On Thu, 8 Dec 2005, Legolas wrote:

> You can give a look to the attached source code to see what I am talking
> about. Please note that the example does not cover all the possible usages
> of this new kind of API, very flexible. A basic support of chained buffers
> has been implemented, but I think it should be designed better.

Yes, I believe something like this is what we need. I think your general
approach is fine and quite similar to what I've had in mind and tried to
express previously.

If you want to proceed and starting poking on libcurl internals to actually
implement this, I really must urge you to change source code style as libcurl
looks quite differently (2-space indented, 80 columns, open brace on next line
when declaring functions and preferably no extra typedefs for structs).

I plan to have a go at the zero copy stuff myself as part of my hiper work,
should time permit. I hope to have a go at it closer when I'm done with the
*socket() API (and HTTP pipelining), but time-wise that won't be until later
in spring/summer 2006. If you start the work on zero copy before this, it
might ease my work or I can let my part of the work be to merge your work into
the main code.

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