# $Id: TODO,v 1.216 2009/10/08 09:04:51 michaels Exp $

netperf
   TCP REQUEST/RESPONSE TEST is slower with new server by ~7-8% (5100 vs 5800)

* look into if we can reduce the stacksize of clientmethod_gssapi(),
  as it creates problems with e.g. solaris' host command.
  Do we really require so big gssapi buffers?

* can we use fstat(2) or similar on the sockets to track them instead?

* we evaluate rfc931 twice if both the client-rule and socks-rule specifies
  rfc931, that is unnecessary.

* connections blocked due to authenticationerrors or other things should
  say block(n), not pass(n).

* change gethostby* to getnameinfo(3)/getaddrinfo(3), which should hopefully
  be threadsafe on most platforms.

* should try to get route for v4 first, since it's faster setup time?

* (bandwidth?) module could support a limit on the total amount transfered
  per time (day/hour) also.  Usefull?

* connecttimeout should mean "timeout for connect(2) on behalf of socks
  client".

* make request-children support multiple clients.

* change _t struct's to typedef.

* share host/addr cache across processes.

* clean up namespace (including for dlib, function names can collide with apps)

* make requestchildren support multiple clients too.

* shrink io_t, too big.  Can we find out what the maximum amount of 
  bytes needed for importing/exporting gssapi auth is?

* IPV6 support

* osdep.h:move defines from osdep.h to configure/autoconf.h

* check for more configure related code that is no longer used
