might change it. I don't *think* it does it when the timeout is 0,0 but it
is better to be sure...

testing so that all test files are build first properly. David Byron reported.

Andrés García helped out!

A patch to bypass MS' sillyness with regard to IPv6 and getaddrinfo().

The CURLDEBUG part is to avoid redefinition warning caused by memdebug.h. If
ENABLE_IPV6 isn't enabled, it doesn't matter since we never call
getaddrinfo(). Allthough we could to support weird protocols like SOCK_RDM
that Win-2K/XP has.

a check for there *_r-versions too

new configure script detects them and now this code acts according to what
API that was detected

1. Try with _THREAD_SAFE instead of _REENTRANT, as AIX seems to require it
   and if _REENTRANT is required we should already have it set since one of
   the previous tests.
2. Added API-detection for what kind of strerror_r() that is provided. The
   POSIX style or the glibc style.

Tor Arntsen provided the necessary feedback these changes are based upon.

'maketgz'

while perl often is installed in /usr/local/bin or elsewhere

Added a more verbose comment about what strerror_r() can set errno to in
case of failure.
This file still doesn't build on Solaris due to a missing 'sys_nerr' symbol.

Curl_strerror() that attempts to be thread-safe _and_ works on Windows too!