is updated even if everything is read in one single pass, as the windows
functions apparantly does more often than other systems.

Really, we should only #define this in one file, not both here and in
connect.c!

Reminder for the future: when we're using malloc() we MUST include <stdlib.h>
as otherwise 64bit archs go bananas.

Bug report #517687

was an extra CRLF added to the post data

content-type header, it was used for each part and thus without this it
failed MISERABLY. *smacks forhead*

on my test machine! :-)

This was a real bummer!

is given

removed the initial CRLF in the formpost, as they are part of the request
and should be written by the code in http.c!

this work-around is not needed anymore!

  consuming operation that should only be needed to do once. We patched
  libcurl to now only seed on the first connect when unseeded. The seeded
  status is global so it'll now only happen once during a program's life time.

request of its own, it just changes the keyword of a request.

on IPv4-linux machines

pointed out this ugliness.

little fix seems to correct it. another case where we just returned and
didn't shut off the reading. This bug is introduced in 7.9.3 due to the
new internal "order".