On Wed, 31 Mar 2004, Roy Shan wrote:

> I wrote up a dummy dns server which never responds to query. So I can
> reproduce the timeout problem easily.

Nice job! Did it turn out like anything you'd consider sharing that we could
use to produce some test cases with to make it easier to test stuff like this?

> if the dns server drops some query, select always returns 0. In that case,
> ares_process never gets called. Then, it never times out.
>
> My fix is: remove line 493. That is always calling ares_process after
> select.

You're a star! I take it this change makes my previously added timeout code
for the ares lookup unnecessary?

> 1. each handle has its own areschannel. Can the design be handles sharing a
> channel?

Yes it can. We already have the share interface that allows data to be shared
among curl handles and we could make the areschannel sharable using that. It
is in fact mentioned in the docs/TODO document.

> 2. Curl_wait_for_resolv blocks and waits. Can it become non-blocking?

All occurances of Curl_wait_for_resolv() should be fixed to use non-blocking
mechanisms indeed. At least if they can get called when the multi interface is
used.

It is however quite a heavy task to turn into reality.

-- 
     Daniel Stenberg -- http://curl.haxx.se -- http://daniel.haxx.se
      Dedicated custom curl help for hire: http://haxx.se/curl.html