$Id: TODO,v 1.10 2005/04/19 21:48:09 mjt Exp $

The following is mostly an internal, not user-visible stuff.

* rearrange an API to make dns_query object owned by application,
  so that it'll look like this:
   struct dns_query *q;
   q = udns_query_alloc(ctx);
   udns_query_set(q, options, domain_name, flags, ...);
   udns_query_submit(ctx, q);
 or
   udns_query_resolve(ctx, q);

* allow NULL callbacks?  Or provide separate resolver
  context list of queries which are done but wich did not
  have callback, and dns_pick() routine to retrieve results
  from this query, i.e. allow non-callback usage?  The
  non-callback usage may be handy sometimes (any *good*
  example?), but it will be difficult to provide type-safe
  non-callback interface due to various RR-specific types
  in use.

* DNS_OPT_FLAGS should be DNS_OPT_ADDFLAGS and DNS_OPT_SETFLAGS.
  Currently one can't add a single flag bit but preserve
  existing bits... at least not without retrieving all current
  flags before, which isn't that bad anyway.

* dns_set_opts() may process flags too (such as aaonly etc)

* a way to disable $NSCACHEIP et al processing?

* initialize/open the context automatically, and be more
  liberal about initialization in general?

* dns_init(do_open) - make the parameter opposite, aka
  dns_init(skip_open) ?

* for the above.  Use separate routine for initializing the context
  from system files, to not link stuff reading resolv.conf if it's
  not needed.  So that automatic init will not be possible.

* allow TCP queue?

* detect servers which don't understand EDNS0 (returning FORMERR),
  and fall back to pre-EDNS0 for them.

* for the above to work, we have to be able to find query object by
  only ID, not ID + qdn.

* And oh, qID should really be random.

* more accurate error reporting.  Currently, udns always returns TEMPFAIL,
  but don't specify why it happened (ENOMEM, timeout, etc).

* check the error value returned by recvfrom() and
  sendto() and determine which errors to ignore.

* maybe merge dns_timeouts() and dns_ioevent(), to have
  only one entry point for everything?  For traditional
  select-loop-based eventloop it may be easier, but for
  callback-driven event loops the two should be separate.
  Provide an option, or a single dns_events() entry point
  for select-loop approach, or just call dns_ioevent()
  from within dns_timeouts() (probably after renaming
  it to be dns_events()) ?

* implement /etc/hosts lookup too, ala [c-]ares??

* sortlist support?

* windows port?  Oh no please!..  At least, I can't do it myself
  because of the lack of platform.
  Ok ok, the Windows port is in progress.  Christian Prahauser
  from cosy.sbg.ac.at is helping with that.