HOST ToDo -- Things that could be fixed/improved/added.

Patches to close any of these items will be gratefully accepted!  Other
suggestions and comments are also encouraged.

- implement host_h_strerror() to return descriptions for extended error
  codes from get_info() -- or maybe don't over-load h_errno in
  get_hinfo() but instead also implement global host_errno?

- util.c:canonical() should not use geth_byname() but should only use
  the results from the get_info(T_A) so as to avoid accidentally seeing
  YP or /etc/hosts entries that would not actually be valid.  Note that
  fixing this involves makeing sure callers and users of the return
  status that get_info() stores in h_errno don't accidentally assume
  it's a valid input to h_strerror().  This will fix the problem with
  /etc/hosts entries causing fake "not canonical" errors when the wrong
  hostname appears in h_hname.

- do not use gethostbyname()/gethostbyaddr() or their newer replacements
  to do basic A and PTR lookups and checks [i.e. addr.c:check_addr() and
  addr.c:check_name()], but instead use the low-level interfaces.
  Provide separate options for using those functions just so that the
  native implementations can be tested.

- check that the names in the answer records match the names we asked
  for (taking CNAMEs into account where they're allowed).

- build more mechanisms to allow checking using only authoritative
  sources, i.e. in the same way '-C' does; e.g. MX, NS, and PTR lookups,
  which might require looking up two separate sets of NS records, one
  for the the MX, NS, or PTR domain and one for the target canonical
  host's domain.  Need to deal with cases where the target name is the
  zone name.

- allow for setting the (stream) read timeout (conn_timeout) at runtime.

- use setservbyname() to get the local NAMESERVER_PORT value.

- allow runtime specification of destination port in server with ':PORT'

- allow runtime specification of source port with '-O'

- clean up the host_res_*() mess.

- don't strip the trailing dot from PTR targets that have only one label.

- maybe '-P' should be the default for '-C', but then we'd need
  something like a '--not-parent' or '--listed-ns' variant to explicitly
  check the nameservers listed inside the zone and it all gets a whole
  lot messier....

- check for both A and AAAA records.  The check for canonical hostnames
  in NS and MX records should be extended to include also AAAA records.
  Originally asked by Matti Aarnio <matti.aarnio@sonera.fi>

- try to build a regression test suite (but how do we feed it known
  bogus junk to test error handling?  nc?).

- think about using autobuild.
