On Fri, 23 Jun 2006, Armel Asselin wrote:

>> But - I believe in starting doing things simple, so I believe this kind of
>> advanced system with multiple pipes and priorities etc will be subject for
>> inclusion later. We should first make pipelining work with a more
>> simplistic plain one-pipe-per-server-until-full-then-start-a-new apprach.

> could the user have the notion of the "pipe" and tell "add this easy_handle
> to this pipe"

In most cases an app will have a hard time to figure out if the URL can in
fact be used for a specific pipe.

> or "how many pending reequest in this pipe"? (pipe==connection?)

I certainly can imagine that you'll be able to query an easy handle to get to
know how many pending requests the pipe it belongs to has.

> or does it already exist?

Pipelining support doesn't exist yet so nothing of this exists.

> with this two simple function, the user could decide by itself to which pipe
> to add and implement on top of libcurl "the advanced system"... or maybe two
> modes: the simplistic you describe or the manual one.

I appreciate you trying to come up with an approach that avoid having to
support the complex scenario Dan described, but I think your way is a bit too
simple as it won't allow all that Dan said without adding a substantial amount
of code. And even so, then it still won't be _sure_ since for example
pipelining with multi-homed servers etc cannot be figured out by any app by a
mere name resolve since libcurl might in fact resolve it differently.

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