "stra\195\159e.de."
     "dig axfr ." contains xn-- names

  GNU libidn:
     "dig straße.de" sends a query for "strasse.de"
     "dig axfr ." contains names in non-ASCII charsets

  GNU libidn (with CHARSET=ASCII):
     "dig straße.de" fails (FAIL)
     "dig axfr ." contains xn-- names

  GNU libidn2:
     "dig straße.de" sends a query for "xn--strae-oqa.de".
     "dig axfr ." contains xn-- names (FAIL)

  GNU libidn2 (with LANG=en_US.ASCII):
     "dig straße.de" fails (FAIL)
     "dig axfr ." contains xn-- names

Note, something has gone wrong in configure with building idnkit; I
had to build it from an unpatched source tree.

It occurs to me that we could simplify things by checking for
IDN_DISABLE in dighost.c itself.  Then we could have the same
documentation for all three libraries.  (Though it might be worthwhile
to add some text explaining differences between IDN 2003 and 2008.)

That still leaves the problem that libidn2 doesn't seem to be
working on the output side -- I only see xn-- names from libidn2,
regardless of environment.

Here's a diff showing the patch as I've modified it for ISC style
compatibility.  Files that are automatically generated (man pages,
configure, config.h) have been omitted.  (I thought it might be easier
if we're both working in the same code.)