default netrc file. We now read and uses HOME _before_ we use getpwuid() to
better allow users to move around HOME to use different .netrc files without
having to rely on even blacker magic.

use as aggressive warning options as possible for the used compiler version.

item 25 is now dealt with using the curlx_ prefix, mentioned in detail here:
http://curl.haxx.se/mail/lib-2004-02/0215.html

officially using library-code when building the app (at least for the platforms
that don't have a strtoll() on their own).

linked separately by the application. This function is not provided by the
libcurl API. It can only be accessed by apps if they compile and use this
particular source code.

now scan the PATH ourself to find the path to (g)libtool

figure it out here anymore to build debug builds.

argument to also not force the casual includer to know about the socklen_t
type.

sockaddr *. This has the added benefit that source files that include
memdebug.h doesn't have to know about "sockaddr".

my_getaddrinfo() (caused a warning by the IRIX MIPSPro compiler). Also
clarified the situation for the 3-arg version of gethostbyname_r() with a huge
comment.

When using it, we now set the include path to better find the devel curl
headers, and we check for the socklen_t type since the curl memdebug stuff
needs it.

be. Simply skip the pruning.

memory leaks etc.

code fine on ares!

to memdebug the ares stuff as well.

same channel during the whole curl handle's life until curl_easy_cleanup().

is a valid pointer, but instead always depend on the passed-in dns pointer.
This happens to be NULL when the connection is re-used...

and memset

to properly work with SOCKS5 proxies. I also included some ascii art describing
the SOCKS5 response, as RFC1928 describes. Jeff provided details in bug
report #741841 and here: http://curl.haxx.se/mail/lib-2004-02/0181.html

seems to correct some host cache screw-ups I could reproduce.

SO_BINDTODEVICE is present, to prevent compiler warnings about the variable