On Mon, 7 Feb 2005, Gisle Vanem wrote:

> Ah, if Curl_wait_for_resolv() returns != CURLE_OK, 'conn' should never be
> touched again. This is a problem with C-ares too (but I failed to trigger
> that situation). Removing Curl_disconnect() at end of
> Curl_wait_for_resolv(), the crash is gone. But that causes a memleak for
> unknown hosts:

What if you forcibly set conn->bits.close to TRUE? Then the disconnect will be
made later on in Curl_done().

> IMHO the resolvers should never invalidate 'conn'. Is there a nicer way to
> clean up?

I agree that invalidating conn there will lead to problems such as this. Let's
avoid that.

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