Just commited the following:

http://cool.haxx.se/cvs.cgi/curl/ares/config-win32.h.diff?r1=1.15&r2=1.16
http://cool.haxx.se/cvs.cgi/curl/ares/setup.h.diff?r1=1.26&r2=1.27
http://cool.haxx.se/cvs.cgi/curl/ares/setup_once.h.diff?r1=1.25&r2=1.26
http://cool.haxx.se/cvs.cgi/curl/lib/config-win32.h.diff?r1=1.53&r2=1.54
http://cool.haxx.se/cvs.cgi/curl/lib/config-win32ce.h.diff?r1=1.19&r2=1.20
http://cool.haxx.se/cvs.cgi/curl/lib/setup_once.h.diff?r1=1.28&r2=1.29
http://cool.haxx.se/cvs.cgi/curl/src/config-win32.h.diff?r1=1.30&r2=1.31

The meat is in setup_once.h

There exists the posibility that this might break any platform that
was including config-win32.h even when not building a WIN32 target and
relaying on HAVE_WS2TCPIP_H not being defined in order to get the
socklen_t definition. I think that this is very unlikely to happen.

Platforms that are including lib/config.dos could probably be verified.

As a following comment to previous mail, bare bones msvc 6.0 without
PSDK also lacks the header definitions for IPv6 and LDAP support, and
who knows what else.

We can try to do our best to support ancient compilers. But I really
think that it isn't to much to ask users who are using a 1998 compiler
to install updated headers released in 2003 for their compiler. After
all its nearly 2008. :-)

-- 
-=[Yang]=-