On Wed, 10 Sep 2008, Dmitriy Sergeyev wrote:

>> A) It assumes that the multi perform function is called again before the
>> returned failure is noticed. This may be true in most cases, but I bet we
>> can think of scenarios where it won't be called, or at least won't be
>> called again for quite some time and that may end up rather strange
>> behavior to applications.
>
> I suppose then, good way to solve this is to never call write-callback on
> curl_easy_pause at all. And modify curl_easy_pause in such a way, that it
> will only alter some transfer state. But write-callback will be called only
> on multi perform call.

That wouldn't solve this problem at all, since it too would rely on the
application to call curl_multi_perform() "soon" but there's no guarantee that
there will be any further action on the connection and there may not be any
timeout that triggers either.

The more I think about this problem, the more I lean towards that the current
approach more or less is the only reliable one.

-- 
  / daniel.haxx.se