> Given the current facts and feedback, I guess this is the best idea so
far. It will of course make CURLOPT_POSTFIELDS act in a non-obvious way
compared to all the other setopt() arguments (thus it may lead to some
surprising bugs for
users) and CURLOPT_COPYPOSTFIELDS would need to have the size set before
it can copy the fields (if binary contents are desired).

The attached patch should do it !

I did not tested it intensively, because I do not have anymore time to
work on it until Monday. So it's provided here for people in a hurry and
those wanting to test...

Have a good week-end.