On Wed, 18 Jul 2007, Guenter Knauf wrote:

> I have found and fixed this problem, it was problem in the user/kernel
> boundary code for recv(). This code didn't properly handle the error case of
> passing a zero length receive buffer ro recv(). Don't know why curl does a
> braindead thing like that but it does.

It would indeed be interesting to know _when_ he gets that (and why the code
would take such a route). The logs you showed before from your problems didn't
seem to hint about any such, afaicr.

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html