On Wed, 24 Dec 2008, Igor Novoseltsev wrote:

> While continue testing my product I encountered CURLM_CALL_MULTI_PERFORM
> status returned by curl_multi_socket_action.
> According the curl_multi_socket_action documentation I have to call it
> again.
> When I call it again, I have no events on the socket, therefore I
> provide ev_bitmask=0.
> In this case the patch logic doesn't work since it uses ev_bitmask.

Euh... you're right of course. How frustrating!

At a larger scope I think I want to get rid of the CURLM_CALL_MULTI_PERFORM
thing completely and instead make the multi interface handle all of those
situations internally instead, since this is just not ever really used for
anything by apps but simply always dealt with by a loop anyway.

In the short term, we could possibly attempt to fix the
CURLM_CALL_MULTI_PERFORM case for curl_multi_socket() only as that would
address this immediate problem.

-- 
  / daniel.haxx.se