On Tue, 25 Dec 2007, Dmitry Kurochkin wrote:

> Can you give some list of what should be implemented to get proper
> pipelining in libcurl (at least the scenario described above)? In am ready
> to work on this if it does not require too much time.

Cool!

I would say that we should make the "not open enough" connections to be
acceptable for pipelining, but the connection must not be used until it is
connected fine (and when built with asynch DNS resolving, it could even be
pipelined before the name resolution is complete), and gracefully handle the
case where the connect/resolve fails and there's one or more handles pipelined
on it.

For me, it's always been a problem that we don't have decent Pipelining
support in the test http server so we haven't been able to write up a bunch of
decent test cases and working with these things against live HTTP servers. I
would really like to see that changed too as it will make working on these
things a lot easier.

> At the moment I use libwww for HTTP. And it does pipelining properly. But it
> is not maintained upstream and is likely to be removed from Debian, so I am
> interested in moving to libcurl.

I am also very interested in making this work in libcurl! It's just this old
and boring not-enough-time problem that's been bugging me! ;-(

(I believe libwww is more or less a dead end these days.)

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